i3-12100f is almost 100 usd cheaper in most builds you would use extra cash for better gpu or just get i5-12400f with the same / lower price and higher clocks but lets wait for benchmarks.
For Intel, those are the cheapest of the cheap boards with the very limited H610 chipset. If you want something acceptable you will pay a lot more. AMD on the other hand has cheap good option with B550 or even B450.
I am interested in your belief that most builds needs GPU more than CPU, or that most builds even have GPUs. iGPUs have ~70% of the market. Most builds are going to spend the cash on CPU performance.
If you're buying the CPU separately, it means you'll be building the PC yourself, not a prebuilt. Upgrading the GPU instead of the CPU will give you better graphics performance, which is what most of the people building one want. The only reason iGPUs have such a large section of the market is because of people buying regular prebuilt to use as a general computer, not a gaming one.
That's not obviously true to me. Is graphics-intensive gaming - and I would point out that defining "gaming" as GPU-intensive would be too narrow - really that large of a market? I've personally built dozens of PCs and the last time I bought a GPU it was a 3DFX Voodoo.
The only "new" chip is the X3D one (with 3d v-cache). The others are the same chips as last year, just at a different price point (eg. the 5700X is basically a cheaper version of the 5800X, only being marginally slower).
Kinda' goes to show the resourcefulnes of salvaging for repurposement based on yields right? Sort of reminds me of the now ancient Black Edition releases with unlockable cores with a big fat YMMV sticker on top.
it depends on market situation but usually those kinds of SKUs (if they're released worldwide in large volumes) are not related to yields but rather artificially disabled.
People hear about binning and assume that every product decision has to be related to binning, but usually it's not, it's just market segmentation. AMD had over 80% of Zen2 chiplets coming off the line with 8 fully-functional cores, and clock bins are generally selected such that most units will pass, by design. And that's at launch, on a new node, in 2019. Numbers have only gotten better over the last 3 years.
AMD already has a bin for iGPUs with a defect - it's 5600G/5600U/5600H/etc. And they have 5300G below that allowing even more defects. There's very very few APUs coming off the line with tons of GPU defects but 6 workable cores, or a defective PCIe controller but only a defect in the iGPU part and not the rest, etc.
The problem is that AMD has tons of supply of high-binned parts but the lowest demand for those parts. And they have the highest demand for low-binned parts but the lowest supply of those parts. How do you mesh those two curves? Disable cores on a high-binned part and sell it as a lower SKU. That's why those "black edition with unlockable cores" existed - those unlockable cores were locked off for market segmentation. Nowadays they just don't let you turn it back on.
(Which isn't to say that none of the 5600X/etc are the result of a dead core/etc - but a lot of them aren't, probably most of them aren't, given the likely >90% yields for 8-core at this point. And you pick your 5600X bin such that most 5800X failures can be sold as 5600X, meaning there's very little that falls through the cracks without being just utterly broken. True binning-generated "we have this pile of chips, let's do something with it" style SKUs tend to have extremely limited availability as a result, it's shit like the Ryzen 4700S or the Ryzen 3100, or the NVIDIA 1650 KO.)
Anyway, it's not a coincidence this is coming a few months after the Alder Lake launch. This is market-driven, Alder Lake is not only faster but in many cases it's cheaper as well. AMD coasted a little bit while motherboard supply firmed up for Intel, but they finally have to respond. I'm sure they're selling lots of Milan but consumer marketshare matters too, and AMD is losing steam there with the price increases, with Intel back on top on performance, and with Intel undercutting their pricing heavily.
I concur, as someone who only pays attention to the processor market twice a decade I'm clueless when it comes to navigating the matrix of CPU market segmentation. It makes me yearn for automobile-style naming where you just have a year, a make, and a model.
Largely those style names do exist in the current product names. Since Intel styled the Core i-series 3, 5, and 7 (and the later 9) over 12 years ago, AMD is following their lead with Ryzen 3, 5, 7, and 9.
The first digit is the generation and then the second digit and suffix letter designate additional features. So a Ryzen 7 5700X is a 5th gen Ryzen 7 meaning it has 8 cores. It is a step down in price and performance to a 5800X and also has no Graphics capabilities (those have a G suffix). Intel uses different letter suffixes such as the Core i7 12700KF being a 12th generation Core i7 with overclocking features (K) and no graphics (F).
It is far worst with GPUs. Certainly an industry wide naming convention reset would be nice but everyone is not going to cooperate like that.
It does take a minute to read up on but at least Intel has *mostly* followed this naming convention for 12 product generations so far.
Even within one vendor it's super confusing. i3? i5? i7? Y series? U series? I think 12th gen Intel may even use different naming conventions already... A Y series i7 of olde delivered a significantly different experience than an i3 U series (literally the difference between being able to run some workloads comfortably vs. locking the machine up). And then marketing materials make it harder because they will drop the CPU model name for the nondescript "i5 CPU". I think it's probably somewhat intentional.
if they don't call out mobile/ULV processors with their own series (Y and U series) then people whine that they're trying to "sneak lower-performing processors into the lineup using the same numbering".
similarly, marketing a processor as being a "5700U" or "5700G" despite it being slower than a desktop 5700X could be seen as equally deceptive. That's not really any better than "i7" vs "m7" or whatever.
unfortunately you're just going to have to learn the naming convention, there really isn't a good solution for that, given the wide range of applications that a given series might be applied to. You just happen to think that AMD's naming convention is worth taking the time to learn while throwing your hands up at the Intel naming. Same with people who think Intel is just awful for internally codenaming all their products after Lakes and Coves while eagerly memorizing every single painter and city name that AMD uses in their codenames - says more about your priorities than their naming scheme.
> You just happen to think that AMD's naming convention is worth taking the time to learn while throwing your hands up at the Intel naming. Same with people who think Intel is just awful for internally codenaming all their products after Lakes and Coves while eagerly memorizing every single painter and city name that AMD uses in their codenames - says more about your priorities than their naming scheme.
The person who was saying they didn't understand the Intel naming convention because U and Y SKUs were too complex to understand.
That's normal. Mobile and ULV SKUs are usually coded differently. AMD calls them "U" and "H" skus I believe.
The person in question just doesn't understand the Intel naming convention, which is fine, but AMD and every other company does the exact same thing. It's not that Intel is uniquely confusing, it's that the individual here doesn't feel that Intel's naming convention is worth the brain space. Which is also fine, but it's not a problem with the naming convention.
As a tangential observation (people do make those in discussions!) architecture/product codenames are another place this comes up. There are many enthusiasts who will eagerly memorize that Rembrandt > Renoir but think the idea that Rocket Lake > Coffee Lake is perplexing and confusing. Or at least that's a repeated theme in many of these naming discussions.
I'm doing my best here to say this politely, but a lot of people clearly just don't value those two bits of knowledge equally. And dipping into rhetorical "who are you even responding to!?" doesn't really further the discussion either.
Naming isn't hard and naming discussions aren't interesting.
As someone who understands the intel naming convention quite well and could understand the AMD naming convention if I had a reason to I can still empathize with random-consumer-x who does not need to have that any practical understanding of the naming convention and could be confused as to what they are really purchasing...
Well, sadly, there is more than one use-case for computers so we need multiple power brackets, so U and Y SKUs are going to continue to exist. It's appropriate to call it out with the naming scheme, but that's exactly what AMD and Intel have done here.
There is no solution which is going to be 100% intuitive to someone who specifically doesn't know anything about what they're looking for. If you move those products out to their own separate series, that's what Intel did with the Y series ("m7-xxxY" line - contrast to "i7"). You specifically don't like that. If you mark them within the existing series, that's what Intel and AMD do with the U series. You specifically don't like that either. If you move them into a single series, you end up with something like the Intel Ice Lake/Tiger Lake naming convention, where there is some part of the name that means "cores" and some part of the name that means "power" and part that means "graphics". Other people really didn't like that, because now you have one name that means 5 different things.
(And this is what I mean about the naming discussion being dumb and boring - whatever you think is how it should be done, someone else hates that, and thinks it is too complex and requires too much knowledge on the part of buyers. It's bikeshedding, product naming is low-stakes so everyone has an opinion on it and is very upset that AMD and Intel are ignoring their urgent forums posts. At the end of the day it's just not that interesting, nor are any of these naming schemes that difficult if you bother to learn what they mean.)
Anyway, it's unfortunate that there are features and distinctions which laymen may not understand, but that's a fact of life, there's things car people really care about that a Camry Buyer doesn't know, and that's fine.
Someone who bought an expensive truck with a base-model trim might be upset that a Camry with a top trim is "nicer", because they didn't understand what a "trim" is before they laid down their money, and that's unfortunate but it's not exactly hidden either, nor should we call to get rid of trim levels because one person didn't understand. Someone else might be really upset that their SUV doesn't tow like a truck even though they got the nicest trim level on the SUV.
Again, sorry if this seems frustrated, but this is a topic that has been bikeshedded endlessly. The stakes are low, there's multiple reasonable options available, and there's a whole lot of people who are all really upset that AMD and Intel aren't taking their forums posts on the topic seriously. Naming conventions are fine, they're good enough to not matter.
Intel and AMD use a similar naming scheme, but get a little confusing with suffixes, especially AMD.
Both include a target market, generation, and performance level. With Intel, it's "i{market}-{generation}{performance}", with market being 3 (budget), 5 (mainstream), 7 (enthusiast), 9 (high-end). The numbers used in the "performance" level varies, but higher is always better. For example, i7-3770 is the third-gen chip targeted towards enthusiasts and was the highest performer in its generation. My i9-9900 is the top end model for the 9th gen. Intel will also use suffixes "K" to mean it has an unlocked multiplier (making overclocking easier) and "F" which means it does not have an on-die GPU, so you'll need a discrete GPU card.
AMD is similar, they'll call them "Ryzen {market} {generation}{performance}", ie, Ryzen 5 5600. But where AMD goes crazy is with the damn suffixes that express an additional performance level that is impossible to decipher.
A Ryzen 2700 is a second-generation CPU with 8c/16t.
A 3700 is a third-generation CPU with 8c/16t
A 4700g is a second-generation CPU with 8c/16t and a GPU
A 4500h is a second-gen CPU with 6c/12t and a GPU
There is no 4600G.
An X suffix generally means higher clockrate; an H suffix generally means high efficiency (lower power draw); a U suffix means ultra-efficient (very low power draw).
In the 5000 and 6000 series, possible suffices include X, H, U, HX, HS, but not G.
Some CPUs are OEM-only.
Some CPUs are only sold in packaging for laptop/tinybox manufacturers.
In general, if you get integrated graphics, you lose an entire processor generation, but more recently you only lose top-end features.
for a while, a lot of that could be summarized with "APUs are 1000 higher than their desktop generation" (4000 series APUs = Zen 2 = 3000 series desktop) but then AMD went and made the 5000 laptop series split between generations, a 5700U is a Zen2 part and a 5800U is a Zen3 part.
Intel did split the 10th gen but even then they changed up the naming between the two series.
Heh, I think the best we can do is embarrass them and complain when they do something stupid and confusing like some of the Ryzen 5XXX series being Zen 2 and others are Zen 3 cores. Not that Nvidia and Intel hasn't done the same thing.
Using http://ark.intel.com has been very useful for figuring out the details of Intel CPUs. Not found anything as useful for AMD or Nvidia.
I'd love to see proper testing of all the current chips from both AMD and Intel with Specter v2 mitigations applied. Performance hits of up to 50% are being shown in some workloads.
AMD really also should be benched with KPTI turned on, since they've got a Meltdown-style vulnerability (discovered by the same team) they've left unpatched because "KPTI fixes it", but they also want KPTI off-by-default because it tanks their benchmarks. It also completely breaks KASLR (and that's been broken on AMD for a while thanks to prior work from the same team).
Sadly people don't seem to take AMD vulnerabilities very seriously, Specter/Meltdown were trumpeted from the rooftops but when AMD leaves vulnerable defaults because mitigations would tank their benchmarks then it's fine, and everyone continues to benchmark in vulnerable configurations since it's "manufacturer-recommended".
There seems to be a mindset for many that because the initial vulnerabilities didn't affect AMD that they're invulnerable forever.
I can't think of anything less relevant for gaming workloads than spec. ex. security. In that use case you'd obviously disable all of the mitigations, to get the highest possible speculative performance.
The biggest part of this announcement is the bios updates for 300 series motherboards in my opinion. People who bought into the AM4 ecosystem 5 years ago should be able to update to these brand new zen3 cpus.
Have we ever had such a long lived socket and chipset before? Supporting brand new products for 5 years?
Socket 7 was around for a long time and supported a multitude of different processors from different companies. You could usually put a socket 5 processor into a socket 7 board as well (but not the other way). And that was a lot larger jump between the first and last processor.
It looks like AM3+ went from 2011 to the first generation of Ryzen launch in 2017. Still you are right, socket 7 takes the crown for longevity and performance boost.
Anecdotally, the increase in performance from my Ryzen 5 1600 to the Ryzen 7 5700G was roughly a 70% increase in performance. Despite that feeling like a solid upgrade it is nothing on the practical doubling between Original Pentium/AMD k6 releases in performance year over year.
Probably depends on your definition of "shortest lived". There's quite a few sockets that only got one generation.
AM1 comes to mind, there were only ever about 5 processors compatible with it and really only two were ever intended for retail market and only one of them ever really existed off the drawing board. That's probably one of the rarest consumer sockets ever produced, at least recently.
There have also been a few HEDT sockets that saw short life and small numbers. The W-3175X platform and AMD Quad FX (Quadfather) both were exceptionally short lived in themselves, but on paper supported a decent number of server processors due to socket compatibility. They are probably some of the smallest numbers sold. Quadfather never lived past a single board and neither of them probably sold more than a few thousand units.
TR4 and TRX40 both went down as exceptionally short-lived HEDT sockets with only one generation each and no cross-compatibility with server chips that share their sockets. But they probably sold higher numbers than W-3175X and Quadfather.
WRX80 is probably one of the lowest-volume sockets around since it's basically OEM only and a niche of the niche HEDT market, but again then you've got two generations, even if the volume sold is small.
Most Intel products really aren't contenders here, "one generation for the socket" is table-stakes here, it's volume that really decides it imo. Nobody can really say Intel doesn't produce volume for their stuff, a short-lived socket for Intel probably sold 100x the amount that Quadfather sold. Weird niche products like W-3175X or Kaby Lake-X are the exception but again then you've got other products in the same socket.
It reduces potential e-waste but depending on the board there certainly will be issues getting the best performance out of the newer zen 3 cpus. No pcie 4.0 support being a big one.
Right, but that's my point, if PICe 3.0 is not a significant bottleneck to an RTX 3090, it can hardly be seen as one for anyone running non-enthusiast builds.
Most games can run fine on Thunderbolt external GPUs, and that's a PCIe 3.0x4 link (roughly) that, on most Intel implementations, is also bottlenecked by the chipset[0]. Game assets are typically static, so available bandwidth mainly impacts how fast you can stream data into VRAM. Of course, with things like DirectStorage and consoles all embracing SSD-to-GPU streaming and hardware decompression, game developers might actually bother to actually use the extra bandwidth they've been given. However, that's still a way's off and definitely not a requirement to enjoy most games today.
The reason why PCIe 4.0 actually became a thing was because of enterprise storage arrays. M.2 slots and U.2 connectors don't have enough pins for 16 lanes, and using up so many of those lanes for one device makes no sense if you need to stick 10 or 20 of them in a server. That's also a use case that doesn't really make sense on AM4, unless you have a bifurcation[1]-capable motherboard or are spending way too much money on M.2 carriers with PLX chips in them.
[0] AFAIK, Intel wanted Thunderbolt direct-to-CPU but there was some weird driver/certification nonsense with Microsoft or something, and going through the chipset apparently made it easier for vendors not named Apple to support it. I don't remember the details.
[1] The ability to drive multiple PCIe devices off the same slot by splitting the slot's lanes. Most M.2 carrier boards are wired up to work this way because proper PCIe hubs are absurdly expensive... because the entire market for such chips are just storage array vendors.
Not generally, but there are a few edge-cases. AMD recently released the 6500XT GPU which has PCIe 4.0 but only x4, and when used in PCIe 3.0 motherboards that obviously becomes 3.0x4. Furthermore, AMD only put 4GB of VRAM on it, which isn't good enough for modern games (it's actually less than the previous 5500XT, where AMD made a big advertising push about "4GB isn't enough for modern games anymore") so it swaps all the time, which amplifies the PCIe bottleneck. The card loses something like 15% performance when used on a PCIe 3.0 motherboard. It's not like it won't run, but that's a significant amount of performance, that's like a half a tier of performance and far more than you see on (eg) 3090.
Unfortunately we're still in something of a transition period. AMD blocked vendors from supporting PCIe 4.0 on 300/400 series boards that might have been capable of the required signal integrity (particularly on the first slot). AMD doesn't support PCIe 4.0 on their Zen3 APUs at all either - and some of the new processors are based on the APU die even though they don't have graphics, so they are limited to PCIe 3.0 as well. And obviously Skylake/Coffee Lake/Comet Lake stuff is all PCIe 3.0 based since they're ancient. So there are definitely scenarios where you might think "throw in a cheap dGPU" and are still stuck with PCIe 3.0.
Anyway though, what I would caution you here is, the 3090 has lots of VRAM, so it doesn't swap. What the 6500XT shows is, low-end cards can be more susceptible to PCIe bottlenecking - they have less VRAM, so they swap more, which increases the pressure on the PCIe bus. 3090 results are not representative of a worst case scenario just because they do drawcalls really fast (high framerate), there are other parts of the pipeline where PCIe load can be generated. If you are swapping due to low VRAM, that's still PCIe load.
Similarly - simply using a card placed into a PCIe 3.0x4 slot is not representative of Thunderbolt 3 results either, despite both links being 3.0x4. Thunderbolt is usually implemented as an additional standalone chip attached to the chipset - so it's not a CPU-direct link, there's multiple hops with higher latency there. There is also contention for the chipset bandwidth - the chipset has a 3.0x4 link, so the GPU alone can saturate it, but there's also NVMe traffic (particularly for pre-11th gen Intel where NVMe has to attach to the chipset) and network traffic (chipset provides the network controller), etc. It's bidirectional so you can read at 4GB/s while you write 4GB/s to the GPU, but there's also just general contention for commands/etc. So performance results on Thunderbolt will be worse than a card attached to the chipset, which will be worse than a card attached to CPU-direct PEG lanes, even if lane count is the same.
(the exception to this might be Ice Lake and newer Intel laptop chips, where the Thunderbolt controller is actually part of the CPU itself, the performance impact of that should be less. However, this does not apply to desktop chips, including Alder Lake.)
AMD does seem to be trying to goose PCIe 4.0 demand by making some of it's products perform best with it (which is a mistake IMO, since it makes their competitors products more appealing) but it seems like the 6500XT was originally a mobile chipset so it wasn't really meant for the DIY desktop market. But shortages meant it was pressed into this duty anyway.
I was reading a techpower article on the 6500XT recently, because their power usage testing is excellent aand the 6500XT has an incredibly low idle draw versus other desktop cards.
I get it's good, but what took them so long? Wouldn't many people who had 300 series motherboards already have upgraded? They were stuck for 18 months without a path forward.
They're trying make switching to Intel less appealing when upgrade time comes. Until 12th gen, they didn't seem to view that as much of a threat. Now it apparently is. Competition is good!
interesting. I hadn't considered AMD doing this due to Intel 12th gen.
If true, that's a sign that AMD isn't doing this to help customers, but to hurt intel
> AMD isn't doing this to help customers, but to hurt intel
Seems correct. ASRock already allowed 5000 series CPUs on their X370/B350/A320 motherboards with a beta BIOS released in 2020[1]. AMD told them to stop[2]. That beta BIOS was never officially released.
AMD, like Intel, was almost certainly gonna take advantage of a lack of competition to try and maximize their returns from their existing chips. Without competitive pressure from Intel they have no incentive to release better chips.
It’s not that AMD is good and Intel is bad, or vice versa. It’s that only Intel is bad and only AMD is bad.
What we need is a healthy back and forth competition between both the companies (and some ARM sprinkled in as well).
I bought a 1700 in 2017 when they launched, upgraded to a 3800X 2 years ago and now I might even get to run a 5800X on my B350 board that I paid CA$130 for.
This is what good competition looks like. As a customer I love it!
I am sure they aren't BUT it might push them to innovate to make the motherboard itself more interesting? Alternatively, money-wise if I had known that the board might last me 7-8 years I probably would've bought a better one that (probably) has a better margin for Gigabyte.
For example, my B350 doesn't have a USB-C port, LAN is only 1Gbps, only one M.2 SSD slot.
> How far we've come. I remember having a semi-decent on board 100mbps LAN was only reserved for highest end.
These cpus are much more than 10x faster than the CPUs you would have used then.
Networking speeds in desktop devices haven't kept pace. Not just with cpu speeds but with storage speeds-- which is particularly obnoxious when you want to use network attached storage.
I assume this is due to a mixture of internet and wireless creating an extremely low bandwidth least common denominator and that running >1GB over copper is kinda problematic (finicky, power hungry, etc) -- and the industry seems to reason (perhaps correctly) the the customers in this segment can't handle fiber.
I personally have 40GBE attached desktops (and 100gbe attached stuff within the server rack)-- thankfully quite economically due to surplus enterprise hardware, so I'm well aware that it can be done... but to do it in a small formfactor system is still a problem, e.g. for any of these non-apu systems your normal mini itx systems will use all their slot space for a graphics card and not have room for a mellanox adapter.
> These cpus are much more than 10x faster than the CPUs you would have used then.
Actually, it's not only about the CPU performance, because NIC performance drew a cosine wave w.r.t. processor power over the years. Earlier NICs had much more machinery inside them and were expensive. Then Realtek came and we witnessed the era of (almost) software LAN adapters, then the silicon became cheap, and some of the stuff moved back into the LAN controller. So modern, higher end cards do not do significantly more offloading, but they handle other things like virtualization of cards between many VMs, or other higher level tasks.
From what I've seen, making a faster CPU is easy, but a faster and wider fabric is harder to weave due to distance, switching related noise and other physics related phenomena. Cache and good memory controllers are also relatively easy, but expensive. Also, RAM and higher voltage doesn't go well together, because RAM uses a lot of energy for its size, so it comes with heat and stability problems and other lifetime related problems (yes, you can fry RAMs).
Storage is another hard technology. Spinning rust is limited by physics, vibration, rotation speed and other mechanical stuff. SSDs came crashing, but before they were cheap and Sun didn't got swallowed by Oracle, they did nice ZFS arrays with some very expensive SSD caches and spinning drives combined. I've seen whattheheckisthat amount of speed from an array with a size of 6-8U total and a couple of IB interfaces fitted to it at the factory. Currently we get that amount of speed from Lustre arrays with some SSDs and a lot of spinning disks.
With the current backbone capacity of a standard desktop computer, even on the high end, 40Gbps network is overkill unless you're going to ingest that data directly at the CPU. Yes, absorbing the data at the disks are possible, for what cost and what use case? 10Gbps is understandable, but I still consider 1Gbps as a sweet spot for general purpose computing. If you're going to transfer vast amounts of data from a storage array to a local system, yes you can go higher, but it's still a niche. Also as you've said, going over 1Gbps is problematic from a signal integrity point of view, and fiber is too fragile and expensive for average customer.
These Mellanox cards get quite hot when they're constantly utilized, so in a little ITX box, both the card and the system will be cooked. The process becomes faster if you use active fiber cables, because connectors sink the heat inside the case via heat sinks attached to the connectors on the cards. Even if you use them in a system room, building a large network with them requires expensive equipment which has adequate bandwidth and cable length. Also, running an IB network needs other software machinery to keep the network up. You can directly run ethernet, or TCP over IB, which kinda defeats the purpose and adds additional penalty and reduces the performance.
All in all, higher speed network is still doesn't bring too much of a value to the home of the average user, and is not still that cheap for the enthusiast. Also the hardware is not scaled down for the home. Cards are big & hot, connectors/cables are bulky and the fabric gear is noisy, bulky and high maintenance for home.
Yes they work rather nicely, but I'd not want them at home. I have enough of them at the system room already.
Mostly agree, and just bought a ZFS server for home, 5 disks (RAIDz2) and 2xNVMe 2TB for root (256GB) and cache (the rest).
GigE is pretty annoying in that it can't keep up with a single spinning disk (let alone 5), so large files (family videos, movies, system images, backups, etc) can take annoying long. Hell even wifi access points are becoming bottle necked by GigE. Fortunately 2.5Gbe looks like a reasonable compromise. Rarely a bottleneck and support is common on cable modems (motorola among others), access points, motherboards (pretty much anything newish from AMD/Intel), and even small SBCs if you run one as a firewall, router, or pi-hole.
There are some switches and routers around and are often fanless till the port counts get high.
Even quality 10G cards seem to run pretty hot, have huge heat sinks, and the sfp+ I've worked with tend to get pretty hot. Seems like there is a newer gen 10G chip, but they are rare and expensive but do run substantially cooler.
Boards like the Intel quad port 10G X710-T4L (the cool one) vs the X710-T4 (the hot board) run at about half the power (14 vs 29 watts). I'm shopping around for the dual port version (X710-TSL) which peaks at 9.6 watts and uses 8.2 watts typically. No idea how other (non-intel) 10G boards do on power use.
40GB is overkill, but 10GB still bottlenecks for me. I'm surprised to see anyone consider 1GB a sweet spot-- it can't keep up with a single conventional disk, and it's causes obvious slowdowns for fairly conventional operations.
And indeed, the heat from higher speed cards is a nuisance... though 10G copper is awful in that respect.
> too fragile and expensive for average customer
I think that's a bit overstated. It's more fragile but not enormously so. Terminations are a nuisance, which favors pre-made patch cables. I'd pretty much always prefer to use 10g over fiber than copper.
But I'll freely admit that having spent more than a decade as a network engineer my expectations may be weird. I will say that after many years of using fiber connected equipment at home for the desktop/server systems, my spouse has never broken a cable or connector (nor have I, of course).
I've never-- professionally nor personally-- had good luck with copper 10g and try hard to avoid it. If I have to use well made/tested pre-made patch cables to make it work I might as well use fiber (or dac cables) and trust that literally anything I buy will just work.
> Also, running an IB network needs other software machinery to keep the network up.
I used to use IB because the IB only nics were super cheap on ebay-- but I don't anymore. It's easy enough to use 10/40/100GBE. The extra software was a nuisance compared to the value it provided.
> and is not still that cheap for the enthusiast.
I mean, you can get 40gbe capable cards on ebay for like $50ish (usually generation or two old mellanox dual ib/eth cards)-- main cost to them beyond their power consumption is knowing how to use the utility to put them in ethernet mode. Dac cables and LR2 optics at similar prices. Particularly if most of your usage can be covered by dac cables, I think it's pretty affordable-- not 1gbe affordable, which is close to free on on a cost per port basis, but when you're bottlenecked on network bandwidth there isn't really an alternative, except for keeping data locally on the systems.
> But I'll freely admit that having spent more than a decade as a network engineer my expectations may be weird.
I think you're correct in that regard. I'm an HPC administrator for ~15 years, and I manage quite a few systems with a sizeable storage and related network machinery up to a certain point. While I can understand your point, I've written my comment from the eyes of a bona fide standard consumer which doesn't do anything higher end on their homes.
> 40GB is overkill, but 10GB still bottlenecks for me. I'm surprised to see anyone consider 1GB a sweet spot-- it can't keep up with a single conventional disk, and it's causes obvious slowdowns for fairly conventional operations.
My sweet spot comment assumes streaming content consumption, occasional system to system file transfers, in home entertainment streaming and some gaming maybe. If I'd have to move my photography or video archive to a NAS instead of internal drives, a fatter network would be indeed required. To be bottlenecked by 10gbps, you really need to have a busy network with a lot of going on. None of my personal networks carry that amount of data any given time, and they don't need to do it to be honest. Moving 10gbps fairly constantly is not fairly conventional, TBH.
> I think that's a bit overstated. It's more fragile but not enormously so. Terminations are a nuisance, which favors pre-made patch cables. I'd pretty much always prefer to use 10g over fiber than copper.
Yes, modern fibers can endure relatively unbelievable amount of abuse, but it's still well inside the boundaries of a normal user's limits, and the threshold is still lower compared to fiber. Many people I know can damage a fiber without intending or knowing it.
> I used to use IB because the IB only nics were super cheap on ebay-- but I don't anymore. It's easy enough to use 10/40/100GBE. The extra software was a nuisance compared to the value it provided.
When you're doing HPC, the value provided by the features are indispensable. Latency, MPI offloading, DMA are all essential things you need for a really performing cluster. So we prefer IB rather than Ethernet. So horses for courses, obviously. I'd also rather not use IB for a home network.
> I mean, you can get 40gbe capable cards on ebay for like $50ish...
Consider the price as the money + setup + maintenance. For you, the last two items have no cost virtually. For me they'd be negligible. For a regular consumer, they're much bigger items to wrestle with.
So, as a people who knows how to play with bigger tools is really a privilege and luxury. However, I have enough servers at work, and I don't want to manage stuff at home. So if that lower and, dumb stuff works as I want and doesn't make my life worse, I'm a happy camper. As a result, I over-engineer my home stuff very minimally.
I try to push my builds as close to 10 years as possible. I've had good luck by trying to time it around important motherboard features that make a big difference & allow for minor upgrades over time.
- Caveats - These are software dev & light multimedia editing computers. By the time I get rid of one, it's had as much RAM added as possible. I've done this twice now. Would not work on a gaming rig.
Hm, also some features (e.g. USB C/3.1 headers) were only available on high-end gaming MoBos for quite a long time, making the whole build power-hungry and needlessly fragile.
Before that, MoBo manufacturers even made extensive use of additional chips for networking, USB, mora SATA ports, RAID and what not, which could make the whole system more complicated/fragile (all the additional drivers!).
I nowadays try to avoid that and aim for boards close to "reference" (whatever the chipset offers out of the box).
I didn't mean to call out anything against gaming rigs. Some of my components have been marketed towards gamers. Many of the better motherboards often are, excluding the very serious motherboards.
I just never bought a very expensive video card & I can't comment on how long a good one would last. I imagine it depends on to many variables from the games you play, the settings of the games you're okay with, the card, etc..
I go for more like 8 years, because I like some overlap so I'm not stuck without a system.
Generally I'd buy midrange parts, from well known vendors, and pay a bit extra for SSD, ram, etc. Avoid anything with a small muffin fan (primarily chipsets these days), get a quality case, 140mm or bigger fans, quality power supply (a 5 year warranty is a good start). My current system is from 2015 and was a Xeon E3-1230 v5, 32GB ECC ram, GTX 1070, and a M.2 NVMe. The E3-1230 was cheaper with ECC then the similar i7 was without and just slightly slower (3.2 vs 3.4 GHz). Had no problems so far, and I'm just starting to look at a replacement. Looking closely at the expectations for Zen 4 CPUs and the current Alder Lake (or follow on ). I'm definitely sold on DDR5, especially since with a 2 dimm system (128 bits wide) you get 4 channels instead of 2 with DDR4. Should be that much better at not stalling your cores on cache misses.
If I bought today it would be tough decision between the AMD 5700x (8c/16T with a 65 watt TDP) vs the similar Intel Alder Lake. I do think AMD has the edge for systems that are low power and Intel the advantage for high power. Personally I'd rather have quiet, cool, and reliable instead of an extra bit of performance that nobody will care about in a few years. I'd consider a Apple M1 mini ... if it had the M1 pro or Max. Don't think the studio is quite worth it, but I am chipping in few $ a month to Marcan for a linux port.
you can always get a 10gbe (or 2.5gbe, 5gbe, etc) network card for $50 or so, and M.2 NVMe can be added with $10 pcie addon cards as well (it's just a physical adapter to the pcie slot, electrically it's still nvme).
unfortunately now we get into the discussion about 3.5 slot GPUs that overhang all but 1 or 2 of your pcie slots, and the general lack of PCIe connectivity on consumer boards... you can sorta work around it sometimes with vertical mounts/etc. But it's annoying and takes work/planning.
I miss the days of X99 and having 28 lanes at my disposal, and every GPU being 2-slot or at most 2.5 slot such that I could actually use those lanes.
It's also been really annoying watching the arms race between GPUs and motherboard slot spacing. For a long time, GPUs were 2 and 2.5 slot, so motherboards went to 3-slot spacing for their slots. But then 3-slot and 3.5-slot GPUs became common, which overhangs the middle slot (or puts the GPU right up against a card placed in the middle slot). What we really need now is for motherboards to go back to 2-slot spacing so you get a slot in the #4 spot to give 3- and 3.5-slot cards some breathing room...
This isn't just good competition. It's damn good engineering. We have to remember that this wasn't something that AMD was forced to do by the competitive landscape. Someone pushed hard for this decision internally, and that person or team deserve recognition. Someone also had to actually build it, and they deserve even more recognition.
Competition is fantastic, but the people who rise to the competition are, in my opinion, the real heroes.
I remember something that there was a technical issue with that "compatibility"... Was it something about power ratings or the chipset/MoBo becoming such a bottleneck that this was essentially useless?
Or am I just thinking about an entirely different thing...?
It really does not fit the narrative of what AMD did with Zen so far, that's for sure.
If you go back to last year you will see a lot of AMD fanboys arguing such a thing, based on pure conjecture. One person replied to one of my comments saying 5000 can not run on 300 series BIOS because of [insert-technical-hypothetical-nonsense] reasons. But suddenly... it can!
The reality is that AMD fanboys haven't woken up to the fact that AMD is no different than Intel. All the same marketing and segmenting games exist on AMD. They were the underdog and then suddenly, they are on top and acting just like Intel.
No. There was a lot of rationalizing from fanboys attempting to retcon a technical justification for what was clearly a business decision from AMD, but none of the technical explanations hold up.
Most of the stack has a similar TDP to Zen1/Zen+/Zen2. The 5950X has the highest TDP, 105W, which is the same as the 3950X, which is allowed. In practice the 5950X actually pulls a bit less power. Furthermore, the A320 boards (even lower-end) got official support, despite a lot of those boards having utter trash VRMs.
(furthermore, there's nothing that guarantees that every A520 board has a VRM that is capable of supporting a 5950X either! it's not like board generation is some guarantee of VRM quality, there are X370 boards that can handle a 5950X and A520 boards that can't. That's not how support is determined.)
16MB bios isn't a limitation either. B450 boards are the literal exact same silicon - B350 is to B450 as Z170 is to Z270, basically, it's rebranded silicon - and B450 boards with 16MB BIOS got support. There are B450 boards with 16MB bios that support literally the entire range of chips. And once again, so did A320 boards with 16MB BIOS.
Releasing official support for A320 really just blew a hole in every retconned justification that people tried to make for AMD. It's uncouth to say it here on HN, but there really is a large number of people who are super emotionally attached to the AMD brand and willing to stretch to absurd lengths to justify what are clearly business decisions.
And now they're just reversing their previous policy. Wasn't a problem at all, actually.
What you probably mean someone pushed hard internally for a decision to NOT limit old platforms.
From a technical point of view "Chipset" no longer exists. Every single chip going into Socket AM4 is a SoC able to work 100% standalone. Any differentiation and market segmentation you see around "chipsets", which are by the way just a fancy northbridges hanging off standard PCIE, is fake profit maximization theater.
So please help me understand: I have an MSI X370 mainboard, and as far as I understood until now, the Ryzen 7 3700X processor for around 300 Euros is the newest one I can get that still runs on my system (that would be relevant for being able to update to Windows 11, since I have a Ryzen 1700 now).
Now I can use the 5800X3D, but nothing in between those, or is there some other wrinkle I haven't understood?
Yeah, people who didn't get a board with the best VRMs must be kicking themselves :D
Really though, there's some cheap mATX boards with high quality VRMs that can run a Ryzen 9 with no issues. Though tbf, if the processor starts but is not stable, an undervolt or maybe clock cap could work, at the cost of a slight performance drop.
It's nice to have the option to just upgrade the processor.
BTW my system is limited to 4K30hz on the HDMI for reasons I never figured out. Haven't found a good Displayport->HDMI adapter to fix it, and not sure why the CPU/mobo combination won't do it on HDMI. I don't do gaming on it so not much of a problem, but I'd prefer 60hz.
Under the HDMI section indicates up to 5Gbps video bandwidth (in addition to 8 channel audio). So on the surface it sounds like I shouldn't even get 30fps, but I do. Thank you for the info, it looks like the board isn't up to the task via HDMI so I will make the effort to find a Displayport adapter. That PC has a LOT of life left in it and I'd like to get the most out of it ;-)
This has been my biggest pain point with Intel. Why are sockets changed, seemingly, every generation? There's no point in having a socket in that case. Might as well solder the CPU right to the board.
I don't think so. Selling CPU and chipset for DIY market isn't very profitable part of business (but customers are vocal and likely influencer). It adds compatibility test cost so somewhat for money.
Part of the answer there is it's easier from an engineering standpoint to get a new socket than it is to make a new architecture work with an existing socket.
Sort of the same way in the software world where it's easier to add a new method then it is to enhance an existing method. New methods have no legacy baggage to worry about. Tweak an old one wrong and you run the risk of breaking a bunch of people.
AMD's commitment to forward compatibility is pretty nice, but definitely more work on their part.
Then again, AMD's legacy is making sockets/slots work for their processors :D (That was the whole point of the original athlon)
Does this mean A300 boards may get official support for the Ryzen 5 4600G? Because I'd love to slap one into my DeskMini A300. With a (now pulled) Beta BIOS, an A300 ran a Ryzen 7 Pro 4750G just fine, and it'd be a big jump from my R3 3200G.
It sounds like it depends on your board’s manufacture providing it for your system but it looks like AMD is releasing the BIOS firmware to allow x370, b350, and a320 boards to support all the new cpus just announced today.
In some cases your existing cpu might not be supported using the new firmware. So if you update the bios, you might have to swap the processor before you can boot the computer again.
A300/A320/X300 were always the exception to the Zen3 lockout code. You could run a 5950X on a $30 trash A320 board with a 16MB BIOS and a 3-phase VRM, but not a Crosshair VI Hero.
anyway, there's nothing stopping vendors from releasing support for A300, Asrock simply has abandoned the Deskmini A300 and doesn't want to support it anymore. That one isn't on AMD, Asrock just seems to have abandoned it.
AMD was aiming for that when AM4/Zen was announced, knowing that this was a major pain point for enthusiasts and builders/"upgraders".
This would have been my main reason to build an AMD system, although I was easy to win, I always built AMD systems IIRC... And I ended up postponing it again and again, since I don't game and the reasonable (budget) options were always lagging behind regarding technology/efficiency - most of all, my Phenom (actually a rebranded Athlon with 2/6 cores disabled) was still doing OK. Now there will be AM5... Didn't follow the news, is it supposed to live as long as AM4?
Would it even pay off for AMD to repeat that effort?
The differences between the 5600 and 5500 are interesting. The $199 5600 has 6 cores, 32MB of L3, no GPU, and uses the chiplet packaging. The $159 5500 has 6 cores, 16MB of L3, a disabled/fused off GPU, and is one larger, monolithic die.
I thought the chiplet packaging was supposed to result in lower costs due to better yields on smaller dies? I guess the extra cache takes up more than enough die space to make up for that?
The 5500 is less performant, therefore lower cost. It only exists to sell salvaged dies, and won't likely appear in large quantities (similar to how the 3300 / 3100 were impossible to find).
None of their chiplet designs currently have GPU. All of their APUs are currently monolithic. The 5500 is just a neutered 5600g. Probably mainly trying make some money on dies with bad GPUs.
The yield issue is going to be much more of a factor when a process node is young and defect rates are high. The big savings is in engineering hours where you can re-use a design across multiple products.
Everything I've been reading has been that 7nm has been having excellent yields and chiplets do bring their own costs. Lisa Su (AMD CEO) indicated in an interview a while back that there is a price floor at which chiplets make sense. Another thing to consider is that most of the chiplets used in the consumer parts could be considered rejects from Epyc. (i.e. they need to make a lot of relative duds to get the small number of 'golden samples' they use in their server chips)
the 5500 is a monolithic die, it's based on the monolithic APU lineup rather than the chiplet enthusiast lineup.
This also implies other limitations like PCIe 3.0 and half the cache of the equivalent enthusiast lineup, because that's how the laptop chips were designed. The APU lineup has always performed a bit worse than the equivalent desktop lineup as a result.
Lower clock speeds but more cache. Also interesting that they stopped at the 8-core chips and aren’t adding the cache to the 12 and 16 core parts.
It looks like the extra cache die isn’t good for thermals, so maybe it’s not viable on the 12 and 16-core chips without sacrificing too much clock speed.
Looks to be a strictly gaming-focused CPU play. Not a bad move given the way the enthusiast market works, but it doesn’t do much for a lot of non-gaming workloads.
> Looks to be a strictly gaming-focused CPU play. Not a bad move given the way the enthusiast market works, but it doesn’t do much for a lot of non-gaming workloads.
How so? 8 cores still covers plenty of light-workstation workloads, where the extra cache should lead to an overall performance improvement (despite the minor drop in top clock speeds). With current software architectures, it's still hard to make full use of 12 to 16 cores.
you're rather missing the point about the exclusion of 3D V-cache on the higher-end processors. If it was a productivity-focused play then you'd see it included on those as well.
The 5800X is a good processor and it's true that people do productivity on it, nobody disputed that and there's no need to rush to its defense. But you've missed the point about the overall way the upgrade is (or rather, isn't) being rolled out.
Parent comment pointed out that the whole 3d-cache business might come with challenging thermals on high-core-count desktops. There might be all sorts of technical reasons why V-cache is not being used in this particular segment as of yet.
thermal problems are obvious based on AMD disabling overclocking on the 5800X3D. Even when gains have become minimal, they didn't explicitly disable them, so clearly either potential gains are literally zero, or overclocking might actually damage the cache/lead to conditions where it could be bricked or lead to glitches that might allow control of the PSP (the cache die being separate, and completely exposed, certainly is an interesting attack "surface").
Thermals have always been the giant asterisk on die stacking. Everyone knows it's going to be a problem. But this is the first consumer stacked-die product (that I am aware of?) so we don't really have any sense for overclocking on those processors. The tentative implication here is - bad. Even with a single cache die, it's still got a whole processor underneath it heating it up. There are implications here for GPUs as well, as both AMD and NVIDIA are expected to release stacked-die products next generation. It's looking like they will have to keep clocks under control to make that happen - maybe that is a counter-argument to the "500-600W" rumors (for both brands).
But multi-die products have twice the area to dissipate their heat over - just like 5950X doesn't run at higher temperatures than 5800X. Higher power, yes, but twice the area to dissipate it over means thermals are about the same. That's not really the reason.
A single-die limitation also wouldn't rule out a 3D version of the 5600X - it's no longer "the bottom of the stack" and finally there are value SKUs underneath it, it would be appropriate to re-release it as a more performance-oriented 5600X3D SKU.
Anyway, my personal opinion is this is going to be a very limited product that primarily exists for "homologation". AMD can say that they've released it, it's officially a consumer product, so it can re-establish AMD's place on top of the gaming benchmark charts, but it's going to be a 9900KS-style super-limited SKU that doesn't see any real volume production at least until AMD has sold their fill of Milan-X. It just exists to put AMD back on the top of the benchmark charts now that Intel has retaken it with Alder Lake.
The 5800X is the best processor for them to do that. 5950X sets up some small regressions for inter-CCD problems/etc. 5800X is enough for games for now anyway.
Is this a sign zen4 is delayed? I get they need new motherboards and socket, and DDR5 etc.., but launching zen3 stuff now is strange. A lot of us already upgraded.
I don't take it that way. Zen 4 is likely to only release high-end parts at first. Given the timing of these releases, I'd expect Zen 3 to serve as the mid- to low-end of the product range at least through mid- to late-2023 if not longer. Given the way AMD sandbagged these parts this generation, they'll probably hold off as long as possible next generation for low-end Zen 4 too.
I couldn't get past the scalpers in 2020. That was already a _slightly_ delayed refresh driven by a combination of less than stellar AMD GPUs (Linux, so nVidia's not a good option) and cryptominer hell to that point.
It's going to be 2 years later, and at this point I'd rather jump directly from DDR3 to DDR5.
Not heard that, but Alder lake has taken back most of the market share the Zen 3 managed to take from Intel. So a response sooner than later will help protect marketshare. In particular they are doubling down on the lower power chips that ADL is having problems competing with. ADL is fast, but power hungry. Thus chips like the 5700x with a 65 watt tdp instead of the previous 8c/16t chip with a 105 watt TDP.
I'm debating whether to buy this. I have an 1800X on a B450 board with DDR4 3000 RAM. I use the PC for gaming and compiling, basically, so I should see huge benefits from the extra cache. I could double my CPU performance and breathe new life into this computer for not much money.
However, that computer is getting long in the tooth. The RAM is slow compared to modern stuff. The motherboard is cheap, has poor UEFI, and I zapped the ethernet port with static - destroying it. Do I really want to invest more into it when I could wait 6 months and build an all new PC with DDR5 and Zen 4? It would be a lot more expensive but I have a lot more money now than when I built this computer.
3000 RAM is good enough. If you upgrade the CPU and GPU you'll get effectively the same performance as any shiny new build and can last another 5+ years. You can get back the ethernet port with a PCI card or USB.
That being said, if you can afford it get the shiny new stuff. You deserve it.
I like this advice. Both practical (I agree with all points) and yet acknowledges the emotions we feel when buying stuff with a side of financial sense.
Did you read the article? The AMD Ryzen 7 5800X3D is being released for gaming performance. I checked and most of the games I've played over the past few years do not even launch on macOS. That's kind of a problem for using Apple Silicon in a gaming PC (among other problems).
Crazy how people would prefer spending 300 - 450 USD to upgrade a rig they already own, instead of throwing it away and buying a apple device with a new M1 cpu that costs +700 USD
Er, studio is minimum $2k, well above the average price of a home built AMD system with the same specs (32GB ram and 512GB SSD). It also runs OSX instead of Windows and Linux,
Not really the same market at all. The 5700X (2nd most expensive on a list of 7 CPUs) costs $300, 32GB ram of DDR4-3600 is $150, 512GB SSD is $60, motherboard, case, and power supply is likely another $300. So $810 for a system, not including a GPU, but $1200 is more than enough for something decent.
Not to discount how cool healthy competition is, but i recently got a Ryzen 5 1600 (used), which came out in 2017 and its 6 cores and 12 threads have been enough for every workload that i've wanted to deal with, even in 2022.
It was a nice upgrade up from Ryzen 3 1200 that i had previously and honestly, i don't see myself upgrading again for the years to come, unless things become visibly slow and sluggish because of Wirth's law (let's assume that in 2030 most of my desktop apps would run Electron instances).
It does gaming, it does programming, it renders videos, it does 3D modelling and other content creation, it compiles code and runs Docker containers and oftentimes multiple of those at the same time with no overclocking and no issues (e.g. render a video in kdenlive while playing a game in the background).
Somehow it feels like either hardware has scaled to the point where new improvements are pretty incremental, or maybe software's ability to be wasteful with the provided resources thankfully still hasn't caught up with these improvements - you no longer really need to get a new CPU every 2 or 3 years, even the Ryzen 3 1200 which also came out in 2017 was adequate (i just got a good deal on the upgrade), which is really nice to see!
Even my homelab servers still run used Athlon 200GE CPUs from 2018 because of the 35W TDP and are on par with cloud resources that i can otherwise afford for my 24/7 available cloud stuff, just much cheaper when you consider how long they last.
Also, there's something really cool about every device in my home (apart from laptops) running on the same CPU socket and all of the parts being interchangeable with no driver update weirdness. Though it'd be even better if the Ryzen CPUs were the variety with iGPUs, given the GPU prices still otherwise being pretty unfortunate (in case my RX 570 would break).
The only reasons that i see for upgrading my setup in the next 5 years would be one of the following:
- the stocks of used CPUs drying up or prices rising (AliExpress still has them pretty affordable)
- the motherboards themselves getting deprecated or there being other unforseen issues
Here's hoping that won't happen and that i can enjoy CPUs that are suitable for my needs (and help avoid the e-waste issue at least somewhat), while those more well off financially than me can dabble with the more recent releases to their heart's content!
For the first time in 2022 I maxed out the 3600x I got when it came out (2018?). And it wasn’t a game, it was my girlfriends R script to do metabarcoding of some bacteria. 100% usage across every core sustained for 15min.
However what shocked me was that my base M1 pro was a few minutes faster. (Same amount of ram)
Still fast enough that it’s not worth an upgrade, but we’ll see once she starts transcriptomix
I do wonder if the days of dimms and long lived sockets of AM4 is over. AM4 maxes out at DDR4-3200 x 128 bit = 50GB/sec and more importantly just 2 pending cache misses. A bit more with overclocking, but not much.
Apple M1 = 128 bit x LPDDR4X-4266 = 68GB/sec and I believe 8 memory channels (8 pending cache misses). A modest core count (4 fast + 4 slow) helps keep the memory bandwidth from being a bottleneck.
The M1 pro doubles this to 256 bits, 16 channels, and 200GB/sec, which is a significant help for the integrated GPU and hits levels that AMD APUs can not match.
M1 max doubles again to 512 bits, 32 channels and 400GB/sec.
M1 ultra doubles again to 1024 bits, 64 channels, and 800GB/sec.
Not sure AMD can really compete and the APUs are severely limited by memory bandwidth, unless you buy a PS5 of XboxX. I'm hoping that AMD takes a page from the Apple playbook and ships a consumer CPU with dramatically more bandwidth and allows users to skip the current curse of discrete GPUs that run hot, are not available at MSRP, and are hard to get even at 2x MSRP.
I think you would find an APU that powerful would suffer the same fate as discrete GPUs in that it would be low-stock and/or expensive. Same thing with the M1 Ultra, starts at $2k it looks like
Dunno. Mining is all about hashrate/$ and scalpers are all about the % profit. Seems like AMD could easily make today's APU with 2x the bandwidth (less than 1/5th of the Apple Studio/M1 max) make a tidy profit and provide 2x the APU performance and make the APUs a good fit for a much larger fraction of the GPU intensive applications.
Not like they aren't shipping by the millions in the XboxX and PS5, there's obviously a demand for them and AMD is obviously capable of making them (they make the CPU in both).
159 comments
[ 0.17 ms ] story [ 21.3 ms ] threadi3-12100f is a 100$ 4 core chip, which should line up with Ryzen 3 4100 at $100 or possibly Ryzen 5 4500 at $129.
[1]: https://www.newegg.com/p/pl?N=100007627%20601394305%20601361...
[2]: https://www.newegg.com/p/pl?N=100007625%20601292786%20601312...
People hear about binning and assume that every product decision has to be related to binning, but usually it's not, it's just market segmentation. AMD had over 80% of Zen2 chiplets coming off the line with 8 fully-functional cores, and clock bins are generally selected such that most units will pass, by design. And that's at launch, on a new node, in 2019. Numbers have only gotten better over the last 3 years.
AMD already has a bin for iGPUs with a defect - it's 5600G/5600U/5600H/etc. And they have 5300G below that allowing even more defects. There's very very few APUs coming off the line with tons of GPU defects but 6 workable cores, or a defective PCIe controller but only a defect in the iGPU part and not the rest, etc.
The problem is that AMD has tons of supply of high-binned parts but the lowest demand for those parts. And they have the highest demand for low-binned parts but the lowest supply of those parts. How do you mesh those two curves? Disable cores on a high-binned part and sell it as a lower SKU. That's why those "black edition with unlockable cores" existed - those unlockable cores were locked off for market segmentation. Nowadays they just don't let you turn it back on.
(Which isn't to say that none of the 5600X/etc are the result of a dead core/etc - but a lot of them aren't, probably most of them aren't, given the likely >90% yields for 8-core at this point. And you pick your 5600X bin such that most 5800X failures can be sold as 5600X, meaning there's very little that falls through the cracks without being just utterly broken. True binning-generated "we have this pile of chips, let's do something with it" style SKUs tend to have extremely limited availability as a result, it's shit like the Ryzen 4700S or the Ryzen 3100, or the NVIDIA 1650 KO.)
Anyway, it's not a coincidence this is coming a few months after the Alder Lake launch. This is market-driven, Alder Lake is not only faster but in many cases it's cheaper as well. AMD coasted a little bit while motherboard supply firmed up for Intel, but they finally have to respond. I'm sure they're selling lots of Milan but consumer marketshare matters too, and AMD is losing steam there with the price increases, with Intel back on top on performance, and with Intel undercutting their pricing heavily.
The first digit is the generation and then the second digit and suffix letter designate additional features. So a Ryzen 7 5700X is a 5th gen Ryzen 7 meaning it has 8 cores. It is a step down in price and performance to a 5800X and also has no Graphics capabilities (those have a G suffix). Intel uses different letter suffixes such as the Core i7 12700KF being a 12th generation Core i7 with overclocking features (K) and no graphics (F).
It is far worst with GPUs. Certainly an industry wide naming convention reset would be nice but everyone is not going to cooperate like that.
It does take a minute to read up on but at least Intel has *mostly* followed this naming convention for 12 product generations so far.
I’m not sure what is better. Amds weird numbers of apples ultra max pro m1
similarly, marketing a processor as being a "5700U" or "5700G" despite it being slower than a desktop 5700X could be seen as equally deceptive. That's not really any better than "i7" vs "m7" or whatever.
unfortunately you're just going to have to learn the naming convention, there really isn't a good solution for that, given the wide range of applications that a given series might be applied to. You just happen to think that AMD's naming convention is worth taking the time to learn while throwing your hands up at the Intel naming. Same with people who think Intel is just awful for internally codenaming all their products after Lakes and Coves while eagerly memorizing every single painter and city name that AMD uses in their codenames - says more about your priorities than their naming scheme.
Who are you talking to?
That's normal. Mobile and ULV SKUs are usually coded differently. AMD calls them "U" and "H" skus I believe.
The person in question just doesn't understand the Intel naming convention, which is fine, but AMD and every other company does the exact same thing. It's not that Intel is uniquely confusing, it's that the individual here doesn't feel that Intel's naming convention is worth the brain space. Which is also fine, but it's not a problem with the naming convention.
As a tangential observation (people do make those in discussions!) architecture/product codenames are another place this comes up. There are many enthusiasts who will eagerly memorize that Rembrandt > Renoir but think the idea that Rocket Lake > Coffee Lake is perplexing and confusing. Or at least that's a repeated theme in many of these naming discussions.
I'm doing my best here to say this politely, but a lot of people clearly just don't value those two bits of knowledge equally. And dipping into rhetorical "who are you even responding to!?" doesn't really further the discussion either.
Naming isn't hard and naming discussions aren't interesting.
There is no solution which is going to be 100% intuitive to someone who specifically doesn't know anything about what they're looking for. If you move those products out to their own separate series, that's what Intel did with the Y series ("m7-xxxY" line - contrast to "i7"). You specifically don't like that. If you mark them within the existing series, that's what Intel and AMD do with the U series. You specifically don't like that either. If you move them into a single series, you end up with something like the Intel Ice Lake/Tiger Lake naming convention, where there is some part of the name that means "cores" and some part of the name that means "power" and part that means "graphics". Other people really didn't like that, because now you have one name that means 5 different things.
(And this is what I mean about the naming discussion being dumb and boring - whatever you think is how it should be done, someone else hates that, and thinks it is too complex and requires too much knowledge on the part of buyers. It's bikeshedding, product naming is low-stakes so everyone has an opinion on it and is very upset that AMD and Intel are ignoring their urgent forums posts. At the end of the day it's just not that interesting, nor are any of these naming schemes that difficult if you bother to learn what they mean.)
Anyway, it's unfortunate that there are features and distinctions which laymen may not understand, but that's a fact of life, there's things car people really care about that a Camry Buyer doesn't know, and that's fine.
Someone who bought an expensive truck with a base-model trim might be upset that a Camry with a top trim is "nicer", because they didn't understand what a "trim" is before they laid down their money, and that's unfortunate but it's not exactly hidden either, nor should we call to get rid of trim levels because one person didn't understand. Someone else might be really upset that their SUV doesn't tow like a truck even though they got the nicest trim level on the SUV.
Again, sorry if this seems frustrated, but this is a topic that has been bikeshedded endlessly. The stakes are low, there's multiple reasonable options available, and there's a whole lot of people who are all really upset that AMD and Intel aren't taking their forums posts on the topic seriously. Naming conventions are fine, they're good enough to not matter.
Both include a target market, generation, and performance level. With Intel, it's "i{market}-{generation}{performance}", with market being 3 (budget), 5 (mainstream), 7 (enthusiast), 9 (high-end). The numbers used in the "performance" level varies, but higher is always better. For example, i7-3770 is the third-gen chip targeted towards enthusiasts and was the highest performer in its generation. My i9-9900 is the top end model for the 9th gen. Intel will also use suffixes "K" to mean it has an unlocked multiplier (making overclocking easier) and "F" which means it does not have an on-die GPU, so you'll need a discrete GPU card.
AMD is similar, they'll call them "Ryzen {market} {generation}{performance}", ie, Ryzen 5 5600. But where AMD goes crazy is with the damn suffixes that express an additional performance level that is impossible to decipher.
A Ryzen 2700 is a second-generation CPU with 8c/16t.
A 3700 is a third-generation CPU with 8c/16t
A 4700g is a second-generation CPU with 8c/16t and a GPU
A 4500h is a second-gen CPU with 6c/12t and a GPU
There is no 4600G.
An X suffix generally means higher clockrate; an H suffix generally means high efficiency (lower power draw); a U suffix means ultra-efficient (very low power draw).
In the 5000 and 6000 series, possible suffices include X, H, U, HX, HS, but not G.
Some CPUs are OEM-only.
Some CPUs are only sold in packaging for laptop/tinybox manufacturers.
In general, if you get integrated graphics, you lose an entire processor generation, but more recently you only lose top-end features.
Intel did split the 10th gen but even then they changed up the naming between the two series.
Using http://ark.intel.com has been very useful for figuring out the details of Intel CPUs. Not found anything as useful for AMD or Nvidia.
https://mlq.me/download/amdprefetch.pdf
Sadly people don't seem to take AMD vulnerabilities very seriously, Specter/Meltdown were trumpeted from the rooftops but when AMD leaves vulnerable defaults because mitigations would tank their benchmarks then it's fine, and everyone continues to benchmark in vulnerable configurations since it's "manufacturer-recommended".
There seems to be a mindset for many that because the initial vulnerabilities didn't affect AMD that they're invulnerable forever.
With protections turned on
With protections turned off
Have we ever had such a long lived socket and chipset before? Supporting brand new products for 5 years?
Anecdotally, the increase in performance from my Ryzen 5 1600 to the Ryzen 7 5700G was roughly a 70% increase in performance. Despite that feeling like a solid upgrade it is nothing on the practical doubling between Original Pentium/AMD k6 releases in performance year over year.
AM1 comes to mind, there were only ever about 5 processors compatible with it and really only two were ever intended for retail market and only one of them ever really existed off the drawing board. That's probably one of the rarest consumer sockets ever produced, at least recently.
There have also been a few HEDT sockets that saw short life and small numbers. The W-3175X platform and AMD Quad FX (Quadfather) both were exceptionally short lived in themselves, but on paper supported a decent number of server processors due to socket compatibility. They are probably some of the smallest numbers sold. Quadfather never lived past a single board and neither of them probably sold more than a few thousand units.
TR4 and TRX40 both went down as exceptionally short-lived HEDT sockets with only one generation each and no cross-compatibility with server chips that share their sockets. But they probably sold higher numbers than W-3175X and Quadfather.
WRX80 is probably one of the lowest-volume sockets around since it's basically OEM only and a niche of the niche HEDT market, but again then you've got two generations, even if the volume sold is small.
Most Intel products really aren't contenders here, "one generation for the socket" is table-stakes here, it's volume that really decides it imo. Nobody can really say Intel doesn't produce volume for their stuff, a short-lived socket for Intel probably sold 100x the amount that Quadfather sold. Weird niche products like W-3175X or Kaby Lake-X are the exception but again then you've got other products in the same socket.
As long as people recognise thay a brand new CPU in a 5 year old board will have some compromises...
Which given the rates of #internetdrama these days, is probably not likely.
Doesn't seem like it (RTX 3090 is probably at the edge of what I'd consider "reasonably priced" anyway): https://www.reddit.com/r/hardware/comments/ikrteg/nvidia_rep...
The reason why PCIe 4.0 actually became a thing was because of enterprise storage arrays. M.2 slots and U.2 connectors don't have enough pins for 16 lanes, and using up so many of those lanes for one device makes no sense if you need to stick 10 or 20 of them in a server. That's also a use case that doesn't really make sense on AM4, unless you have a bifurcation[1]-capable motherboard or are spending way too much money on M.2 carriers with PLX chips in them.
[0] AFAIK, Intel wanted Thunderbolt direct-to-CPU but there was some weird driver/certification nonsense with Microsoft or something, and going through the chipset apparently made it easier for vendors not named Apple to support it. I don't remember the details.
[1] The ability to drive multiple PCIe devices off the same slot by splitting the slot's lanes. Most M.2 carrier boards are wired up to work this way because proper PCIe hubs are absurdly expensive... because the entire market for such chips are just storage array vendors.
Unfortunately we're still in something of a transition period. AMD blocked vendors from supporting PCIe 4.0 on 300/400 series boards that might have been capable of the required signal integrity (particularly on the first slot). AMD doesn't support PCIe 4.0 on their Zen3 APUs at all either - and some of the new processors are based on the APU die even though they don't have graphics, so they are limited to PCIe 3.0 as well. And obviously Skylake/Coffee Lake/Comet Lake stuff is all PCIe 3.0 based since they're ancient. So there are definitely scenarios where you might think "throw in a cheap dGPU" and are still stuck with PCIe 3.0.
Anyway though, what I would caution you here is, the 3090 has lots of VRAM, so it doesn't swap. What the 6500XT shows is, low-end cards can be more susceptible to PCIe bottlenecking - they have less VRAM, so they swap more, which increases the pressure on the PCIe bus. 3090 results are not representative of a worst case scenario just because they do drawcalls really fast (high framerate), there are other parts of the pipeline where PCIe load can be generated. If you are swapping due to low VRAM, that's still PCIe load.
Similarly - simply using a card placed into a PCIe 3.0x4 slot is not representative of Thunderbolt 3 results either, despite both links being 3.0x4. Thunderbolt is usually implemented as an additional standalone chip attached to the chipset - so it's not a CPU-direct link, there's multiple hops with higher latency there. There is also contention for the chipset bandwidth - the chipset has a 3.0x4 link, so the GPU alone can saturate it, but there's also NVMe traffic (particularly for pre-11th gen Intel where NVMe has to attach to the chipset) and network traffic (chipset provides the network controller), etc. It's bidirectional so you can read at 4GB/s while you write 4GB/s to the GPU, but there's also just general contention for commands/etc. So performance results on Thunderbolt will be worse than a card attached to the chipset, which will be worse than a card attached to CPU-direct PEG lanes, even if lane count is the same.
(the exception to this might be Ice Lake and newer Intel laptop chips, where the Thunderbolt controller is actually part of the CPU itself, the performance impact of that should be less. However, this does not apply to desktop chips, including Alder Lake.)
I was reading a techpower article on the 6500XT recently, because their power usage testing is excellent aand the 6500XT has an incredibly low idle draw versus other desktop cards.
PCIe 4.0 is one of those things that if you need it, then you already know you need it. But almost no one does.
Seems correct. ASRock already allowed 5000 series CPUs on their X370/B350/A320 motherboards with a beta BIOS released in 2020[1]. AMD told them to stop[2]. That beta BIOS was never officially released.
https://wccftech.com/asrock-amd-ryzen-5000-cpu-bios-support-...
https://wccftech.com/amd-warns-motherboard-makers-offering-r...
It’s not that AMD is good and Intel is bad, or vice versa. It’s that only Intel is bad and only AMD is bad.
What we need is a healthy back and forth competition between both the companies (and some ARM sprinkled in as well).
This is what good competition looks like. As a customer I love it!
For example, my B350 doesn't have a USB-C port, LAN is only 1Gbps, only one M.2 SSD slot.
How far we've come. I remember having a semi-decent on board 100mbps LAN was only reserved for highest end.
Same for M.2.
These cpus are much more than 10x faster than the CPUs you would have used then.
Networking speeds in desktop devices haven't kept pace. Not just with cpu speeds but with storage speeds-- which is particularly obnoxious when you want to use network attached storage.
I assume this is due to a mixture of internet and wireless creating an extremely low bandwidth least common denominator and that running >1GB over copper is kinda problematic (finicky, power hungry, etc) -- and the industry seems to reason (perhaps correctly) the the customers in this segment can't handle fiber.
I personally have 40GBE attached desktops (and 100gbe attached stuff within the server rack)-- thankfully quite economically due to surplus enterprise hardware, so I'm well aware that it can be done... but to do it in a small formfactor system is still a problem, e.g. for any of these non-apu systems your normal mini itx systems will use all their slot space for a graphics card and not have room for a mellanox adapter.
Actually, it's not only about the CPU performance, because NIC performance drew a cosine wave w.r.t. processor power over the years. Earlier NICs had much more machinery inside them and were expensive. Then Realtek came and we witnessed the era of (almost) software LAN adapters, then the silicon became cheap, and some of the stuff moved back into the LAN controller. So modern, higher end cards do not do significantly more offloading, but they handle other things like virtualization of cards between many VMs, or other higher level tasks.
From what I've seen, making a faster CPU is easy, but a faster and wider fabric is harder to weave due to distance, switching related noise and other physics related phenomena. Cache and good memory controllers are also relatively easy, but expensive. Also, RAM and higher voltage doesn't go well together, because RAM uses a lot of energy for its size, so it comes with heat and stability problems and other lifetime related problems (yes, you can fry RAMs).
Storage is another hard technology. Spinning rust is limited by physics, vibration, rotation speed and other mechanical stuff. SSDs came crashing, but before they were cheap and Sun didn't got swallowed by Oracle, they did nice ZFS arrays with some very expensive SSD caches and spinning drives combined. I've seen whattheheckisthat amount of speed from an array with a size of 6-8U total and a couple of IB interfaces fitted to it at the factory. Currently we get that amount of speed from Lustre arrays with some SSDs and a lot of spinning disks.
With the current backbone capacity of a standard desktop computer, even on the high end, 40Gbps network is overkill unless you're going to ingest that data directly at the CPU. Yes, absorbing the data at the disks are possible, for what cost and what use case? 10Gbps is understandable, but I still consider 1Gbps as a sweet spot for general purpose computing. If you're going to transfer vast amounts of data from a storage array to a local system, yes you can go higher, but it's still a niche. Also as you've said, going over 1Gbps is problematic from a signal integrity point of view, and fiber is too fragile and expensive for average customer.
These Mellanox cards get quite hot when they're constantly utilized, so in a little ITX box, both the card and the system will be cooked. The process becomes faster if you use active fiber cables, because connectors sink the heat inside the case via heat sinks attached to the connectors on the cards. Even if you use them in a system room, building a large network with them requires expensive equipment which has adequate bandwidth and cable length. Also, running an IB network needs other software machinery to keep the network up. You can directly run ethernet, or TCP over IB, which kinda defeats the purpose and adds additional penalty and reduces the performance.
All in all, higher speed network is still doesn't bring too much of a value to the home of the average user, and is not still that cheap for the enthusiast. Also the hardware is not scaled down for the home. Cards are big & hot, connectors/cables are bulky and the fabric gear is noisy, bulky and high maintenance for home.
Yes they work rather nicely, but I'd not want them at home. I have enough of them at the system room already.
GigE is pretty annoying in that it can't keep up with a single spinning disk (let alone 5), so large files (family videos, movies, system images, backups, etc) can take annoying long. Hell even wifi access points are becoming bottle necked by GigE. Fortunately 2.5Gbe looks like a reasonable compromise. Rarely a bottleneck and support is common on cable modems (motorola among others), access points, motherboards (pretty much anything newish from AMD/Intel), and even small SBCs if you run one as a firewall, router, or pi-hole.
There are some switches and routers around and are often fanless till the port counts get high.
Even quality 10G cards seem to run pretty hot, have huge heat sinks, and the sfp+ I've worked with tend to get pretty hot. Seems like there is a newer gen 10G chip, but they are rare and expensive but do run substantially cooler.
Boards like the Intel quad port 10G X710-T4L (the cool one) vs the X710-T4 (the hot board) run at about half the power (14 vs 29 watts). I'm shopping around for the dual port version (X710-TSL) which peaks at 9.6 watts and uses 8.2 watts typically. No idea how other (non-intel) 10G boards do on power use.
And indeed, the heat from higher speed cards is a nuisance... though 10G copper is awful in that respect.
> too fragile and expensive for average customer
I think that's a bit overstated. It's more fragile but not enormously so. Terminations are a nuisance, which favors pre-made patch cables. I'd pretty much always prefer to use 10g over fiber than copper.
But I'll freely admit that having spent more than a decade as a network engineer my expectations may be weird. I will say that after many years of using fiber connected equipment at home for the desktop/server systems, my spouse has never broken a cable or connector (nor have I, of course).
I've never-- professionally nor personally-- had good luck with copper 10g and try hard to avoid it. If I have to use well made/tested pre-made patch cables to make it work I might as well use fiber (or dac cables) and trust that literally anything I buy will just work.
> Also, running an IB network needs other software machinery to keep the network up.
I used to use IB because the IB only nics were super cheap on ebay-- but I don't anymore. It's easy enough to use 10/40/100GBE. The extra software was a nuisance compared to the value it provided.
> and is not still that cheap for the enthusiast.
I mean, you can get 40gbe capable cards on ebay for like $50ish (usually generation or two old mellanox dual ib/eth cards)-- main cost to them beyond their power consumption is knowing how to use the utility to put them in ethernet mode. Dac cables and LR2 optics at similar prices. Particularly if most of your usage can be covered by dac cables, I think it's pretty affordable-- not 1gbe affordable, which is close to free on on a cost per port basis, but when you're bottlenecked on network bandwidth there isn't really an alternative, except for keeping data locally on the systems.
I think you're correct in that regard. I'm an HPC administrator for ~15 years, and I manage quite a few systems with a sizeable storage and related network machinery up to a certain point. While I can understand your point, I've written my comment from the eyes of a bona fide standard consumer which doesn't do anything higher end on their homes.
> 40GB is overkill, but 10GB still bottlenecks for me. I'm surprised to see anyone consider 1GB a sweet spot-- it can't keep up with a single conventional disk, and it's causes obvious slowdowns for fairly conventional operations.
My sweet spot comment assumes streaming content consumption, occasional system to system file transfers, in home entertainment streaming and some gaming maybe. If I'd have to move my photography or video archive to a NAS instead of internal drives, a fatter network would be indeed required. To be bottlenecked by 10gbps, you really need to have a busy network with a lot of going on. None of my personal networks carry that amount of data any given time, and they don't need to do it to be honest. Moving 10gbps fairly constantly is not fairly conventional, TBH.
> I think that's a bit overstated. It's more fragile but not enormously so. Terminations are a nuisance, which favors pre-made patch cables. I'd pretty much always prefer to use 10g over fiber than copper.
Yes, modern fibers can endure relatively unbelievable amount of abuse, but it's still well inside the boundaries of a normal user's limits, and the threshold is still lower compared to fiber. Many people I know can damage a fiber without intending or knowing it.
> I used to use IB because the IB only nics were super cheap on ebay-- but I don't anymore. It's easy enough to use 10/40/100GBE. The extra software was a nuisance compared to the value it provided.
When you're doing HPC, the value provided by the features are indispensable. Latency, MPI offloading, DMA are all essential things you need for a really performing cluster. So we prefer IB rather than Ethernet. So horses for courses, obviously. I'd also rather not use IB for a home network.
> I mean, you can get 40gbe capable cards on ebay for like $50ish...
Consider the price as the money + setup + maintenance. For you, the last two items have no cost virtually. For me they'd be negligible. For a regular consumer, they're much bigger items to wrestle with.
So, as a people who knows how to play with bigger tools is really a privilege and luxury. However, I have enough servers at work, and I don't want to manage stuff at home. So if that lower and, dumb stuff works as I want and doesn't make my life worse, I'm a happy camper. As a result, I over-engineer my home stuff very minimally.
- Caveats - These are software dev & light multimedia editing computers. By the time I get rid of one, it's had as much RAM added as possible. I've done this twice now. Would not work on a gaming rig.
Before that, MoBo manufacturers even made extensive use of additional chips for networking, USB, mora SATA ports, RAID and what not, which could make the whole system more complicated/fragile (all the additional drivers!).
I nowadays try to avoid that and aim for boards close to "reference" (whatever the chipset offers out of the box).
I just never bought a very expensive video card & I can't comment on how long a good one would last. I imagine it depends on to many variables from the games you play, the settings of the games you're okay with, the card, etc..
Generally I'd buy midrange parts, from well known vendors, and pay a bit extra for SSD, ram, etc. Avoid anything with a small muffin fan (primarily chipsets these days), get a quality case, 140mm or bigger fans, quality power supply (a 5 year warranty is a good start). My current system is from 2015 and was a Xeon E3-1230 v5, 32GB ECC ram, GTX 1070, and a M.2 NVMe. The E3-1230 was cheaper with ECC then the similar i7 was without and just slightly slower (3.2 vs 3.4 GHz). Had no problems so far, and I'm just starting to look at a replacement. Looking closely at the expectations for Zen 4 CPUs and the current Alder Lake (or follow on ). I'm definitely sold on DDR5, especially since with a 2 dimm system (128 bits wide) you get 4 channels instead of 2 with DDR4. Should be that much better at not stalling your cores on cache misses.
If I bought today it would be tough decision between the AMD 5700x (8c/16T with a 65 watt TDP) vs the similar Intel Alder Lake. I do think AMD has the edge for systems that are low power and Intel the advantage for high power. Personally I'd rather have quiet, cool, and reliable instead of an extra bit of performance that nobody will care about in a few years. I'd consider a Apple M1 mini ... if it had the M1 pro or Max. Don't think the studio is quite worth it, but I am chipping in few $ a month to Marcan for a linux port.
unfortunately now we get into the discussion about 3.5 slot GPUs that overhang all but 1 or 2 of your pcie slots, and the general lack of PCIe connectivity on consumer boards... you can sorta work around it sometimes with vertical mounts/etc. But it's annoying and takes work/planning.
I miss the days of X99 and having 28 lanes at my disposal, and every GPU being 2-slot or at most 2.5 slot such that I could actually use those lanes.
It's also been really annoying watching the arms race between GPUs and motherboard slot spacing. For a long time, GPUs were 2 and 2.5 slot, so motherboards went to 3-slot spacing for their slots. But then 3-slot and 3.5-slot GPUs became common, which overhangs the middle slot (or puts the GPU right up against a card placed in the middle slot). What we really need now is for motherboards to go back to 2-slot spacing so you get a slot in the #4 spot to give 3- and 3.5-slot cards some breathing room...
Competition is fantastic, but the people who rise to the competition are, in my opinion, the real heroes.
https://wccftech.com/asrock-amd-ryzen-5000-cpu-bios-support-...
Then AMD told them to stop it. Which is why that BIOS never got an official release and never appeared on ASRock's website.
There are people that have been running 5000 CPUs on ASRock for over a year now.
Conversely, you can run first gen Ryzen on B550/X570. Which many people seem to bizarrely think is impossible. Possibly due to AMD's marketing.
Or am I just thinking about an entirely different thing...?
It really does not fit the narrative of what AMD did with Zen so far, that's for sure.
https://wccftech.com/amd-warns-motherboard-makers-offering-r...
The reality is that AMD fanboys haven't woken up to the fact that AMD is no different than Intel. All the same marketing and segmenting games exist on AMD. They were the underdog and then suddenly, they are on top and acting just like Intel.
https://www.youtube.com/watch?v=h74mZp0SvyE
Most of the stack has a similar TDP to Zen1/Zen+/Zen2. The 5950X has the highest TDP, 105W, which is the same as the 3950X, which is allowed. In practice the 5950X actually pulls a bit less power. Furthermore, the A320 boards (even lower-end) got official support, despite a lot of those boards having utter trash VRMs.
(furthermore, there's nothing that guarantees that every A520 board has a VRM that is capable of supporting a 5950X either! it's not like board generation is some guarantee of VRM quality, there are X370 boards that can handle a 5950X and A520 boards that can't. That's not how support is determined.)
16MB bios isn't a limitation either. B450 boards are the literal exact same silicon - B350 is to B450 as Z170 is to Z270, basically, it's rebranded silicon - and B450 boards with 16MB BIOS got support. There are B450 boards with 16MB bios that support literally the entire range of chips. And once again, so did A320 boards with 16MB BIOS.
Releasing official support for A320 really just blew a hole in every retconned justification that people tried to make for AMD. It's uncouth to say it here on HN, but there really is a large number of people who are super emotionally attached to the AMD brand and willing to stretch to absurd lengths to justify what are clearly business decisions.
And now they're just reversing their previous policy. Wasn't a problem at all, actually.
From a technical point of view "Chipset" no longer exists. Every single chip going into Socket AM4 is a SoC able to work 100% standalone. Any differentiation and market segmentation you see around "chipsets", which are by the way just a fancy northbridges hanging off standard PCIE, is fake profit maximization theater.
Now I can use the 5800X3D, but nothing in between those, or is there some other wrinkle I haven't understood?
Basically all Ryzen 5000 series chips can become compatible if your motherboard OEM releases an updated BIOS.
I currently have a 2600 paired with an X370 and have been meaning to try APUs and play around with passthroughs more with the RX580.
Not to say for certain I plan to get it, but it's great that the support is there the day AMD announces it!
Really though, there's some cheap mATX boards with high quality VRMs that can run a Ryzen 9 with no issues. Though tbf, if the processor starts but is not stable, an undervolt or maybe clock cap could work, at the cost of a slight performance drop.
It's nice to have the option to just upgrade the processor.
A great upgrade for the old Mellori-ITX running a 2400G. https://github.com/phkahler/mellori_ITX
BTW my system is limited to 4K30hz on the HDMI for reasons I never figured out. Haven't found a good Displayport->HDMI adapter to fix it, and not sure why the CPU/mobo combination won't do it on HDMI. I don't do gaming on it so not much of a problem, but I'd prefer 60hz.
Under the HDMI section indicates up to 5Gbps video bandwidth (in addition to 8 channel audio). So on the surface it sounds like I shouldn't even get 30fps, but I do. Thank you for the info, it looks like the board isn't up to the task via HDMI so I will make the effort to find a Displayport adapter. That PC has a LOT of life left in it and I'd like to get the most out of it ;-)
[1] https://www.cpu-world.com/CPUs/K7/AMD-Athlon%20650%20-%20A06...
Sort of the same way in the software world where it's easier to add a new method then it is to enhance an existing method. New methods have no legacy baggage to worry about. Tweak an old one wrong and you run the risk of breaking a bunch of people.
AMD's commitment to forward compatibility is pretty nice, but definitely more work on their part.
Then again, AMD's legacy is making sockets/slots work for their processors :D (That was the whole point of the original athlon)
In some cases your existing cpu might not be supported using the new firmware. So if you update the bios, you might have to swap the processor before you can boot the computer again.
anyway, there's nothing stopping vendors from releasing support for A300, Asrock simply has abandoned the Deskmini A300 and doesn't want to support it anymore. That one isn't on AMD, Asrock just seems to have abandoned it.
This would have been my main reason to build an AMD system, although I was easy to win, I always built AMD systems IIRC... And I ended up postponing it again and again, since I don't game and the reasonable (budget) options were always lagging behind regarding technology/efficiency - most of all, my Phenom (actually a rebranded Athlon with 2/6 cores disabled) was still doing OK. Now there will be AM5... Didn't follow the news, is it supposed to live as long as AM4?
Would it even pay off for AMD to repeat that effort?
I thought the chiplet packaging was supposed to result in lower costs due to better yields on smaller dies? I guess the extra cache takes up more than enough die space to make up for that?
This also implies other limitations like PCIe 3.0 and half the cache of the equivalent enthusiast lineup, because that's how the laptop chips were designed. The APU lineup has always performed a bit worse than the equivalent desktop lineup as a result.
It looks like the extra cache die isn’t good for thermals, so maybe it’s not viable on the 12 and 16-core chips without sacrificing too much clock speed.
Looks to be a strictly gaming-focused CPU play. Not a bad move given the way the enthusiast market works, but it doesn’t do much for a lot of non-gaming workloads.
How so? 8 cores still covers plenty of light-workstation workloads, where the extra cache should lead to an overall performance improvement (despite the minor drop in top clock speeds). With current software architectures, it's still hard to make full use of 12 to 16 cores.
The 5800X is a good processor and it's true that people do productivity on it, nobody disputed that and there's no need to rush to its defense. But you've missed the point about the overall way the upgrade is (or rather, isn't) being rolled out.
Thermals have always been the giant asterisk on die stacking. Everyone knows it's going to be a problem. But this is the first consumer stacked-die product (that I am aware of?) so we don't really have any sense for overclocking on those processors. The tentative implication here is - bad. Even with a single cache die, it's still got a whole processor underneath it heating it up. There are implications here for GPUs as well, as both AMD and NVIDIA are expected to release stacked-die products next generation. It's looking like they will have to keep clocks under control to make that happen - maybe that is a counter-argument to the "500-600W" rumors (for both brands).
But multi-die products have twice the area to dissipate their heat over - just like 5950X doesn't run at higher temperatures than 5800X. Higher power, yes, but twice the area to dissipate it over means thermals are about the same. That's not really the reason.
A single-die limitation also wouldn't rule out a 3D version of the 5600X - it's no longer "the bottom of the stack" and finally there are value SKUs underneath it, it would be appropriate to re-release it as a more performance-oriented 5600X3D SKU.
Anyway, my personal opinion is this is going to be a very limited product that primarily exists for "homologation". AMD can say that they've released it, it's officially a consumer product, so it can re-establish AMD's place on top of the gaming benchmark charts, but it's going to be a 9900KS-style super-limited SKU that doesn't see any real volume production at least until AMD has sold their fill of Milan-X. It just exists to put AMD back on the top of the benchmark charts now that Intel has retaken it with Alder Lake.
The 5800X is the best processor for them to do that. 5950X sets up some small regressions for inter-CCD problems/etc. 5800X is enough for games for now anyway.
It's going to be 2 years later, and at this point I'd rather jump directly from DDR3 to DDR5.
However, that computer is getting long in the tooth. The RAM is slow compared to modern stuff. The motherboard is cheap, has poor UEFI, and I zapped the ethernet port with static - destroying it. Do I really want to invest more into it when I could wait 6 months and build an all new PC with DDR5 and Zen 4? It would be a lot more expensive but I have a lot more money now than when I built this computer.
That being said, if you can afford it get the shiny new stuff. You deserve it.
And can't run most of the games I've played over the past two years.
That parent comment was particularly out of place since the 5800XD is a gaming-focused CPU.
Not really the same market at all. The 5700X (2nd most expensive on a list of 7 CPUs) costs $300, 32GB ram of DDR4-3600 is $150, 512GB SSD is $60, motherboard, case, and power supply is likely another $300. So $810 for a system, not including a GPU, but $1200 is more than enough for something decent.
It was a nice upgrade up from Ryzen 3 1200 that i had previously and honestly, i don't see myself upgrading again for the years to come, unless things become visibly slow and sluggish because of Wirth's law (let's assume that in 2030 most of my desktop apps would run Electron instances).
It does gaming, it does programming, it renders videos, it does 3D modelling and other content creation, it compiles code and runs Docker containers and oftentimes multiple of those at the same time with no overclocking and no issues (e.g. render a video in kdenlive while playing a game in the background).
Somehow it feels like either hardware has scaled to the point where new improvements are pretty incremental, or maybe software's ability to be wasteful with the provided resources thankfully still hasn't caught up with these improvements - you no longer really need to get a new CPU every 2 or 3 years, even the Ryzen 3 1200 which also came out in 2017 was adequate (i just got a good deal on the upgrade), which is really nice to see!
Even my homelab servers still run used Athlon 200GE CPUs from 2018 because of the 35W TDP and are on par with cloud resources that i can otherwise afford for my 24/7 available cloud stuff, just much cheaper when you consider how long they last.
Also, there's something really cool about every device in my home (apart from laptops) running on the same CPU socket and all of the parts being interchangeable with no driver update weirdness. Though it'd be even better if the Ryzen CPUs were the variety with iGPUs, given the GPU prices still otherwise being pretty unfortunate (in case my RX 570 would break).
The only reasons that i see for upgrading my setup in the next 5 years would be one of the following:
Here's hoping that won't happen and that i can enjoy CPUs that are suitable for my needs (and help avoid the e-waste issue at least somewhat), while those more well off financially than me can dabble with the more recent releases to their heart's content!However what shocked me was that my base M1 pro was a few minutes faster. (Same amount of ram)
Still fast enough that it’s not worth an upgrade, but we’ll see once she starts transcriptomix
Apple M1 = 128 bit x LPDDR4X-4266 = 68GB/sec and I believe 8 memory channels (8 pending cache misses). A modest core count (4 fast + 4 slow) helps keep the memory bandwidth from being a bottleneck.
The M1 pro doubles this to 256 bits, 16 channels, and 200GB/sec, which is a significant help for the integrated GPU and hits levels that AMD APUs can not match.
M1 max doubles again to 512 bits, 32 channels and 400GB/sec.
M1 ultra doubles again to 1024 bits, 64 channels, and 800GB/sec.
Not sure AMD can really compete and the APUs are severely limited by memory bandwidth, unless you buy a PS5 of XboxX. I'm hoping that AMD takes a page from the Apple playbook and ships a consumer CPU with dramatically more bandwidth and allows users to skip the current curse of discrete GPUs that run hot, are not available at MSRP, and are hard to get even at 2x MSRP.
Not like they aren't shipping by the millions in the XboxX and PS5, there's obviously a demand for them and AMD is obviously capable of making them (they make the CPU in both).
Only went for AMD so I could reuse the motherboard. But next time I'll go back to Intel.