> in a HandBrake-based video transcoding task, Ryzen scored a narrow win [against a stock i7-6900k]
Man, if that holds true in more benchmarks, that's fantastic news. AMD chugged along by "competing" further and further down Intel's product line more or less since Sandy Bridge, if they can genuinely compete with the i7 now I can't wait to see the next few years. Intel has always seemed content to release incremental improvements until they get kicked in the pants by somebody else, I wouldn't be surprised if they have a bunch of stuff waiting in the wings to maintain their edge over AMD.
I think AMDs downfall started way before SB. AMD was competitive against Core and Core 2, but already Nehalem was rather... troubling. SB and the following iterations where just the death stab in many markets for AMD; after SB AMD didn't even try to compete with Intel in the server markets (their platform is six years old. Their newest server processors were released in 2012).
AMD fell the moment they failed to improve over K10.
Purchasing ATI didn't seem to help them much, either. Perhaps it was to offset the losses that they saw coming, but it also seemed to distract them a bit.
I never understood what were the failure and what could AMD have done better. When intel started to be ahead in single threaded IPC, AMD tried to innovate by bringing low consumption, multi threaded, with embedded GPUs cheap SoC. It seemed quite in touch with the early netbooks where people would need a blend of capabilities (hardware video accel, compositing / CSS3 able GPU) instead of a compilation workhorse.
I don't know if the designs were really absurd, or if the market couldn't understand AMD's value prop. Or if the programmers couldn't (or wouldn't) tap into their APU idea.
APUs didn't take off except in low-end-ish notebooks and low-end desktop computers not because the idea is bad, or the graphics performance was poor (it wasn't), but because the CPU part was subpar.
The gaming consoles show how well the concept works, although, sadly for AMD, these deals may be high volume but don't have high margins.
AMD dedicated graphics have only minor issues imho, they're very competitive with nVidias offerings. But in the professional and compute segments they fail to take off, because nVidia established the popular, proprietary CUDA standard there. I don't think there's a (major) technical reason why AMD has much less market share there.
> The gaming consoles show how well the concept works, although, sadly for AMD, these deals may be high volume but don't have high margins.
They're likely fairly consistent income, however, even if the margins are relatively slim. For a company in AMD's state, constant income no matter how small is likely a good thing.
Always good to have economic mass, market presence. I have no data, but I could see a guy owning an AMD based game console thinking about other AMD products for its PC.
The problem with that is unless that guy figures out what's inside his console, he would never realize that an established PC vendor was providing key parts inside. I've never had a game console, but I don't remember seeing AMD stickers on consoles or their packaging.
One problem with APUs is that they make it so difficult to rank offerings by performance. A convenient benchmark blend is a reasonably useful approximation of actual utility for conventional general purpose CPUs. But when GPU becomes part of the mix, you get a lot more uncertainty if you do not investigate the details. For an entry level system, where APUs are at home, few people are willing to invest that much effort into their buying decision.
AMD marketing tried to put GPU performance in the spotlight. The unfortunate effect was that a moderately informed consumer then had the choice between a somewhat predictable level of graphics performance paired with whatever CPU that came with it if buying AMD, versus a somewhat predictable level of CPU performance paired with whatever GPU that came with it if buying Intel. In most market segments, uncertainty about graphics performance has much less impact on buying decisions than uncertainty about general purpose compute. The (involuntarily, given their portfolio) focus of AMD's marketing on the graphics part of their APUs made it close to impossible to successfully communicate value-for-money advantages in their price range (assuming they had it).
No, it was early designs (intentionally?) fucked up that deviated strongly from AMD's reference design.
Rumors went around that Intel paid manufacturers to do it intentionally, but nothing was ever proven. However, given how Intel has had similar behavior in the past, it isn't an unreasonable rumor either...
No you didn't I thought you were referencing the old blender demo. They apperantly did another one.
Would be interesting to see the boost stats in that case, 3.7 boost is a bit low for a 6900k part tho everyone I know that had it rubs it around 4.2-4.3ghz.
With the stock cooler, you can expect the CPU to start at 3.7GHz and come down to ~3.3GHz or so near linearly over about a minute, which is about how long the demo lasted.
If AMD left a benchmark to run as a space heater right up to when they started the timed one, the CPU would start at 3.3GHz.
There is no stock cooler for the 6900k nor any other enthusiast part intel makes and this has been the case for a while now there is also no official oem reference design for the cooler either.
As for the boost, I haven't seen a load that would prevent a core i7 from boosting permanently on air cooling, with water you'll have no problems keeping it at 4.3-4.5 even with an AIO...
> Tre 6900k was not only not boosting they downclocked it to a locked 3.0ghz....
This is incorrect. They did a "both CPUs locked to 3.0GHz" test a few months back, today they did a "Intel CPU running absolutely stock, AMD one locked to without boost" test.
Yeah Ryzen looks like a big win, however it's been the entry level i7's that have been the sweet spot for gaming, offering fewer cores that can be clocked from 4.5 to 5+GHz quite easily. Coupled with the current paradigm where most games don't scale beyond 4 cores much at all, gamers haven't really wanted the 6900k anyway. Long term Ryzen will continue to become more useful as GameDevs learn how to make better use of cores, but I don't see the need to replace my 5 year old 2600K just yet.
Usually when top products have close performance price war starts so it's possible that "10% more expensive than the Intel" will be much lower than that Intel price today.
But Intel also has an ability to put pressure on OEMs (because of monopoly position and massive supply volume, so if they will not sell CPUs to some vendor as a punishment for buying AMD, AMD will not be able to supply enough CPUs on their own) and brand loyalty, so probably they will keep their "high-end" price points.
Oop yes it says so on the image on the page. Depending on price point this could be very interesting. I'm particularly looking forward to seeing what they can offer in high-core-count models.
Shame they still seem to have a 15-year-old in charge of marketing though. Should have just stuck with "Zen".
it seems like their naming scheme is to take a three-syllable name, and cut off the first syllable. Triathalon -> Athlon, Centurion -> Turion, Horizon -> Rizen.
Also, it sounds like "Ryzen" is their consumer product line. Their server product line might be called something else. It makes sense to have multiple product lines with different names and to refer to the common microarchitecture as "Zen".
The i7-6900K is a Broadwell-E part (i.e., not their "mainstream" high-end parts)—while it implies an eight-core part against an eight-core part is competitive, it doesn't say anything about four-core against four-core (which as far as I know at Intel are different designs from different masks, and that's ignoring the fact that now the competition there is Skylake and a different microarchitecture too).
They were real-world though. The Bulldozer architecture is optimal for some applications, just not very many that mattered at the time (and even now). I think the architecture was just way ahead of its time. It anticipated the software practices of the future and tried to build a CPU for it within the confines of today. Such futurism is impressive, although kinda foolish, for such a capital-intensive product.
The final result is that the Bulldozer and derivated cores, has aged very well. Any FX core works better now that they was released, because actually a lot of programs have in mind multicore that when the original FXs was released.
There will most certainly be a server variant. Currently on schedule for Q2 (this article is about the desktop variant, which is scheduled for Q1. Maybe at CES?)
For both companies, you typically need a different socket/chipset/mainboard for the server chips.
AMD typically includes ECC support in "consumer" chips. Not all boards, do. I'm worried too, though, that this still hasn't been confirmed yet. An 8-core at 3.4+ GHz sounds perfect for a workstation.
ASUS is one of the few that lists it on their spec sheets. But I believe Gigabyte also supports it on a number of current boards, along with some other manufacturers.
The AMD server line has usually been named Opteron.
And by the way, Xeon CPUs aren't usually faster than regular Intel chips. They may have more cache or more cores, which might speed up some programs. But otherwise, they're the same. Or even slower since their motherboards don't overclock well, overclocking not being a big selling point for servers.
One of my machines has a low-end Xeon because at the time it was basically 1:1 performance comparable to an i7, but cost ~$40 less because it had no integrated graphics.
The other day I thought to look into refurbished server and workstation boxes. Lots of last-gen Xeon machines to be had for 25% of their original retail, often likewise outfitted with a workstation-spec NV Quadro GPU - maybe a good choice if you've been working on laptops for years like me and could make use of more cores.
Biggest difference with i7 vs xeon, xeon can handle sometimes upwares of 1tb of ram now. i7 maxes out at 64gb. You be the judge. i7 were usually 32gb, especially the LGA1100 series. The LGA2011 were @ i7 handled 64gb. LGA 2011 Xeon's are all 64GB+. Also support for ECC ram. So basically if you are not pressed for raw CPU, and you need just a lot of cpu for simulations, VM's etc, xeon is your go to CPU. Xeon LGA2011 series has some crazy CPU Socket configurations, you can get a lot depending on the motherboard vendor. I think some as many as 4+ CPU Sockets. i7 cannot run in parallel with another socket I believe. Just like with the LGA2011 1600 vs 2600 series xeons.
The previous generation (5820k, etc) were only officially rated for 64GB, but they too worked fine with 128GB in practice. Intel just never updated the spec sheets once 16GB DIMMs became available.
> They may have more cache or more cores, which might speed up some programs. But otherwise, they're the same.
It's like saying that mercedes with 2.0 engine is the same as 65 AMG. So writing that they are the same is like writing that every car is the same because they all drive.
You do not overclock in 99.99% cases Xeon server cpus.
You do not buy 12 core xeon to use one core, don't you? of course Xeon is faster for the use case you are buying it. A LOT faster in many cases than what you called "regular" chips. You are writing about home use cases from what I see and Xeon series is mainly for server use.
The E3 xeons use the same silicon as the desktop ships, but they do enable ECC. Amusingly the cheaper E3s are often cheaper than desktop equivalents, but have hyper threading enabled. A few times I've managed an E3 based system (4-5 generations out so far) CPU cheaper than the similar I7. Often something like $270 vs $340.
Well, since you ask, and speaking as someone who purposely bought AMD with ECC RAM in the past - because she'd want it that way.
It's a bit cheaper, and the downside is an occasional bad cat pic, or maybe an extra browser crash per month to add to the 399 one already gets from poor code ...
> Also with Intel you can upgrade to more performant Xeons
Xeons are not necessarily more performant, in fact, often they are less performant than their desktop brethren...
They're server CPU's, designed for long uptime, typically underclocked compared to the same silicon in an i7/i5 package (they focus on stability, not sheer performance), support ECC memory (again, not performance related, rather stability), support more memory, etc.
Is there a MSRP for the RYZEN flagship that they showed off today yet, because I havent seen one.
But as long as it stays under the $1100 of the 6900K I think they can secure the mid to high-end enthusiast grade market which is good because I am tired of intel controlling that portion of the market
If they're still making an adequate profit at 500$ they're gonna grab a very healthy chunk of the market. It's more than half intel's competing product. Even price cuts ..
It will depend on how well they'll be able to push the CPU in 4 core mode.
If they can't reach at least 4ghz it wouldn't matter.
Games are not optimized well beyound 3-4 cores these days I can disable 2-3 cores on my 6 core CPU and maybe lose 5% fps in CPU intensive games.
Intel can push their clocks pretty darn high especially with OC which now comes out of the box with every enthusiast grade motherboard.
If you end up having a 3.4ghz part going against a 4.5ghz part the 4 extra cores might not matter.
Even in professional use cases adobe doesn't scale above 4 threads like at all and many video encoders also don't scale some even don't scale beyond 2.
The only thing that really scales well universally is 3D rendering.
Good points overall, but I wonder what about new usages like neural networks ? Seems like new tricks might make multi threading a bit more trendy compared to previous years.
> Even in professional use cases adobe doesn't scale above 4 threads like at all and many video encoders also don't scale some even don't scale beyond 2.
I've mostly used open source encoders, but they all scaled well with more cores, and the Adobe suite has come a long way. It wouldn't make sense for a heavy user to get anything less than 6-8 cores using Adobe.
Probably the real price that such a processor should have. But we've gone so long without real competition in the PC chip market, we can't even tell anymore when Intel is ripping us off.
Heck, even 99% of the "tech media" didn't notice (or didn't care to make a big deal out of it) when Intel swapped Core-based Celerons and Pentiums for significantly weaker, but much more profitable to build, Atoms. The consumers never stood a chance against Intel's progressive rip-off.
Let's see how the market settles in 2-3 years, after the Zen architecture is fleshed out across device lines, and after Qualcomm commences its entrance in the notebook market with 50%-100% of "Core i7" performance (depending on which Core i7 [1]).
Per-thread performance tends to matter much more than number of cores for the enthusiast market, which has centered around the 6700K (which I've got in my own machine right now). The 6700K retails for $330 and is about 14% faster per-thread than the 6900K, which I'm betting is a tougher proposition for AMD to manage.
This chip definitely seems rad, don't get me wrong--I just think the enthusiast market is sensitive to what this isn't, at present, doing.
I'd say the popularity of the 6700K has a lot to do with its price point. Paired with a GTX 1070 or RX 480, for around $1000 you can build a very good gaming PC.
If you want 6 cores you can get a 6800k for $150 more, but it also has a higher TDP and requires a more expensive motherboard (X99 vs Z170 chipset). This is going to put the total price up by at least $300.
As you say most games still favour single-core performance so paying extra isn't worth it. DX12 and Vulkan will start to change that though, so let's see...
DX12/Vulkan isn't going to be a serious thing for a few years, though. At that point, I think that Optane is going to be a serious feature and upgrading to post-Skylake chipsets is gonna be on the table for those enthusiasts.
You're thinking the old AMD architecture, which was quite different from that of Intel, and it sucked. The new Zen architecture is much more similar to that of Intel. It's possible third party reviews will show its performance to be a little less than what AMD is showing now, but it won't be anywhere near comparable to the old AMD CPUs (it will be much better and much more competitive to Intel CPUs).
> You're thinking the old AMD architecture, which was quite different from that of Intel, and it sucked. The new architecture is much better
I've been hearing that for the last eight years and it has never been proven to be true, in real world benchmarks and $$$/performance by the time equivalently priced tiers of competing varying $125, $195, $250 and $350 CPUs are available for purchase.
If you have been hearing that for the last eight years, you have been listening to some very questionable sources. CPU architectures take very long to replace, and all the AMD CPUs on the market currently are based on the BD core or it's derivatives. BD was released in 2011, and was very different from Intel CPUs.
We don't know yet whether Zen is as good as Intel CPUs, but the statement "You're thinking the old AMD architecture, which was quite different from that of Intel, and it sucked. The new Zen architecture is much more similar to that of Intel." Is absolutely correct, and is something that was not correct about any AMD CPU released in the past 8 years.
I used to only buy AMD.
Im not exactly sure where they are up to now (probably about 8 years if not more since I touched them)
but i definately remember the bulldozer hype and subsequent flop. lies about the number of cores and other things that just made me say "my computing is to important to my life to play that game".
I was also privy over the years to some test builds of machines including head to head amd vs intel. it is truly painful to say that not only were they often more expensive (new) for the performance, they were also horribly unreliable, pausing, and doing all sorts of stuff that made us sad.
hope thats changed, but saying there is any similarity between the current in use intel vs amd range is very far from the mark.
"Given AMD's recent poor track record in terms of $$$/performance for desktop CPUs..."
Are you kidding or did you mistype? Current AMD processors don't generally compare favorably to Intel with respect to some metrics, like absolute performance per core or performance per Watt. However if there is any metric where they match or surpass Intel, it's actually perf/$.
The reason is simple: perf/$ is the only metric where AMD has some control: they can decide to lower and lower their price down to wherever they need to be. If AMD wasn't competitive on at least perf/$ they would be in a much worse situation.
Here is one example I witnessed first hand recently when doing some video work: a good old $100 3.2 GHz 8-core Phenom matches or surpasses most ~$200 Intel processors when encoding MPEG2 files to H.264 with the Linux x264 encoder.
Even there, they're not currently competitive if you look at the lower end skylake (i3 and pentium branded CPUs) processors versus the same price range $100 to $150 AMD CPUs. I would dearly wish that were not the case, because more competition is better for the buyer...
I think the CPU on my 2015 home server cost even less (€40-€50). I went with an AMD Athlon because Celerons (which are about the same price) didn't support AES-NI, which I need for reading from my encrypted disks.
> if you look at the lower end skylake (i3 and pentium branded CPUs) processors versus the same price range $100 to $150 AMD CPUs.
This is bullshit through and through.
Between $100 and $150, you get any of AMD's FX 6xxx and 8xxx line. We're talking about 6 and 8-core processors whose performance are, in the very least, comparable with some of Intel's i5 and i7 offerings.
And Intel's comparable offerings cost over twice what AMD's processors sel for.
It's not bullshit. Those FX-processors are 4 years old by now and indeed do not compare favorably even against the smallest current Core-Processor, the i3-6100, in almost any metric. That is not surprising, as Skylake is a lot newer, and has a lot faster cores, which matters most for typical workloads.
The FX-8320E when overclocked can compete in multi-threaded computational workloads, but will use a lot more energy. For games (which is the main area where users stress their processor), the i3-6100 will still be faster. See http://www.anandtech.com/bench/product/1709?vs=1682. The bigger FX-8s are too expensive and would be pitted against i5s, where they always lose.
The only recent exception I ran across is Watch Dogs 2, which is sufficiently optimized for multiple cores and AMD processors to make the FX-8370 for once competitive against Intels i5. But that is the first game since the release of those processor where that is happening.
What does work for AMD is the Athlon X4 8.0K, the area under 100€. Intels Alternative is the Pentium G4.00, which is worse for most things.
Games are a very specific workload, and I find the focus placed on gaming performance in CPU reviews to be ridiculous. For another specific workload: software that can use 100% of 8+ threads, the FX CPUs are probably ahead of Intel CPUs in the same price range. They do run hotter, but that's not necessarily a huge drawback in a cold country.
No, they are not faster anymore. Really check the anandtech-benchmarks, http://www.anandtech.com/bench/product/1709?vs=1646. Also in areas that are not gaming performance, but resemble more pure computations – like video encoding – the i5 easily beats the fastest FX-8-processor. Same is true when comparing against the FX-9590. If I see that right in every single benchmark.
The results on that page seem to indicate that the i5 is faster (sometimes up to twice as fast) than the FX on single-threaded workloads, while the FX is faster (sometimes up to twice as fast) than the i5 on multi-threaded workloads. They're evenly matched on other tests (e.g. Handbrake & "Agisoft PhotoScan") - I would bet these aren't getting 100% core usage out of the FX. I'm surprised by how close they are in games.
If those results are accurate, they do seem to support that AMD FX CPUs are still competitive with new Intel CPUs in the same price range.
It's anandtech, it will be accurate. I'll now go one by one through each of them and count where the FX is faster: 17 (out of 86). Okay, that is even more than I thought. However, the differences where the FX is faster are most often very small, apart from the benchmark Linux-Bench 7-Zip, while the other way around the differences are often very big.
The FX has the potential to be a fast processor in specific workloads. Yes, that fits to what you said. Certainly not when single-threaded, but in multi-threaded workloads it can still be fast. But faster than the i5 it is only very rarely (which, yes, is less harsh on the FX than I was in the comment above). It is a pity that its potential (4 years ago it compared to a weaker i5) could not be used in meaningful ways in common software.
By now it is clearly not a good choice anymore: It uses more energy, depending on what you compare it with it is more expensive or costing the same, and it is on an old platform (AM3+) that won't see further updates.
> They're evenly matched on other tests (e.g. Handbrake & "Agisoft PhotoScan
No. We don't seem to have the same benchmarks. I'm looking at http://www.anandtech.com/bench/product/1709?vs=1646, where the i5 is clearly faster in the Agisoft and Handbrake benchmarks. It is mainly in the artificial multi-threaded benchmarks as well as Linux 7zip and Linux OpenSSL that the FX can win. For games the FX is almost always behind, but often close, however, those are partly limited by the gpu and thus not too meaningful.
> If those results are accurate, they do seem to support that AMD FX CPUs are still competitive with new Intel CPUs in the same price range.
No. They mean get an i5: Way better in single threaded workloads which is what counts for 90% of software, faster in realistic multi-threaded workloads. Get an FX if all you do is run OpenSSL and 7zip under Linux, or if already having a good AM3+ board (for that scenario it still is good enough).
There are 87 benchmarks there. I do not believe they re-do all 87 each time someone releases a processor. It's more likely that they are reusing old 8370 results and comparing with new 6600 results. I question whether all variables (OS, drivers, minor version numbers of software, etc.) are the same. (I'm not saying that re-doing the 8370 tests would favor AMD, just that it would give more accurate results). They don't mention what settings they used for the software, and what exactly each program is stressing. Splitting the PhotoScan results into five seems very arbitrary. Two decimal places in each result with no error bars, no indication of how they performed the measurement, no indication of the number of times they performed the test leads me to think that it's not a very well conducted set of tests.
> > They're evenly matched on other tests (e.g. Handbrake & "Agisoft PhotoScan
> No. We don't seem to have the same benchmarks.
We clearly have the same benchmarks. "evenly matched" was a poor choice of words, but I don't see a ~23% advantage for the 6600 to be very significant, when other results have the 8370/6600 100% faster than the 6600/8370.
> They mean get an i5: Way better in single threaded workloads which is what counts for 90% of software, faster in realistic multi-threaded workloads.
For browsing, office and most games the 8370 and 6600 are close enough that the average user won't notice a difference. Most users don't spend any time waiting for software that eats 100% of <= 4 cores, which is where the i5's advantage lies. What you find "realistic", I may not, and vice-versa. It depends on the user, and what he's using his computer for.
The i5 may well be the better choice for most users. It certainly has a number of advantages: less power-hungry, newer chipset/features, more upgrade options. My point is simply that the FX cpus are still price and performance competitive with new i5s. Which is better depends on the needs of the user.
> Most users don't spend any time waiting for software that eats 100% of <= 4 cores, which is where the i5's advantage lies.
Well, every gamer does, and this exactly describes what web-browsing does. However, the FX is of course not too slow for that. Just slower.
> It depends on the user, and what he's using his computer for.
Sure. I was joking above with "get The FX if all you do is 7-Zip on Linux", but yours is of course a correct statement. If you identified correctly that the FX will perform better for your workload, then it is a better choice (if okay with higher energy usage and not having further upgrades, which just might be the case, and it is not like intel is known for long socket support anyway). I for example use Funtoo, which is a Gentoo variant, meaning I compile stuff all the time. I'm on an older processor, but I assume the FX would be a good choice for me.
> It's more likely that they are reusing old 8370 results and comparing with new 6600 results.
I think exactly for the FX-8370 that might be not true, as it is the processor you'd have wanted to compare the new skylake-i5 against. But for the others and older models exactly that will happen. I see no big issue there.
> I don't see a ~23% advantage for the 6600 to be very significant, when other results have the 8370/6600 100% faster than the 6600/8370.
~23% advantage is very significant to me. That is way off of measurement errors. 100% difference just matters even more ;) Also, that it is most often pro-i5.
Ah, I just hope Zen will turn out as great as I hope and is announced and I can start recommending and defending those processors :)
That case seems* to have been dismissed. My (very brief) reading of it is that there is no industry-standard definition of what a "core" is (other than possibly being able to simultaneously execute), and what can or cannot be shared among cores. I see no issue with AMD calling a module of two integer units and one floating point unit "two cores"
I've always wondered how much that brought in, and how much they depended on it. AMD (ATI) have been in every Nintendo console since Gamecube until now (but not Switch), and they also were in the Xbox 360.
> AMD has been teetering on the edge of bankruptcy for quite awhile now. It's unclear how much rope they have left.
They have actually dug themselves out of that hole over the past two years or so. Their engineers are good, but their CFO is a wizard. Somehow he managed to take a company with bad short-term prospects and very questionable long-term ones that was up to it's eyeballs in very expensive short term debt and convert that into cheap long-term debt. Then once things started going better for them and the stock started going up, they have been turning their debt into equity pretty much as soon as possible.
Buying CPUs on eBay is generally a bad idea. When people consider price/performance they don't generally include second hand cpu markets in the equation. Also, xeon MBs are more expensive.
A xeon e5 2665 is rated at 115W thermal, so two of them at full load (let's say, calculating prime numbers) would be 230W... plus the overhead for fans, motherboard, hard drive other heat producing components in the system. I would guess that the lenovo workstation in the ebay link is probably 310W electrical load as measured by a kill-a-watt if you run simultaneous full cpu, ram and disk benchmarks on it.
Depends a lot on your electrical cost. Some people live in places where the power costs $0.05 per kWh. Some people live in places where it cost $0.36 per kWh (Hawaii).
There are some awesome deals on ebay right now, but I'm getting to the time when it's better to separate cpu performance and things with a lot of RAM, and have a fairly small, low powered quiet workstation...
If somebody wants a machine with a lot of xeon cores and a ton of ram (but noisy), you can buy a dell poweredge r910 server with 4 sockets x 8-core xeons and 128GB of RAM for like $1100. Makes an excellent xen or kvm hypervisor dom0/host development machine. It's just noisy, so in a home environment, put it in your garage or something on the far end of a cat5e cable.
Apples vs oranges. You quote prices of used/refurbished Intel processors. The $100 I quoted was for a brand new AMD FX-8300 processor (which sells for much less used.)
For me personally, I've benefited from another aspect: AMD doesn't constantly change their sockets (edit: or at maintains some level of compatibility).
My wife's laptop died, and she just wanted me to do put together something cheap for her office (she has a tablet and work laptop). I bought myself a new motherboard (keeping my CPU), gave her my old one and put in an older CPU I had.
A year and half later, I upgraded my CPU to an FX-8350. My old CPU (Phenom x4) moved to the basement NAS/Plex server, a nice upgrade to Athlon II CPU (without requiring a new motherboard).
Had I been on Intel, I would simply not done those upgrades because they wouldn't have been worthwhile. Basically every CPU upgrade requires a new motherboard. And in my experience, nearly every time I've had to get a new motherboard, my memory or video card or some other component is now outdated, and suddenly I'm basically building a completely new PC.
I'm in the same boat. Every computer I build is on a shoestring budget, re-using old components, etc. AMD has been a lifesaver. They provide acceptable performance at a price I can afford.
This is exactly what I have done also. It enables me to upgrade a lot of computers with every upgrade. I also end up with a lot of older spare parts, enabling me to build computers for elderly and people in need.
Personally, I hate to have either lying around that would work, but are unable to be paired up. The only instances that I've replaced motherboards but not CPUs are when the board has a problem (leaking capacitors, unbootable BIOS). CPUs generally aren't worth the cost of incremental upgrades anymore, and motherboards are cheaper than the CPU on them.
pcmark and passmark measure a vast range of cpu performance use cases.
currently, amd does terribly in pretty much every measure. absolute performance, performance per $, throughput. absolutely anything that matters.
There are some promoted intel chips that are worse.
but the only thing amd have been doing well at in recent years is hyperbole with a healthy dose of shilling. So i cant disagree with anyone on here that wont believe anything they say until they see it.
Does AMD now have an equivalent to Intel's open-source tboot for DRTM?
In the past, AMD avoided artificial "enterprise" segmentation with security features like IOMMU, but did not always have the necessary BIOS, board (e.g. TPM) and open-source tooling that Intel solutions provided.
During the press event it was suggested the boost mode would operate primarily off of temperature, taking into account high-end cooling setups. Seemed interesting, if not innovative. As far as I'm aware, existing boost modes tend to be static ranges.
I'd really like to see Ryzen meet or exceed Intel i7-6700K/7700K single-core performance using boost mode clocks with a reasonable cooling setup. That would eliminate any reason whatsoever to choose Intel for gaming until their 10nm product arrives Q3 2017.
Yeah, I wouldn't set your hopes that high, if that were the case Intel has 4 to 5 generations of silicon already prepped that they could start churning out in volume within 3 months.
Comparatively, AMD has been building chips to order as of late, since they do not have the money to build a bunch of chips that may or may not sell, so your looking at a 3 to 4 month delay before you can get a new chip like Ryzen after announcement.
The cache arrangement (see pic halfway down) and better branch prediction are interesting though, cache has always been a space where Intel has been able to use way less transistors to store the same amount of data.
>... if that were the case Intel has 4 to 5 generations of silicon already prepped that they could start churning out in volume within 3 months.
Can you elaborate? I'm fairly certain Intel deviating from their existing roadmap to such a degree and on such short notice would be extremely difficult at best.
After the Pentium 4 debacle, Intel Israel came up with the Core micro-architecture to replace it. Since then, Intel has built up a 4 to 5 year buffer of developed silicon so that if any generation or two is worse than their competitors, they won't be caught flat footed & have to pay contra-revenue to incentivize manufacturers to use their chips.
Most notably this can be seen in review samples of chips sent out, where AMD and Qualcomm will send out chips dated as being made shortly before being sent per the packaging, Intel will send out chips that are dated 4 to 5 years old, as they are from the first or second run of silicon that Intel did with that chip design.
It is, hence why they got fined in the EU recently. That being said, so long as AMD & Via are still alive and producing x86-64 chips, they'll avoid the brunt of anti-competitive regulations.
>Intel will send out chips that are dated 4 to 5 years old, as they are from the first or second run of silicon that Intel did with that chip design.
Fabrication process advancements (i.e. die shrinks) in that time span are enough to rule out what you said with near complete certainty—let alone the notion Intel is somehow shelving architectures for 4-5 years at a time.
Architectures may be developed over those spans, sure—but not shelved for that long.
I'm not saying that they have 4 to 5 full micro-architectures shelved (that'd be insane), but from talking to Intel Oregon people they do keep a deep stock that will buy them a few years. Think incremental improvements like you see from Intel every year, not 20% or 30% performance improvements.
Intel desktop 10nm won't actually arrive until late 2018/early 2019 with Icelake. Due to yield issues, those first 10nm Cannonlake parts are only for low power/mobile applications. To compensate, they are adding another 14nm iteration called Coffee Lake with up to 6 cores.
I wonder how this will affect the software side. Programmers and compilers.. how do you interface with a NN based prefetcher to avoid regression and hiccups.
If it's a truly adaptive branch predictor, one could naively believe it will yield good results for most inputs. Even if it's an absurdly deep pipeline with a cache miss penalty worse than the Pentium 4, it would be premature optimization for the vast majority of programs.
Most processors only let compilers give a hint of strongly taken or strongly not-taken. I don't have any numbers which show regressions, but a NN should be able to take the compilers hint.
I believe Intel has been doing this for a while, or at least it's definitely not the first time I've heard of the idea.
‘Neutral network’ is more of a marketing term, though somewhat accurate in that it's a number of perceptrons doing statistical branch prediction. No multilayered networks (as far as I know) have been used in any CPU to date.
Can we all give Mark Papermaster a healthy round of applause? Amazing what he accomplished. Who would've thought AMD would ever even come close to Intel in performance again?
Seeking Alpha is so pushy about registration. 2nd page or below the fold, it stops you. Use mobile and it stops you (and forces you to use their app). Their app continually pushes for registration and is very neutered without it.
If they are this pushy to get my information, how pushy are they in selling it or in spamming me?
Im currently using Throw Away Email, and wow, they are nasty pushy. And yes, I'm polluting their Database. Too bad, so sad :)
Theyre digging in "what kind of brokerage work do you do" and then asking very pointed money-market questions.
When you get to "Suggestions" they aren't suggestions; they force you to enter multiple stocks.
So fuck them. Here's a Copy/Paste.
_______________________________________________
Summary
In my previous article I showed how speculation going into Zen's launch greatly exceeded speculation going into the Athlon and Opteron launches.
In this article, I take the comparisons further.
I compare both the revenues expected/attained by AMD and the market caps into the 3 launches.
The Investors of today like AMD nearly 4x better than when Opteron (or Athlon, for that matter) was looking pretty.
My thesis on Advanced Micro Devices (NYSE:AMD) is as follows:
I am negative on AMD because it is my belief, based on leaked benchmarks, that the company's incoming Zen-based CPUs will still underperform Intel's offerings materially, when it comes to single-threaded performance. I believe AMD will compensate for this the same way it has, historically: by selling CPUs with a higher core count. This puts AMD at an intrinsic margin disadvantage, which is made worse by the fact that AMD is fabless whereas Intel (NASDAQ:INTC) runs its own factories (so AMD loses the foundry margin as well).
I also think AMD got way ahead of itself. In my last article, titled "Advanced Micro Devices: Then And Now", I explained this by showing AMD's behavior versus how it behaved during 2 extraordinary past events: the launch of the original Athlon and original Opteron. In both those instances AMD stepped ahead of Intel while with Zen AMD is only expected to get close to Intel. Still, in both of those instances AMD had nothing like the present run going into those launches.
While I am negative on AMD, I am not negative enough to sell it short. I don't think the timing is right for such a move. I also think that there's the chance Zen gets "close enough" even if it underperforms, allowing for substantial share gains while selling at a discount. Thus, before considering selling short AMD, I'm waiting to see how Zen actually performs.
With that being put out of the way, I'm writing this article because I think the prior one didn't quite filter through. From the comments, one would guess that the message "AMD is much more speculative going into Zen than going into Athlon or Opteron" didn't register. The criticism was far and wide, including such things as:
This time is different because the market is larger.
This time AMD is different because it also has GPUs and custom silicon and whatnot.
That's only share price (it was an adjusted share price …), not market capitalization.
AMD was thought to go bankrupt this time, and back in 1999 or 2003 that wasn't the case.
Etc, etc.
Well, I feel I need to give some more detail here. This detail will show that sentiment today is even more extraordinary than what the previous charts showed. So here we go.
Revenues, Now And Then
Over the last 12 months, AMD has produced $4.24 billion in revenues. For 2017, including the early Zen launch, expectations are for $4.55 billion in revenues. Now, remember:
This includes Zen.
This includes GPUs.
This includes custom solutions (consoles, etc).
Why is this revenue number relevant? Consider the following regarding the past two episodes I mentioned:
Athlon. In 2000, AMD had revenues of $4.64 billion.
Opteron. In 2004, AMD had revenues of $5.0 billion.
In short, AMD's revenues were larger in both those years, versus those expected for 2017 including Zen. This includes Zen, GPUs and everything else, thus rendering the arguments that AMD is &...
The original Xbox One used a Jaguar[0] based chip, which is really quite a different bread than the Bulldozer or Zen micro architectures. (It's closer to what you'd find in a netbook than in a high end machine, minus the custom memory cache/controller setup AMD designed for Microsoft.)
While it's possible they'd switch micro-architectures, if history is any indication, I suspect it's more likely what you'll see is a die-shrink (28nm -> 14nm) combined with higher clock speeds. And hopefully a faster memory controller (I think that's more the bottleneck than raw CPU power). But that's entirely my speculation.
Yes, Scorpio might possibly include Zen cores. AMD has on multiple occasions stated that one of their design goals has been to design a core than can scale from ultra power efficient embedded applications all the way to high performance. I think their motivation is to consolidate all their efforts into one core.
Now I don't think they will be able to fit 8 zen cores into a Xbox SoC especially since the Vega GPU also have to fit in there, but quad core with 4 logical cores would still probably be better than the existing Puma cores.
I wasn't expecting it to compete with the higher level I7, or ANY I7.
hopefully this puts AMD into a position to capture the enthusiast PC market, which Intel has been stagnating on for generations now.
Even the new 7700k there's almost no performance bump from the 6700k, which isn't all that great itself from the upper end I7s from previous generations.
I want to believe, I really do. Intel can't offer a compelling reason to upgrade my 5 year old 3930k--there's no question that the desktop market has stagnated in the face of a weakened AMD. Can anyone enlighten me as to what video encoding and 3d benchmarks might omit that a more well-rounded passmark score would not?
More competition is good, though I dont have many hopes now. I switched to Intel starting with "Cores". I don't see myself upgrading my 4690k anytime soon, but I hope but the time I need, AMD would have some nice offerings.
This being HN, I'm hoping someone more competent than me in the area of CPU architectures can comment, but as a scientific programmer some of the details lead me to adopt a very cautious attitude for now, especially given the 95W TDP. Given the architecture, producing a performance-competitive 8C/16T chip at 68% of Intel's TDP (95W vs 140W) seems very dubious.
Most of the things they're touting in the slides amount to little more than finally catching up to Intel (SMT, branch prediction (minus the neural network thing), cache bandwidth, etc.). Also, some of the areas where they seem to exceed current Intel architectures may turn out to be a mixed blessing. For instance, I'm not sure if doubling L2 caches (512kb/core vs 256kb in Intel CPUs) is a great idea given bandwidth limitations and associativity effects. We also don't know the amount of vector registers these CPUs will have, which will greatly influence the performance of optimal CPU-bound code.
Can anyone with more experience comment on the low level details they've released, particularly the issues per cycle and cache architecture figures?
1. Wider retire (probably only when executing two threads)
2. 32B fetch from icache.
3. Wider issue when executing from uop cache.
4. Larger icache
5. Larger L2 cache helps hit rate.
6. Larger schedulers, leading to larger practical instruction window
7. Split FP/Int pipelines, meaning more execution resources and instruction window in mixed code.
Mixed:
6. Very different branch prediction. AMD has used perceptron predictors before, the talk about AI-driven CPU is just better marketing on this. Perceptrons work much better on some loads, worse on others. What has been released about their newest seems to point to some kind of mitigation of the worst cases of their predictors. Could be great, could be a letdown.
7. Narrower FPU and load-store. This obviously hurts on wide AVX loads, but the adoption of wide AVX so far has been terrible. Narrower FPU gets them shorter distances on chip, leading to higher clocks.
Disadvantages:
8. Their split integer schedulers can have bubbles caused by their inability to move instructions between lanes after dispatch. Part of their prediction talk seems to be focused on mitigating this, but it's still something that Intel CPUs just don't have to worry about.
9. Less AGUs. On any code without rmw ops, they probably can't maintain full memory throughput.
10. Larger L2 cache probably hurts L2 latency.
Current Intel CPUs are in effect designed for 4uop/clock throughput, with both the frontend and the retire limiting them to that over long term. Especially when both threads are running mixed code, this one appears to have a higher peak throughput. How often it can hit that peak is another question -- if I had to make a quess, I'd say that Zen runs highly optimized code better, but the lower latencies of the modern Intel CPU makes them better at worse-optimized code.
It's all wasted ink and pointless flapping of gums until a tech magazine gets their hands on a production model with a set in stone RRP, and does third-party benchmarks. You can't eyeball the actual performance from the marketing material, and neither can anandtech.
197 comments
[ 3.2 ms ] story [ 230 ms ] threadMan, if that holds true in more benchmarks, that's fantastic news. AMD chugged along by "competing" further and further down Intel's product line more or less since Sandy Bridge, if they can genuinely compete with the i7 now I can't wait to see the next few years. Intel has always seemed content to release incremental improvements until they get kicked in the pants by somebody else, I wouldn't be surprised if they have a bunch of stuff waiting in the wings to maintain their edge over AMD.
AMD fell the moment they failed to improve over K10.
I don't know if the designs were really absurd, or if the market couldn't understand AMD's value prop. Or if the programmers couldn't (or wouldn't) tap into their APU idea.
Multiple papercuts made them fall I guess.
The gaming consoles show how well the concept works, although, sadly for AMD, these deals may be high volume but don't have high margins.
AMD dedicated graphics have only minor issues imho, they're very competitive with nVidias offerings. But in the professional and compute segments they fail to take off, because nVidia established the popular, proprietary CUDA standard there. I don't think there's a (major) technical reason why AMD has much less market share there.
They're likely fairly consistent income, however, even if the margins are relatively slim. For a company in AMD's state, constant income no matter how small is likely a good thing.
AMD marketing tried to put GPU performance in the spotlight. The unfortunate effect was that a moderately informed consumer then had the choice between a somewhat predictable level of graphics performance paired with whatever CPU that came with it if buying AMD, versus a somewhat predictable level of CPU performance paired with whatever GPU that came with it if buying Intel. In most market segments, uncertainty about graphics performance has much less impact on buying decisions than uncertainty about general purpose compute. The (involuntarily, given their portfolio) focus of AMD's marketing on the graphics part of their APUs made it close to impossible to successfully communicate value-for-money advantages in their price range (assuming they had it).
Rumors went around that Intel paid manufacturers to do it intentionally, but nothing was ever proven. However, given how Intel has had similar behavior in the past, it isn't an unreasonable rumor either...
Also IIRC, ryzen wasn't using any form of boost while the 6900k was allowed to push its enveloppe.
Not a time to buy an intel cpu.
If there is no room for boost on Zen parts it's gonna be dead in the water for enthusiasts. My 5820k does 4.6ghz and it's a 6 core part.
That said what's even more important is going to be USB 3/3.1, storage and PCIe performance which ATM on AMDs current platforms are abysmal.
Would be interesting to see the boost stats in that case, 3.7 boost is a bit low for a 6900k part tho everyone I know that had it rubs it around 4.2-4.3ghz.
If AMD left a benchmark to run as a space heater right up to when they started the timed one, the CPU would start at 3.3GHz.
As for the boost, I haven't seen a load that would prevent a core i7 from boosting permanently on air cooling, with water you'll have no problems keeping it at 4.3-4.5 even with an AIO...
This is incorrect. They did a "both CPUs locked to 3.0GHz" test a few months back, today they did a "Intel CPU running absolutely stock, AMD one locked to without boost" test.
But Intel also has an ability to put pressure on OEMs (because of monopoly position and massive supply volume, so if they will not sell CPUs to some vendor as a punishment for buying AMD, AMD will not be able to supply enough CPUs on their own) and brand loyalty, so probably they will keep their "high-end" price points.
Also are the 8 cores they're claiming 8 physical cores or logical?
Shame they still seem to have a 15-year-old in charge of marketing though. Should have just stuck with "Zen".
Also, it sounds like "Ryzen" is their consumer product line. Their server product line might be called something else. It makes sense to have multiple product lines with different names and to refer to the common microarchitecture as "Zen".
Also are the 8 cores they're claiming 8 physical cores or logical?
AMD was never expected to beat Intel's best, they are expected to beat Intel on price performance.
Also with Intel you can upgrade to more performant Xeons. What can you upgrade AMD to?
For both companies, you typically need a different socket/chipset/mainboard for the server chips.
Personally, I will be buying a new PC with Ryzen as soon as they get out.
And by the way, Xeon CPUs aren't usually faster than regular Intel chips. They may have more cache or more cores, which might speed up some programs. But otherwise, they're the same. Or even slower since their motherboards don't overclock well, overclocking not being a big selling point for servers.
Biggest difference with i7 vs xeon, xeon can handle sometimes upwares of 1tb of ram now. i7 maxes out at 64gb. You be the judge. i7 were usually 32gb, especially the LGA1100 series. The LGA2011 were @ i7 handled 64gb. LGA 2011 Xeon's are all 64GB+. Also support for ECC ram. So basically if you are not pressed for raw CPU, and you need just a lot of cpu for simulations, VM's etc, xeon is your go to CPU. Xeon LGA2011 series has some crazy CPU Socket configurations, you can get a lot depending on the motherboard vendor. I think some as many as 4+ CPU Sockets. i7 cannot run in parallel with another socket I believe. Just like with the LGA2011 1600 vs 2600 series xeons.
The previous generation (5820k, etc) were only officially rated for 64GB, but they too worked fine with 128GB in practice. Intel just never updated the spec sheets once 16GB DIMMs became available.
And even if you don't use GPUs, now there are M.2 SSDs which commonly use 4X PCIe lanes, see https://en.wikipedia.org/wiki/M.2
It's like saying that mercedes with 2.0 engine is the same as 65 AMG. So writing that they are the same is like writing that every car is the same because they all drive.
You do not overclock in 99.99% cases Xeon server cpus. You do not buy 12 core xeon to use one core, don't you? of course Xeon is faster for the use case you are buying it. A LOT faster in many cases than what you called "regular" chips. You are writing about home use cases from what I see and Xeon series is mainly for server use.
It's a bit cheaper, and the downside is an occasional bad cat pic, or maybe an extra browser crash per month to add to the 399 one already gets from poor code ...
Xeons are not necessarily more performant, in fact, often they are less performant than their desktop brethren...
They're server CPU's, designed for long uptime, typically underclocked compared to the same silicon in an i7/i5 package (they focus on stability, not sheer performance), support ECC memory (again, not performance related, rather stability), support more memory, etc.
But as long as it stays under the $1100 of the 6900K I think they can secure the mid to high-end enthusiast grade market which is good because I am tired of intel controlling that portion of the market
If they can't reach at least 4ghz it wouldn't matter.
Games are not optimized well beyound 3-4 cores these days I can disable 2-3 cores on my 6 core CPU and maybe lose 5% fps in CPU intensive games.
Intel can push their clocks pretty darn high especially with OC which now comes out of the box with every enthusiast grade motherboard.
If you end up having a 3.4ghz part going against a 4.5ghz part the 4 extra cores might not matter.
Even in professional use cases adobe doesn't scale above 4 threads like at all and many video encoders also don't scale some even don't scale beyond 2.
The only thing that really scales well universally is 3D rendering.
For ANNs you want as many threads as possible I would think.
Xeon SOCs and what ever AMD would pull out to compete with 48 and 64 thread parts would probably be something quite different than this.
I've mostly used open source encoders, but they all scaled well with more cores, and the Adobe suite has come a long way. It wouldn't make sense for a heavy user to get anything less than 6-8 cores using Adobe.
Heck, even 99% of the "tech media" didn't notice (or didn't care to make a big deal out of it) when Intel swapped Core-based Celerons and Pentiums for significantly weaker, but much more profitable to build, Atoms. The consumers never stood a chance against Intel's progressive rip-off.
Let's see how the market settles in 2-3 years, after the Zen architecture is fleshed out across device lines, and after Qualcomm commences its entrance in the notebook market with 50%-100% of "Core i7" performance (depending on which Core i7 [1]).
[1] http://www.laptopmag.com/articles/intel-renames-core-m-core-...
This chip definitely seems rad, don't get me wrong--I just think the enthusiast market is sensitive to what this isn't, at present, doing.
If you want 6 cores you can get a 6800k for $150 more, but it also has a higher TDP and requires a more expensive motherboard (X99 vs Z170 chipset). This is going to put the total price up by at least $300.
As you say most games still favour single-core performance so paying extra isn't worth it. DX12 and Vulkan will start to change that though, so let's see...
http://www.anandtech.com/show/10907/amd-gives-more-zen-detai...
3.4+ GHz for 8-core base frequency (TDP is apparently 95W) is good news.
I'll be extremely surprised if it's faster than a 4-core core i5 skylake which runs at 65W TDP, in the $200 per cpu price range.
It's worth mentioning that the i7-6900K has a TDP of 140W [1]
I have to admit when I first saw 95W I wasn't impressed because I didn't realize how high the TDP of the 6900K was.
[1] https://ark.intel.com/products/94196/Intel-Core-i7-6900K-Pro...
I know, I was surprised too... I used to buy AMD, but haven't in some time now. Perhaps I will again...
I've been hearing that for the last eight years and it has never been proven to be true, in real world benchmarks and $$$/performance by the time equivalently priced tiers of competing varying $125, $195, $250 and $350 CPUs are available for purchase.
We don't know yet whether Zen is as good as Intel CPUs, but the statement "You're thinking the old AMD architecture, which was quite different from that of Intel, and it sucked. The new Zen architecture is much more similar to that of Intel." Is absolutely correct, and is something that was not correct about any AMD CPU released in the past 8 years.
I used to only buy AMD. Im not exactly sure where they are up to now (probably about 8 years if not more since I touched them)
but i definately remember the bulldozer hype and subsequent flop. lies about the number of cores and other things that just made me say "my computing is to important to my life to play that game".
I was also privy over the years to some test builds of machines including head to head amd vs intel. it is truly painful to say that not only were they often more expensive (new) for the performance, they were also horribly unreliable, pausing, and doing all sorts of stuff that made us sad.
hope thats changed, but saying there is any similarity between the current in use intel vs amd range is very far from the mark.
Are you kidding or did you mistype? Current AMD processors don't generally compare favorably to Intel with respect to some metrics, like absolute performance per core or performance per Watt. However if there is any metric where they match or surpass Intel, it's actually perf/$.
The reason is simple: perf/$ is the only metric where AMD has some control: they can decide to lower and lower their price down to wherever they need to be. If AMD wasn't competitive on at least perf/$ they would be in a much worse situation.
Here is one example I witnessed first hand recently when doing some video work: a good old $100 3.2 GHz 8-core Phenom matches or surpasses most ~$200 Intel processors when encoding MPEG2 files to H.264 with the Linux x264 encoder.
This is bullshit through and through.
Between $100 and $150, you get any of AMD's FX 6xxx and 8xxx line. We're talking about 6 and 8-core processors whose performance are, in the very least, comparable with some of Intel's i5 and i7 offerings.
And Intel's comparable offerings cost over twice what AMD's processors sel for.
The FX-8320E when overclocked can compete in multi-threaded computational workloads, but will use a lot more energy. For games (which is the main area where users stress their processor), the i3-6100 will still be faster. See http://www.anandtech.com/bench/product/1709?vs=1682. The bigger FX-8s are too expensive and would be pitted against i5s, where they always lose.
The only recent exception I ran across is Watch Dogs 2, which is sufficiently optimized for multiple cores and AMD processors to make the FX-8370 for once competitive against Intels i5. But that is the first game since the release of those processor where that is happening.
What does work for AMD is the Athlon X4 8.0K, the area under 100€. Intels Alternative is the Pentium G4.00, which is worse for most things.
If those results are accurate, they do seem to support that AMD FX CPUs are still competitive with new Intel CPUs in the same price range.
The FX has the potential to be a fast processor in specific workloads. Yes, that fits to what you said. Certainly not when single-threaded, but in multi-threaded workloads it can still be fast. But faster than the i5 it is only very rarely (which, yes, is less harsh on the FX than I was in the comment above). It is a pity that its potential (4 years ago it compared to a weaker i5) could not be used in meaningful ways in common software.
By now it is clearly not a good choice anymore: It uses more energy, depending on what you compare it with it is more expensive or costing the same, and it is on an old platform (AM3+) that won't see further updates.
> They're evenly matched on other tests (e.g. Handbrake & "Agisoft PhotoScan
No. We don't seem to have the same benchmarks. I'm looking at http://www.anandtech.com/bench/product/1709?vs=1646, where the i5 is clearly faster in the Agisoft and Handbrake benchmarks. It is mainly in the artificial multi-threaded benchmarks as well as Linux 7zip and Linux OpenSSL that the FX can win. For games the FX is almost always behind, but often close, however, those are partly limited by the gpu and thus not too meaningful.
> If those results are accurate, they do seem to support that AMD FX CPUs are still competitive with new Intel CPUs in the same price range.
No. They mean get an i5: Way better in single threaded workloads which is what counts for 90% of software, faster in realistic multi-threaded workloads. Get an FX if all you do is run OpenSSL and 7zip under Linux, or if already having a good AM3+ board (for that scenario it still is good enough).
There are 87 benchmarks there. I do not believe they re-do all 87 each time someone releases a processor. It's more likely that they are reusing old 8370 results and comparing with new 6600 results. I question whether all variables (OS, drivers, minor version numbers of software, etc.) are the same. (I'm not saying that re-doing the 8370 tests would favor AMD, just that it would give more accurate results). They don't mention what settings they used for the software, and what exactly each program is stressing. Splitting the PhotoScan results into five seems very arbitrary. Two decimal places in each result with no error bars, no indication of how they performed the measurement, no indication of the number of times they performed the test leads me to think that it's not a very well conducted set of tests.
> > They're evenly matched on other tests (e.g. Handbrake & "Agisoft PhotoScan
> No. We don't seem to have the same benchmarks.
We clearly have the same benchmarks. "evenly matched" was a poor choice of words, but I don't see a ~23% advantage for the 6600 to be very significant, when other results have the 8370/6600 100% faster than the 6600/8370.
> They mean get an i5: Way better in single threaded workloads which is what counts for 90% of software, faster in realistic multi-threaded workloads.
For browsing, office and most games the 8370 and 6600 are close enough that the average user won't notice a difference. Most users don't spend any time waiting for software that eats 100% of <= 4 cores, which is where the i5's advantage lies. What you find "realistic", I may not, and vice-versa. It depends on the user, and what he's using his computer for.
The i5 may well be the better choice for most users. It certainly has a number of advantages: less power-hungry, newer chipset/features, more upgrade options. My point is simply that the FX cpus are still price and performance competitive with new i5s. Which is better depends on the needs of the user.
Well, every gamer does, and this exactly describes what web-browsing does. However, the FX is of course not too slow for that. Just slower.
> It depends on the user, and what he's using his computer for.
Sure. I was joking above with "get The FX if all you do is 7-Zip on Linux", but yours is of course a correct statement. If you identified correctly that the FX will perform better for your workload, then it is a better choice (if okay with higher energy usage and not having further upgrades, which just might be the case, and it is not like intel is known for long socket support anyway). I for example use Funtoo, which is a Gentoo variant, meaning I compile stuff all the time. I'm on an older processor, but I assume the FX would be a good choice for me.
> It's more likely that they are reusing old 8370 results and comparing with new 6600 results.
I think exactly for the FX-8370 that might be not true, as it is the processor you'd have wanted to compare the new skylake-i5 against. But for the others and older models exactly that will happen. I see no big issue there.
> I don't see a ~23% advantage for the 6600 to be very significant, when other results have the 8370/6600 100% faster than the 6600/8370.
~23% advantage is very significant to me. That is way off of measurement errors. 100% difference just matters even more ;) Also, that it is most often pro-i5.
Ah, I just hope Zen will turn out as great as I hope and is announced and I can start recommending and defending those processors :)
http://www.bit-tech.net/hardware/cpus/2011/10/12/amd-fx-8150...
http://www.theregister.co.uk/2015/11/06/amd_sued_cores/
Note that they changed that with ZEN
http://wccftech.com/amd-one-fpu-per-core-design-zen-processo...
* http://www.leagle.com/decision/In%20FDCO%2020160408M22/DICKE....
AMD has been teetering on the edge of bankruptcy for quite awhile now. It's unclear how much rope they have left.
They have actually dug themselves out of that hole over the past two years or so. Their engineers are good, but their CFO is a wizard. Somehow he managed to take a company with bad short-term prospects and very questionable long-term ones that was up to it's eyeballs in very expensive short term debt and convert that into cheap long-term debt. Then once things started going better for them and the stock started going up, they have been turning their debt into equity pretty much as soon as possible.
AMD currently is in no risk of bankruptcy:
http://cdn.wccftech.com/wp-content/uploads/2016/10/AMD-CFO-C...
Intel Xeon E5-2665 ($80 on eBay) = 8 cores, 2.3-3.1GHz
Intel Xeon E5-2660 ($60-65 on eBay) = 8 cores, 2.2-3Ghz (so you can buy two for $120-130)
Intel Xeon E5-1650 ($97 on eBay) = 6 cores, 3.5-3.8GHz
E5-2665 beats FX-8350 hands down, and is cheaper.
http://cpuboss.com/cpus/Intel-Xeon-E5-2665-vs-AMD-FX-8350
http://cpuboss.com/cpus/Intel-Xeon-E5-1650-vs-AMD-FX-8350
I can buy a whole workstation with E5-2665, motherboard, 32GB RAM, 1TB HDD, all for $414, shipped.
http://www.ebay.com/itm/162274529994
I doubt you can find something like that with FX-8350 and cheaper. And it will still be slower.
Depends a lot on your electrical cost. Some people live in places where the power costs $0.05 per kWh. Some people live in places where it cost $0.36 per kWh (Hawaii).
If somebody wants a machine with a lot of xeon cores and a ton of ram (but noisy), you can buy a dell poweredge r910 server with 4 sockets x 8-core xeons and 128GB of RAM for like $1100. Makes an excellent xen or kvm hypervisor dom0/host development machine. It's just noisy, so in a home environment, put it in your garage or something on the far end of a cat5e cable.
I have a AMD FX 8320 workstation, and I assembled it brand new for less than €400.
In fact, if you take away the €130 referring to the price of an AMD FX 8320, you get the remaining budget for the mobo, RAM, and HDs.
My wife's laptop died, and she just wanted me to do put together something cheap for her office (she has a tablet and work laptop). I bought myself a new motherboard (keeping my CPU), gave her my old one and put in an older CPU I had.
A year and half later, I upgraded my CPU to an FX-8350. My old CPU (Phenom x4) moved to the basement NAS/Plex server, a nice upgrade to Athlon II CPU (without requiring a new motherboard).
Had I been on Intel, I would simply not done those upgrades because they wouldn't have been worthwhile. Basically every CPU upgrade requires a new motherboard. And in my experience, nearly every time I've had to get a new motherboard, my memory or video card or some other component is now outdated, and suddenly I'm basically building a completely new PC.
those don't exist
less performance than a $133 intel e5 2650
and neither are worth buying today.
currently, amd does terribly in pretty much every measure. absolute performance, performance per $, throughput. absolutely anything that matters.
There are some promoted intel chips that are worse.
but the only thing amd have been doing well at in recent years is hyperbole with a healthy dose of shilling. So i cant disagree with anyone on here that wont believe anything they say until they see it.
around this time last year we had
http://www.theregister.co.uk/2015/11/06/amd_sued_cores/
Gamers need perf/core. Datacenters need perf/power.
They're only viable for "cheap CPU" in the < $100 range which is low margin.
In the past, AMD avoided artificial "enterprise" segmentation with security features like IOMMU, but did not always have the necessary BIOS, board (e.g. TPM) and open-source tooling that Intel solutions provided.
I'd really like to see Ryzen meet or exceed Intel i7-6700K/7700K single-core performance using boost mode clocks with a reasonable cooling setup. That would eliminate any reason whatsoever to choose Intel for gaming until their 10nm product arrives Q3 2017.
Comparatively, AMD has been building chips to order as of late, since they do not have the money to build a bunch of chips that may or may not sell, so your looking at a 3 to 4 month delay before you can get a new chip like Ryzen after announcement.
The cache arrangement (see pic halfway down) and better branch prediction are interesting though, cache has always been a space where Intel has been able to use way less transistors to store the same amount of data.
http://semiaccurate.com/2016/12/13/amds-zen-becomes-ryzen/
Can you elaborate? I'm fairly certain Intel deviating from their existing roadmap to such a degree and on such short notice would be extremely difficult at best.
Most notably this can be seen in review samples of chips sent out, where AMD and Qualcomm will send out chips dated as being made shortly before being sent per the packaging, Intel will send out chips that are dated 4 to 5 years old, as they are from the first or second run of silicon that Intel did with that chip design.
Fabrication process advancements (i.e. die shrinks) in that time span are enough to rule out what you said with near complete certainty—let alone the notion Intel is somehow shelving architectures for 4-5 years at a time.
Architectures may be developed over those spans, sure—but not shelved for that long.
I had understood that new generations of cpu to be years from beginning of development to store shelves.
We're slowly power gating our way to clockless CPUs.
‘Neutral network’ is more of a marketing term, though somewhat accurate in that it's a number of perceptrons doing statistical branch prediction. No multilayered networks (as far as I know) have been used in any CPU to date.
Intel uses more traditional predictors but with larger buffers. Both approaches have situations that are more or less favorable to them.
http://seekingalpha.com/article/4030174-advanced-micro-devic...
If they are this pushy to get my information, how pushy are they in selling it or in spamming me?
Theyre digging in "what kind of brokerage work do you do" and then asking very pointed money-market questions.
When you get to "Suggestions" they aren't suggestions; they force you to enter multiple stocks.
So fuck them. Here's a Copy/Paste.
_______________________________________________
Summary
In my previous article I showed how speculation going into Zen's launch greatly exceeded speculation going into the Athlon and Opteron launches.
In this article, I take the comparisons further.
I compare both the revenues expected/attained by AMD and the market caps into the 3 launches.
The Investors of today like AMD nearly 4x better than when Opteron (or Athlon, for that matter) was looking pretty.
My thesis on Advanced Micro Devices (NYSE:AMD) is as follows:
With that being put out of the way, I'm writing this article because I think the prior one didn't quite filter through. From the comments, one would guess that the message "AMD is much more speculative going into Zen than going into Athlon or Opteron" didn't register. The criticism was far and wide, including such things as: Well, I feel I need to give some more detail here. This detail will show that sentiment today is even more extraordinary than what the previous charts showed. So here we go.Revenues, Now And Then
Over the last 12 months, AMD has produced $4.24 billion in revenues. For 2017, including the early Zen launch, expectations are for $4.55 billion in revenues. Now, remember:
Why is this revenue number relevant? Consider the following regarding the past two episodes I mentioned: In short, AMD's revenues were larger in both those years, versus those expected for 2017 including Zen. This includes Zen, GPUs and everything else, thus rendering the arguments that AMD is &...That said, puts were fairly cheap so I took some as my last gasp effort to see if AMD actually tanked.
While it's possible they'd switch micro-architectures, if history is any indication, I suspect it's more likely what you'll see is a die-shrink (28nm -> 14nm) combined with higher clock speeds. And hopefully a faster memory controller (I think that's more the bottleneck than raw CPU power). But that's entirely my speculation.
[0] https://en.wikipedia.org/wiki/Jaguar_(microarchitecture)
Now I don't think they will be able to fit 8 zen cores into a Xbox SoC especially since the Vega GPU also have to fit in there, but quad core with 4 logical cores would still probably be better than the existing Puma cores.
hopefully this puts AMD into a position to capture the enthusiast PC market, which Intel has been stagnating on for generations now.
Even the new 7700k there's almost no performance bump from the 6700k, which isn't all that great itself from the upper end I7s from previous generations.
Most of the things they're touting in the slides amount to little more than finally catching up to Intel (SMT, branch prediction (minus the neural network thing), cache bandwidth, etc.). Also, some of the areas where they seem to exceed current Intel architectures may turn out to be a mixed blessing. For instance, I'm not sure if doubling L2 caches (512kb/core vs 256kb in Intel CPUs) is a great idea given bandwidth limitations and associativity effects. We also don't know the amount of vector registers these CPUs will have, which will greatly influence the performance of optimal CPU-bound code.
Can anyone with more experience comment on the low level details they've released, particularly the issues per cycle and cache architecture figures?
http://www.anandtech.com/show/10591/amd-zen-microarchiture-p... (slide 2 on this page)
1. Wider retire (probably only when executing two threads)
2. 32B fetch from icache.
3. Wider issue when executing from uop cache.
4. Larger icache
5. Larger L2 cache helps hit rate.
6. Larger schedulers, leading to larger practical instruction window
7. Split FP/Int pipelines, meaning more execution resources and instruction window in mixed code.
Mixed:
6. Very different branch prediction. AMD has used perceptron predictors before, the talk about AI-driven CPU is just better marketing on this. Perceptrons work much better on some loads, worse on others. What has been released about their newest seems to point to some kind of mitigation of the worst cases of their predictors. Could be great, could be a letdown.
7. Narrower FPU and load-store. This obviously hurts on wide AVX loads, but the adoption of wide AVX so far has been terrible. Narrower FPU gets them shorter distances on chip, leading to higher clocks.
Disadvantages:
8. Their split integer schedulers can have bubbles caused by their inability to move instructions between lanes after dispatch. Part of their prediction talk seems to be focused on mitigating this, but it's still something that Intel CPUs just don't have to worry about.
9. Less AGUs. On any code without rmw ops, they probably can't maintain full memory throughput.
10. Larger L2 cache probably hurts L2 latency.
Current Intel CPUs are in effect designed for 4uop/clock throughput, with both the frontend and the retire limiting them to that over long term. Especially when both threads are running mixed code, this one appears to have a higher peak throughput. How often it can hit that peak is another question -- if I had to make a quess, I'd say that Zen runs highly optimized code better, but the lower latencies of the modern Intel CPU makes them better at worse-optimized code.
(To those asking, PSP is the equivalent of the Intel Management Engine).
noob alert :p
We detached this comment from https://news.ycombinator.com/item?id=13174901 and marked it off-topic.