we simply need a radiator or exchanger design to plug in to our domestic water supply at least I'll be very happy with either free compute or free hot water pov depending.
Who both buys consumer desktop CPUs (without ECC) and runs software that can use AVX512? Games won't use avx512 (not really simd math heavy work on the CPU, and also consoles don't have it). Web browsers won't use AVX512 (similar to games - not really SIMD-friendly raw math computation limited). Video & photo editing maybe will, but GPU acceleration is already more widely available & faster than avx512.
AVX-512 software is appearing. For example, for neural net inference: https://NN-512.com
AVX-512 is a pleasure from the programmer's perspective, much nicer than writing software for GPUs or for earlier SIMD instruction sets like AVX2/SSE
Versus a GPU, the AVX-512 machine model is just much simpler, and there are no drivers or proprietary compilers
Versus earlier SIMD instruction sets, AVX-512 provides all the instructions you want (permutations, floating point conversions, etc.) and they're all masked so edge cases are easy. AVX-512 Foundation is a really nice instruction set and it has a bright future
It's a bit less messy if you ignore the GPU-like Xeon Phi (Knights Landing and Knights Mill) and the failed Cannonlake that Intel definitely would prefer everyone pretend never happened. But that still leaves you with a progression of four levels of expanding support, and the Cooper Lake mess of having BFLOAT16 but not the stuff added in Ice Lake and Tiger Lake.
If you don't cut down the support table to eliminate irrelevant niche products, it's most of the way to being a Gray code.
I think Icelake and Tiger lake not having BF16 is going to be a big issue especially for machine learning applications, I can’t find what Rocket Lake supports yet but I’m guessing it’s going to be similar to Tiger Lake.
There is also the big unknown of what kind of support AMD will bring to the table.
I think Intel is banking on people using OneAPI more than writing AVX-512 code using intrinsics or compiler magic.
At that point compatibility might be not as much of an issue if you ignore potentially drastic differences in performance.
If you limit yourself to the Foundation instruction set (AVX-512F) you get a lot of benefit and maximum compatibility. That's what https://NN-512.com does
Single thread floating point: +19.0%
Multi-thread floating point: +19.5%
Single thread integer: +13.0%
Multi-thread integer: +7.3%
Jesus. Just overclocking the RAM for the 10700k would give a similar performance increase. Along with 291W, this 10nm->14nm backport is not working out well for Intel.
PS My work desktop died on Wednesday afternoon during the SpaceX SN10 flight. I rushed to my local computer store to replace it. Out of the entire store, only 2 motherboards models with Intel's LGA1200 are in stock. All the other motherboards on the shelf were for AMD.
I picked the i5 10400f as it's under deep discount. I asked around and everyone agrees that Intel is competing only on price point right now.
"The new generation Rocket Lake is the combination of two different backported technologies. Intel took the Sunny Cove core from 10nm Ice Lake, and re-built it on 14nm, calling it now Cypress Cove. Intel also took the Xe graphics from 10nm Tiger Lake and re-built those on 14nm, but these are still called Xe graphics."
Just filter out Charlie Demerjian's grudge against Intel.
Companies don’t backport because a core is designed for a process. It’s efficiency and architecture is based on the transistors that it will use from the design start, caches are sized based on the process too, and a lot of the architectural gains are due to the added transistor counts. If you backport a CPU you lose the efficiency of these new transistors so the energy use goes up and the clocks likely go down as well. Cores take up 2x the size, or at least a lot more than the older core did on the older process, so costs go up too. You can either cut out bits of the core and lose performance or eat it on area and therefor cost. Backporting a design not made for portability isn’t a lose/lose proposition, it is a lose/lose/lose/lose proposition. But Intel is desperate so…
Yep, he's been saying the sky was falling for some time, and it turns out he was pretty accurate in his statements on Intel's problems.
This review makes it clear that unless Intel magically gets a process improvement they're entirely screwed.
This is precisely the same kind of design issue that hit with the latter end of the P4 era, where thermal limitations were so severe that they were crippling the chips. I'd have to do some digging, but I absolutely remember reading an interesting set of benchmarks that made it clear that you were 99% likely to thermally throttle those P4s even with proper thermal paste and an improved high-end heatsink/fan.
AMD will make gains for sure at this rate, but TSMC can't serve the whole market. Intel owns fabs, and these rocket lake chips will sell to the mainstream just fine.
I expect intel's fabs to be a big asset in these COVID times with sold out computers.
This chips really don't seem to be a good use of Intel's limited fab capacity. The CPU cores are substantially larger while offering basically no benefits other than AVX-512 support. It looks like Intel would have been better off pairing the upgraded IO and integrated graphics with their existing, smaller Skylake cores.
Pretty much every Rocket-lake score is faster than Skylake across the board.
I'd definitely buy a Rocket Lake over any old Skylake based off of these reviews. Rocket Lake seems a wee-bit inferior to Zen3, but AMD Zen3 is pretty much sold out, so if you need a computer, Intel is still there.
But Intel doesn't need the incremental performance increase in order to sell their processors. With AMD so supply-limited, Intel should be trying to sell smaller, cheaper processors that they can produce in larger volume with higher margins.
> Pretty much every Rocket-lake score is faster than Skylake across the board
Except for gaming, where rocket-lake is across the board slower. Which is important since gaming was Intel's last holdout prior to the arrival of Zen3. So if you're gaming this is worse, and if you're not gaming why would you go with this over a 3900x at basically the same price? Even verse skylake if you're doing productivity stuff a 10920x seems likely to be consistently faster thanks to 50% more cores (and way way more pcie lanes if you're into that)
The only reason to get this seems to be if the unique combination of avx512, no ECC, and only up to 8 cores is somehow ideal for your use case.
Ignoring both sides of the performance debate entirely, Intel found a way to increase utilization of 14nm process in a time when sector-wide demand exceeds supply. It seems very smart to use 14nm because it presents a lower opportunity cost versus getting worse yield on 10nm process node and taking capacity away from laptop chips, an area where AMD has been struggling to meet demand.
TSMC is making huge investments into new fabs. And AMD doesn't have to take the whole market. They just have to take enough that Intel loses too much economy of scale to sell at profitable prices. Intel's fabs don't mean anything if their volumes are too low-- and they're dropping rapidly.
Their fixed costs & massive R&D requirements have to be paid for by shrinking revenue: Projected 1st quarter 2021 revenue is down about 12% over the same quarter in 2020. This isn't COVID-related: AMD's projected revenue for the same period jumped 80% year over year.
Intel is far from okay. They are far from having a safe path due to production bottlenecks by competitors: TSMC's 7nm capacity-- which is what AMD uses-- is on track to increase 18% in just the next quarter alone. I don't know how sustainable that production growth is, but it doesn't have to last long when they're bringing new fabs online and Samsung's fabs may be taking on some of Apple's M1 production, further freeing up TSMC capacity.
I don't think I'm saying anything particularly controversial or surprising or new when I say that Intel probably has about 2 years to figure out their own fab problems before they're well out of things for the next decade or so, having to claw their way back the way AMD has done.
Assuming that you're thinking they'd be looking at a R5 3600 ($179.99). For gaming the 10400F ($109.99) is basically on par with the AMD. AMD is only worth it for productivity related task, and only then it's a ~15% performance hit.
I think Ryzen 5000 series' stock situation is getting better, both 5600X and 5800X has been consistently in stock in Belgium and Netherlands for about 2-3 weeks.
Exactly. In 2020 the 10400f was about 40% more expensive; now Intel is forced to lower the price of their entire lineup to compete against AMD. These days, at most performance tiers, Intel+mobo combo are slightly cheaper than AMD's equivalent.
Of course, Intel's CPUs have less PCIE lanes, no ECC support, lessor upgrade path, generates more heat, etc.
Finally my old desktop was Intel and I am working on a lot of kernel/VM stuff. I don't want to switch architecture halfway. I did that before and it was painful.
> Finally my old desktop was Intel and I am working on a lot of kernel/VM stuff. I don't want to switch architecture halfway. I did that before and it was painful.
What? Aside from edge-cases resulting from slightly divergent microcode implementations you should not be dealing with that. They are both x86-64. (Forgive me if I am missing an elephant in the room.)
> These days, at most performance tiers, Intel+mobo combo are slightly cheaper than AMD's equivalent.
> Of course, Intel's CPUs have [...] lessor upgrade path
One big + for AMD is their socket system, and keeping it the same for multiple generations via a roadmap so people can plan. Contrast that with intel, who comes up with a new socket every generation forcing you to buy a new motherboard and likely new ram if you want to upgrade. AMD's Socket AM4 was released in 2016 and works with Zen+, Zen 2, Zen 3 and Excavator microarchitectures. If you upgraded every 2 years, that's 3 motherboards and memory sets you would have had to purchase if you went with Intel vs a single motherboard/memory and upgrading only the cpu with AMD. That's quite a premium.
Just a small correction, most of the AM4 motherboards released in 2016 and 2017 do _not_ run Zen 3. Only few vendors (I think ASRock is the only one?) added support for Zen 3 on B350 / X370 / A320. The rest only added support on B450 (2018) / X470 (2018) / B550 / X570 / A520. If you were one of the original Zen buyers in 2017, you are unfortunately limited to Zen 2 CPUs, unless if you had an ASRock motherboard.
Spend enough time in front of a specific computer (or little time in front of a lot of different computers) and you'll eventually come across the same thing. In my (around) 20 years in the industry I've had (mostly laptops) die for various of reasons, most often the motherboard and screen giving out, but also the CPU and harddrives. Happened a few time with desktop computers that been 24/7 under stress as well.
Geekbench 5 is skewed because the 11700K uses the AVX-512 unit for the AES test. It scores 4x over previous gen now, and that skews the whole end result.
The next big desktop release will be whatever comes with the new AM5 socket so they have some time to get their act together. I don’t have high hopes that they can actually compete and now Apple also leapfrogged them with non-desktop chips
Either I had really bad timing or bad luck, but somehow I am ending up with Intel even though I really wanted to switch.
Around 2012 I bought a 4-core Core i5 and after some basic upgrades (max RAM to 32GB and a good SSD) it had more-than-acceptable performance for my work until mid-2018. Even when I dabbled with Machine Learning I borrowed a 1070 from my office and was plenty to run my tasks.
Around end of 2018 I started working more in different offices, so I needed a more powerful laptop. No strong offering from AMD yet, so I ended up with a System76 core i7, 6-core 32GB RAM.
2020 came and with it I got back into working at home. So I got excited about upgrading my workstation. I bought all the components to get a Ryzen 5, a 3700X I think. What a shitshow: had to go through three different pairs of RAM to find one that was actually compatible with the motherboard. Then I had to find a way to upgrade the BIOS, and I could only do it by getting an Athlon CPU. BIOS upgraded, then the original 3700 died. I ended up turning the whole system into a very expensive media center just with the Athlon and continued working with the laptop.
A couple of weeks ago my laptop's battery puffed up, so I decided to go look again into a nice desktop. An 12-core Ryzen 5900X is coming out at 900€ and a basic motherboard is 300€. 10-core Intel i9-10900 is 300€ and a motherboard for it is 180€. Guess which one I ended up taking and that I managed to get running without any issue whatsoever?
Here is to hope that in 6-8 years from now I finally manage to make the switch to AMD.
(Edit: why the downvotes, HN? It's an honest account of my experience.)
What an odd experience. I haven't heard of anyone having similar experiences with Ryzen, including myself. Maybe you got some unusually finicky motherboard?
It's a mini-ITX Gigabyte B450, the Aorus line. From what I found online, there were plenty of cases of incompatible RAM that would simply not boot.
The "odd" thing is that this Intel system I put together ended up being the very same brand. I am regretting it now a bit because I realized that the board can only do 4k@30Hz through the HDMI port and the i9 can do 4k@60, but given that my laptop had a similar limitation and I got used to run with my monitor at 1440p, I think I will hold on to it until I feel like it is time to get a dedicated GPU.
Yeah, that is exactly the problem. I'm wondering now if there is any way I can jerry-rig one of the USB 3.2 ports from the mobo to get 4k@60hz to my monitor, but like I said I am ok with staying at 1440 until prices of graphic cards come down.
A 'basic' motherboard two weeks ago was the same as it is today, a low-end B550, and can be purchased for under 100€.
The 10900 trades blows with the 3900x, particularly for multi-threaded, workstation-type workloads where the AMD part pulls ahead with its two extra cores. And the 3900x can be had around the same price as a 10900.
So your "honest account" comes off as a little disingenuous, which may explain the downvotes.
(Yes, the high-end Zen 3 parts are pricey and have very limited availability right now, but that is afflicting the entire industry - see also consoles, graphics cards and many other things. With their own fabs, intel are uniquely placed to supply right now, but their CPUs are more competing with the last gen of AMD's if you're talking about a workstation)
I paired up a 5950x with a $200 TUF 570 board - a middle of the road offering. One thing to consider is the BIOS version of the motherboard. The Asus one had a sticker with the build version, and what they were stocking in Microcenter was just new enough to support zen3 OOTB. A better board would allow for BIOS flashing without the CPU, and a few folks have struggled with that first flashing. Check the version first.
The threadripper boards are starting at $400 for the current generations. I regret spending what I did on the first generation boards, as they dropped support for them on the third generation. I'm waiting to see a zen3 series threadripper before I'll replace my old 1950x. The even older 3930k i7 was able to keep up on single threaded processes... which surprised me. Funny it really took 10 years before I got that wow factor again. The number of cores that I'd consider 'normal' has certainly gone up. Think the 3930k was around $600 in 2011, the 1950x around $800 in 2017, and the 5950x at $800, so that price point has not changed much. The first two were picked up at launch date, the last was a paper launch that took till January to find.
AMD is beating Intel on performance, but with recent price cuts, Intel is offering VERY good value right now and also much better availability. If you don't need the absolute best, Intel is the way to go easily.
That is the most succinct and business-minded summary in the entire thread.
Relatedly, I am surprised by the commenters who believe that the trends in Intel pricing were somehow involuntary. The outperformance of AMD CPUs was clear even prior to the pandemic, and I recall talk at that time about Intel using price pressure to resist AMD's expansion of market share and gross margins. Now that silicon supply is getting tight, market share is all the more relevant because whatever demand AMD can't meet gets caught by Intel. It's not all about benchmarks (although IMO Intel is hanging in there with Rocket Lake). You actually have to get a customer to buy your product in order to generate revenue.
An involuntary price change would be one in which Intel is following a price set by AMD. Think about price takers and price makers.
In this case Intel is undercutting AMD and setting a low price that forces AMD to keep its own pricing lower. This is disadvantageous for AMD because its ability to satisfy spikes in demand is capped; additional orders for unexpected demand have to flow through the TSMC pipeline, which is also used by many other companies. In contrast, Intel can meet demand because it owns its production processes; Intel churns out more product and absorbs the unexpected demand. Per-unit margin matters less for Intel because its share of the unexpected/marginal demand is high.
Your honest account seems more like a series of errors you've invited on yourself. Your motherboard prices are insane. And this litany of spontaneously breaking components make you sound like you're just yeeting the stuff around. Maybe that's all unfair, but it's how it comes across.
Now you're comparing the prices of the i9-10900 to the 5900X, when the 5800X's performance more than matches it. The 5800X is about 10% cheaper than the 10900 here.
If you don't understand the platform and chipsets, use a picker service or even buy a bundle. There are lots of component sellers flogging matched components in tested configurations. You won't get top-spec motherboards but there's no reason to spunk €300 on a desktop motherboard.
> If you don't understand the platform and chipsets, use a picker service
Like I actually did, I have been doing since I can't even remember to build previous workstations, NAS, the dappnode under my desk, and my home server that runs the services I self-host?
Looking around now, it seems that MSI X570 is available at a lower price, but it wasn't when I first checked.
So, yeah, bad timing. But this patronizing is unwarranted.
> The 5800X is about 10% cheaper than the 10900 here.
In Germany it's about 15% more expensive, and my calculation was based on price-per-core than absolute performance, not to mention that the 10900 has a 65W TDP vs 105W for the Ryzens.
>Not to mention that the 10900 has a 65W TDP vs 105W for the Ryzens.
TDP is lie these days .The 65W intel chips will use over 200W in most motherboards.[1] The 10 core you're using will use even more. The 5900X would use 130W max.
Watch your temps, your passive thermal solution may not be up to the task and your CPUs may be throttling or dying prematurely.
Now that is a comment that brings a productive discussion. Thank you. :)
I was indeed weary of keeping only the fanless cooler, so I ended up installing the stock one. Coupled with the fact that it is a big case and it has a lot of fans, I assume that would be enough.
Anyway, what you are saying about the misleading TDP is interesting. Perhaps I will get a 3900X or a 5800X system and if it works out I will return the Intel.
TBF, I still haven't seen enough to convince me to upgrade my desktop from 4690K (which got an RTX2080 in the meantime). Standing plan is to wait for RTX4090 or AMD's equivalent and change the whole box then.
Intel is on my shitlist now for their new "Coalition for Content Provenance and Authenticity" with Microsoft and Adobe. I don't care how fast their CPUs are, they are dead to me.
Basically a large group of companies are forming a group to create an agreed upon fingerprinting system for 'content' (read: everything) to destroy privacy in the name of 'fighting disinformation'.
I'm just going to say, if you don't think this _isn't_ going to be massively abused - take a look at what the CCP, or any authoritarian regime does really, to silence dissent.
Don't have a good enough social credit because you shared an opinion somebody doesn't like one too many times? Intel and Microsoft will just remotely brick all your devices.
> Is there a privacy-respecting solution that you prefer?
Yeah, it's called being media literate, using my brain and being vigilant. I don't need Pat Gelsinger, Satya Nadella, Sundar Pichai, Shantanu Narayen, Tim Cook, Mark Zuckerberg, Jack Dorsey or any other power hungry jackass telling me what to think.
Wont just post it from an official social media account belonging to that journalist fix this issue ?
It starts off with opt in, then they do an update to now opt out, then then do an update to make it harder to opt-out and next thing we know this is something everyone has to live with.
How do we know evidence is valid in a court case? Chain of custody and documentation. Add to that some commitment scheme and trusted time-stamping and publish in real time and you are most of the way there to extremely veridical publishing.
But the main problem is the people most vulnerable to fake media are already bamboozled by bad Photoshop jobs. I don't see how any signing scheme will help.
Ah, so a 50+ years campaign? Yup, that can work if you can have a government with enough attention span.
Edit: don't get me wrong, _not_ doing that for children is evil. But, the results thus far do speak for themselves... And uncle Bob will not take being sent back to middle school all that well.
Honestly everything you say sounds like abusive insanity and not anywhere I'd want to live. I hope you don't make any decisions for other people in any capacity ever because you lack fundamental respect for other people's choices.
The choice anyone else makes to change society to be what they want. Or would they have to lie and take your test to make changes? Only go through your change process?
You have this idea that society is a monolith you can keep people in and out of based on the results of an algorthm, and that's only true for the most authoritarian and oppressive regimes.
That's the world you're trying to create. Social Credit scores as maintained by facebook, etc.
What are you a talking about. A test does not impact the choice anyone has to change society how they want it to be. You are imagining things I have never even hinted at.
Almost everyone get's swimming lessons and tests. And yet, and this might really blow your mind, you can still actually go swimming without actually having had passed the test. No SWAT team will barge down your door and take you away if you go swimming.
Then whats the purpose of the test? Tests are scores used to determine something. You're talking about forcing people to take a test to do what, exactly? Or is it just to indoctrinate them with whatever the test-maker (because it will change every election) comes up with? A thing you must study for but with no reprecussions? Ya right.
Once you create a certification for it, it will become a private passport for access to goods or services, or it will be meaningless. In both cases its either a waste or evil.
With a mandatory test also come mandatory lessons. If you don't pass you have to keep taking the lessons. The lessons will lead to the population having better skills. Seriously I had hoped swimming lessons analogy would make you understand. What is the purpose of swimming lessons and test? Is that done to indoctrinate people that they should not drown? Is having passed a swimming test a private passport for access to goods and services? Get real...
I'm not sure why you keep going down this dark path which I didn't even touch on remotely. You just completely imagine it.
"Think of the elderly" appears to be the new "think of the children" when it comes to erasing privacy (and other freedoms) in the name of "security".
I have definitely fallen victim to that one myself, and have pushed that argument when considering Apple's walled garden in iOS, and how much that has assisted me personally in not having to provide tech support for members of my family.
Unfortunately, between these arguments, and the continued push of individual/consumer responsibility and "vote with your wallet", the vast majority of people do not focus their attention on how technology platforms work, instead focusing their interest and energy in other areas, and will "vote with their wallet" for whatever is the simplest solution at the time.
First order effects are generally not considered when it's not something you care that much about. Second and third order effects are definitely not in mind.
No one said uncle Bob is old. Maybe uncle Bob is your hip cousin who still has no idea how these things work and zero interest to learn it. IMO, the biggest failure of tech discussions is to ignore that technology is there to solve a problem people have. It's not an end, it's a means to an end. People want to know that the video they see which says "filmed in xy" was really filmed in xy and not manipulated afterwards. And saying "yeah, you can just do this yourself, all you have to do is use gpg, make your own signature, distribute it, .." completely misses the point.
>People want to know that the video they see which says "filmed in xy" was really filmed in xy and not manipulated afterwards.
The only people who care enough to give up all privacy on the internet in exchange are deeply misunderstanding what normal people care about. Uncle bob would much rather the meme he made about weed not trace back to him or get to his employer with his fingerprint.
whether some video was filmed in xy doesn't matter unless you're in some sort of political crisis. Bob isn't. You're using a fictional bob to represent your own desires and present them as 'everyman issues'.
Bob doesn't care, and if he did he'd already use authentication. This is your problem, not bob's, but you're using a fake bob as some kind of 'personal shield' for an opinion leaning towards sacrificing freedom and privacy for very little in return.
What real bob wants most is probably less of you interfering in what he can see and read. He's not your kid and your intrusion on his freedom to read and speak are aggression against him.
You can't make uncle Bob any more media literate than he wants to become. So perhaps your energy might be better focused on being a good resource for helping his daughter/your niece become media literate. And then perhaps that may rub off on him via osmosis.
If he doesn't want to be. I don't bother. I also stop taking his incoherent ramblings any more seriously than the limited amount I already did. You can't reason people out of a position they didn't reason themselves into in the first place.
No, this thing is not necessary. That's an excuse for giving up privacy to a group of unaccountable corporations who exist to make money and are often abusive of their positions.
I don't have an answer per se, but if your answer is some sort of content blockchain so everything is traceable, then it doesn't matter that I don't have the answer. That's so repulsive as to not require an alternate solution. It's like giving yourself a lobotomy because you have a headache. It's the wrong answer full stop.
Not sure why this is downvoted, the questions are constructive for the thread. To add a question: why is it bad if it's an optional component? Nobody's forcing you to disclose that you shot a certain picture, unless I'm misunderstanding the concept here.
The endgame is to be able to have your fingerprint in every device you own, and it is not going to be optional. So that no matter if you uploaded a photo from your phone, an ipad, your PC - they will know it was you. It's going to be embedded at every level of the stack, from a file - right down to the silicon (if Intel's involvement is any indication). They want to know every piece of content you create and distribute, not so that they can trace it back to you - but also so that they can ultimate shut you down if you don't live up to their standard of "good citizen".
I can see in the future the prices for "pre-spyware" hardware going up, much like "pre-emissions" is a thing in the car enthusiast community and lately, interestingly enough, farmers (right to repair and all that). Computers are plenty fast already, without the bloat of "modern" software forcing you to upgrade... if we're going to go down that conspiracy rabbithole further, those hostile behaviours all align against the users.
Do you have any proof or even hint for your suggestions or is this just a doom 'n gloom mixed with "they are out to get you" paranoid trope of the week?
Do you think the mass deplatforming of "wrong thinkers" - which we have been seeing a shocking amount lately - is going to be harder, the same, or less difficult with something like the CCPA tracking literally everything you are doing?
When I was first on the internet, there were about 16 million people using it. Now there are allegedly around 5 billion. Roughly 260 million people got connected in the last 6 months or so. How many does it take to be a shocking mass?
Deep fake propaganda is not really a problem. I've spent the past two years working in the media forensics space. Deep fakes ended up being much less of a threat than anticipated. Anti deep fake models are quite proficient now too.
The big problem now is semantic forensics - taking media out of context and spinning a new narrative.
Oh this is dreadful. I moved to AMD this year. Do you know good alternative to After Effects on PC? If I could get that going I could stop my Adobe sub.
Unfortunately, the only viable alternative is Final Cut...which I think is vastly superior but is a much bigger investment (buying an Apple computer instead of just switching software).
Wait, this just looks like cryptographically signing pictures. What's so scary about that? You can do it today from your terminal if you feel like it.
I guess I agree that the 'trusted computing' stuff it seems like they're trying to do is a little scary, but the tech isn't really there yet, at least not on the desktop (look at the fiasco that is Intel SGX) and it's happening with or without whatever this CAI thing is.
I guess a world where your iPhone's camera sends signed frames to the processor's secure enclave which processes them and signs them with a key signed by Apple is... a little different from today? They do basically this for Face ID today.
> In order to bring its best single core performance to the desktop market, Intel had to redesign its 10nm product on 14nm, which combines the high throughput of the design with the high frequency of 14nm
Is that just because they couldn't do it at 10nm? I thought the whole point of die shrinking was to get higher frequency, lower heat etc?
My understanding is that they can't get things to work at 10nm at the high desktop frequencies they use at 14nm. I don't know if that's yield issues, instability, or what... probably a combination of those? My sense is that on laptop, the 10nm stuff is doing... okay.
It sure seems like the hope is to limp along on 14(+++++) for a while longer and leap straight to a kinks-worked-out 7?
The article drops some hints that the 10nm parts aren't stable at desktop power levels. Normally process shrink -> reduced power -> more power budget -> higher frequency. The power savings gives you room to push frequency (and power) back up. But it's not working out for 10nm. They're more efficient than 14nm, but they can't spend the energy they're saving on performance.
I'm just flabbergasted that Intel has Thunderbolt support integrated into it's mobile chips, but on desktop it's business as usual, left to motherboard vendors to implement Thunderbolt themselves with the normal slew of add-on chips.
Super frustrating that desktop users can't expect modern connectivity. It should be on the chip, like mobile. Notably, one can take any old TB4 cable, connect two modern Intel laptops, and get a 40Gbps networking connection between them. Alas, only a couple rare desktops will be able to participate & do this, because for whatever reason this on-chip capability that Intel has was omitted on desktop. I was so excited to be entering an era of enhanced connectivity!! Not yet, I guess.
Intel still did not react properly to their new situation of being objectively worse than AMD. They lowered the prices a bit, that's all. Still trying to segment the market with locked processors, even introducing more limits on ram overclocking, and still refusing to honor warranty when enabling XMP. That's not a company that wants to compete, so I wouldn't expect big moves with regards to connectivity.
Locked processors shouldn't be a thing anymore, ECC should be supported, motherboard chipset segmentation stopped, and it would be nice if the new processors wouldn't be set to 300W energy usage. With the right price Intel's processors would have a chance then even though they are slower. But like this - who would want to buy this? You are right to be flabbergasted, they should do things like integrated thunderbolt support to better the situation.
Nowadays they change their sockets at a much higher rate compared to AMD. Since you have to replace the motherboard to upgrade to a newer generation, the door is open to move from Intel to AMD.
You're probably right, but the Android market has it's marketplace pressures just like PC, and while I don't know the margins, I imagine Apple is making good margins on Macs.
It seems there IS a market for high end androids, but not for PCs, which really is an indictment of mainly Windows but also desktop Linux.
> really is an indictment of mainly Windows but also desktop Linux.
I think it is more about the market size. Only a small part of population needs or wants PCs at all, so the base for marketing the high end is limited. But almost everybody needs a phone, so the base for marketing of pricey phones is many times bigger.
It's not USB4 so it can't connect to other computers. And it's already legacy, with limitations (no sharing mtiple streams on the channel, so no mixed DisplayPort). It's not Thunderbolt so it can't connect to thunderbolt pcie devices.
It's about more than the throughput. It's about usability. TB4 and USB4 both have some great usability leaps over USB3.
You have to give whomever came up with 14nm process. Despite being super old it's still somewhat competitive. It's crazy Intel can't fix their process. They make more cash than amd, nvida etc. They make more cash than tsmc.
I would not be surprised if there was a lot of sunk cost fallacy on Intel's 10nm. They couldn't make it work effectively and it was so far behind that they bled out most of their process leadership. But rather than pivot and reset they've just kept hammering away at it. Of course the decision would not be trivial to make; the lead time for a new process node is significant.
Is it plausible that they wouldn’t be able to get 10nm to work, but would be able to make 7nm work? Or put differently, would abandoning 10nm be tantamount to abandoning the effort to develop a smaller process?
I am a very causal enthusiast but from my understanding there are many ways to skin a cat when it comes to process nodes hence with the company A 10nm vs company B 10nm are not equivalent.
Intel somewhere along the line chose the wrong set of solutions to achieve their 10nm process.
IMO these mistakes will not necessarily mean that some other node will be inherently disadvantaged other than the fact that they may now find themselves scrambling for tech lead. They will still have learned a lot from 10nm and hopefully should be able to feed that into future density improvements.
This far along into the crisis, it feels to me like it's clear that any technical issues are incidental. It seems like there are such dire internal cultural/management issues that will make it impossible to fix the technical ones in any acceptable timeframe.
Absolutely this. The technical issues exist because management made really bad decisions without having any sort of a backup plan.
Complicating it further, Intel made the same mistake over and over again-- they've been trying to get things back on track since roughly Sandy Bridge and have replaced their head designer every chip since that point when things didn't go well. Instead of fixing their design, they throw it out and start completely over. Again. and. Again.
It would require a miracle at the top to fix their decision process.
7 nm and 10 nm are quite different (7 nm relies more heavily on EUV IIRC), so the 10 nm delay doesn't have to affect 7 nm. However, 7nm is also facing significant delays.
I read somewhere that the crunch to get 14nm fixed burned a lot of goodwill among their top engineers, and when they saw delays looming for 10nm as well they looked for greener pastures rather than go through the 14nm crunch again. But yeah, Intel’s leadership should have been making better fallback plans for 10nm and beyond from the moment they knew Broadwell would have to be delayed, all the way back in 2013.
It's really a pipeline problem. You start the core design 2 or 3 years out, with the design rules you expect to be able to fab on. It's relatively easy to fab a 14nm design at 10nm if the 10nm fab shows up early, but it's hard and messy to take a 10nm design and fab it at 14nm if the 10nm isn't working so well.
But when you're "close" to getting your 10nm fab working, it's hard to start a new design for 14nm. If their 7nm had shown up ready to go, that would have saved them, but it looks troubled too. Maybe they'll be able to work out the kinks faster on 7nm.
Maybe they'll design something out of character for Intel that works for the realities of their 10nm process; from what I can tell, their 10nm process is mobile only because it doesn't clock well, they could turn that into a less negative by doing a design they normally couldn't do because of the need for high clockspeeds. It may look copycat of M1, but something that's a lot wider might be viable with a 3Ghz clock target, but isn't viable for 5Ghz that Intel needs to hit on desktop chips.
can the situation be thought of as "Intel has problems with porting all their 14nm tweaks to 10nm" and therefore is it reasonable to expect Intel on 10nm nodes to pick up all the similar node advantages over again?
I give the people working on 14nm a ton of credit for being able to squeeze out as much performance as they have. As long as big datacenter customers worry about AMD's supply chain that will probably keep them from switching en masse, though the big TSMC plant coming on line might go a long way towards alleviating those concerns.
I did not know this. This is interesting, because it gives Intel even less of an excuse for their insane power consumption and poor performance. In recent discussions, Ryzen's performance was commonly credited to the process, along the lines of "if you ported Intels core to 7nm it'd wreck everything else". It seems like more of a fair comparison now.
They are extracting every single dime of their billon dollars worth 14 nm process, like any business would.. time to move on to the next best thing though
I really wanted to build a Zen 3 / RTX 30-something to replace my Ivy Bridge / GTX 1060. After months of chasing the RTX, I finally ordered a prebuilt yesterday with a 10700K / RTX 3080. Now to see if it really is "in stock and ready to ship."
The only thing I feel like I'll be missing in the 10700K compared to the 11700K is AVX512. It's been derided for throttling, but it does look kind of amazing for certain workloads.
One thing I find curious about the Zen 2 and Zen 3 product lines is that the processor I really want is OEM only: first the 3900, now the 5900. I like the idea of twelve cores at an easier-to-cool 65W. I bought an OEM Athlon back in the day, but I don't see a lot of OEM processors for sale these days. Maybe just more fallout from the chip shortage.
You can just get a higher TDP processor and set limits so it's a low-TDP one. It will perform better. The most efficient CPUs go to the highest end parts; not the other way around.
Anyone have any idea when intel's 10nm will hit desktop? Are they making any progress at all? Are they just planning on skipping straight to 7 at this point?
This is half- to full- snark: When they outsource to TSMC.
At some point Intel will get their act together and only be one step behind rather than... christ, are they three steps behind? They just have way too much money.
But it does look like with the management purges, re-hires of old guard, and other changes it'll be a while before they right ship.
There's probably pride to overcome, since a lot of process choices probably are "stake my career" things and now those careers may be coming to an end.
The interesting thing may be that Intel now not only must close a gap in process technology, they may also need to change instruction set and make an ARM offering.
> At some point Intel will get their act together and only be one step behind rather than... christ, are they three steps behind?
They're not really three steps behind. At some point, node measurements became more useless as Intel and TSMC differ on what part of the thing they're measuring. Intel's 14 nm is roughly the same as TSMCs 10, and i10 ~= t7, i7 ~= t5. TSMC is shipping their 5nm chips, and intel is halfway shipping their 10nm chips, so that's somewhere between 1.5 and 2 steps behind.
Agreed on the process measurements, and so you're technically (the best kind!) right, and I remember the days when Intel was on "nm" parity with other fabs and their stuff was better at TDP and other things. But then...
With the code of processes and half-nodes / node+ tiers which easily cost as much as full node shrinks before, well, financially it may be three steps.
Plus the fact you don't chase what's in the marketplace if you're behind, you have to chase where the ball is rolling to.
Very theoretically, ADL-S is meant to be released in September. This seems a bit surprising, given that people won't typically by buying systems with RKL until April - a very tight window.
RKL might turn out to have been a stop-gap measure that was only semi-needed. A lot of the hyperventilating about how much trouble Intel is in here (which is, frankly, moderately justified) ignores the fact that the desktop segment just isn't that important. I mean, it's important to me. I've been waiting for a new desktop for a while (and, despite being ex-Intel, may go Team Red). But the money is in laptops and servers. Desktop is clearly not high in the queue.
It's at the point where I'll either go Team Red (and probably buy a TGL NUC for AVX-512 experimentation) or buy an 8-core TGL "desktop replacement laptop". The power consumption is ridiculous. A lot of my argument for wanting a desktop is not having laptop-style power management wobbles while trying to do fine-grain performance benchmarking, but what kind of wobbles am I going to have on a system like RKL? 291W? 104C? Yeesh.
Supposedly the issues are fixed with 10nm in 10++ or whatever it's called (SuperFin?). We're not going to learn about that until September or whenever ADL tips up.
ECC RAM is more expensive - from a very quick search looks like 16gb ddr4 2666mhz would be about 30% higher price for ECC.
I've used non-ECC RAM for my entire life so far and bit flipping has never caused an obvious error for me, so if I'm going to pay more for ECC RAM I'd need a good reason to do so.
I bought one a few months ago and I can confidently say that the performance is fantastic (the integrated Xe graphics plays all manner of games at full settings and framerate), however the drivers are terrible.
The particular laptop that I bought is the New Dell XPS 13, and it has an integrated 4K screen that randomly flickers, fails to start after sleep, dims and has a plethora of other issues related to power management bugs in the drivers that Dell have told me are a work-in-progress with the Xe. In the last month or so, there has been a new graphics driver or BIOS nearly every week, and insofar the issues are now nearly all resolved.
As for the CPU, the i7-1185G7 runs really hot on idle but thankfully it doesn't get much hotter under extreme load. It feels absurdly fast, but the battery life is quite lackluster as you'd expect due to the heat produced under idle.
Price is the biggest factor. You can get 10 cores with Intel for cheaper than 8 with AMD, plus with Intel you get integrated graphics as well which is really important right now considering the GPU shortage. Also, Intel allows virtual GPUs for virtualization in QEMU which AMD won't do.
The whole situation is a kind of surreal 180-degree turn of what things looked like 5+ years ago and a testament to how far AMD has come.
"Yeah go with Intel if you don't need top performance because the CPUs offer more[0], but slower cores, albeit with much higher power draw and a decent-enough integrated GPU at a cheaper price point."
Are you sure we're not talking about early 2010s era AMD APUs?
[0]: There are as many caveats here as there were 10 years ago for AMD.
If I wanted more than 8 cores, the 12-core ryzen 5900X is $549, can you really get an intel 10-core cpu for the same price, or within 50 bucks plus or minus, that performs better?
I'm seeing the i9-10900K for $467, but in terms of dollars/performance per core I think it's still behind the ryzen which costs $82 more.
The biggest issue you're going to run into, is that the 5900x may be $549 retail or whatever, but where are you going to buy one? I recently opted for a pre-built 4600G PC, and the 4700G is a great price point too... unfortunately AMD decided to really open these up to OEMs.
GPU is not a prerequisite unless you're into gaming or ML. For someone like me, a GeForce 8800GT is sufficient, but I'll take all the cores and RAM I can get.
Some form of GPU with modern display out is needed for some multi-monitor setups (mostly 4k), even if you don't do ML or gaming. Unfortunately there are also many motherboards out there that only support running one 4k monitor at 60 Hz even if the integrated graphics support more.
if you are going to use the GPU solely for the video (output) acceleration then you still have many options that can do that but (fortunately?) can't mine digital currency efficiently.
Don't know about that. Here in France, I've just checked Amazon for a cheap, relatively modern GPU and the offer has dried up (at least those sold by Amazon). I'm looking to run a GPU-passthrough virtual Windows on my linux PC without integrated graphics.
I bought a RX550 a year or so ago so my father could run a 4K screen. It cost around 100 euros at the time. This would be ideal for my use case, but now I can't find one. Are those even useful for mining?
On the nvidia side, all you can find is some GT 710. Not sure those can run 4K screens (didn't check). Even the GT 1030 is hard to find - a few months ago they were everywhere.
I wouldn't recommend buying a 5900X for gaming; you're going to be bottlenecked on your 3080 GTX even with many fewer CPU cores. 5900X is a workstation-class CPU.
I wouldn't look at the 10900k trying to "save a buck" on the high end. Instead the 10850k which is, for most purposes, just a 100 MHz slower 10900k can be found on sale for 350$. Higher availability, and prices that meet or are below recommended MSRP.
Oh wow, I didn't know HN still had knowledgeable commentators! (I thought it was a competition to say BLUE COMPANY BAD GREEN COMPANY GOOD in various ways).
So OP assuming I still want to buy an AMD because of its ECC support, (and can get my hands on a GPU), can you point me somewhere so that I can read about the last point you made. (EDIT: haven't researched but https://www.amd.com/en/processors/ryzen-with-graphics has integrated?)
Can you also list any other factors that I should weight in my decision, specifically against AMD?
I do wonder how good the secure enclave offerings of both platforms are. So far I am left believing that Intel's is actually mature and AMDs is not.
Regarding other factors to consider, I really don't have anything else. If ECC is important, then as you mentioned Intel doesn't offer ECC on these chipsets.
Additionally, AMD's advanced security features like memory guard is only available on the PRO processors, however there are versions of the 4650GE Pro and 4750GE Pro for a descent price point:
More advanced capabilities like SEV are only supported by EPYC from what I can tell, which is what various cloud providers are now using for confidential computing:
I'm just cracking a smile and almost laughing at the thought that 25 years ago I wanted an Intel Pentium but had to settle for a cheaper AMD 586 clone that ran hotter and performed worse. My, my, my how the tables have turned on Intel...
AMD needs to integrate basic gpu in its higher end CPUs. With the current gpu prices where even old low end models are selling for double it will make sense for anyone needing good desktop experience without need for a fast gpu otherwise AMD is missing a market segment.
There is in the Netherlands it seems. The only one available AMD one is a RX 580 for €669, the only available Nvidia one is a €109 GT 1030. The GT 710 cards seem like a huge downgrade in terms of display options compared to my laptop’s integrated graphics.
Like I said, in the Netherlands. I looked at a couple of the biggest stores and looked at what was available to be picked up or sent over right away. I did recheck and found one single store that sells a RX 550 for €95. Look (warning illegal cookie wall): https://tweakers.net/videokaarten/vergelijken/#filter:q1ZKSc...
I'd expect similar capabilities to the host GPU. In terms of how many virtual GPUs can you use with acceleration, unless you're trying to run graphics-intensive tasks on multiple VMs then you're mostly limited by your RAM. Here is more information for the Intel UHD 630:
... or the 5600X (probably equal perf but cheaper / lower power) or the 5900X (about the same price if you factor in power supply + cooling, still lower power).
...Which is weird, because there are other reviews that say the opposite based on a small handful of benchmarks. I'm not sure what conclusions to draw.
This is pathetic. At this point Intel should open up x86 and become like ARM as this is proven to work now. I wonder how badly Intel's mishap set humanity back by releasing essentially the same processors each year...
> The improvements Intel has made to its processor seem to do very little in our gaming tests, and in a lot of cases, we see performance regressions rather than improvements.
> The answer from our side of the fence is that Rocket Lake has some regressions in core-to-core performance and its memory latency profile.
I think at this point we're just waiting for Windows and Linux to be fully supported on the M1 or for ARM to come out with a system on a chip that's documented enough for Windows and Linux to make a port.
Considering how little fab capacity there is for new designs right now (and Apple has probably booked it all out for months/years to come), I can't see this situation changing much. AMD is the closest but they also have insane supply shortages right now.
Why would you wait for that when you can have a faster CPU in the form of the 5800X today? The M1 is a phenomenal laptop CPU, but it's still not faster than AMD's desktop offerings, either. At least not according to the specint results in the article.
The M1 is much faster on some single-threaded workloads than any other CPU. For example on browser benchmarks M1 is a whopping 60% faster than the 2nd place processor. Browser is an important workload for most people.
If you want to run MS Windows then obviously I can't recommend it.
Yah, the M1 is a single threaded beast, but its too power constrained (and apparently not as efficient in a number of cases) to keep up with AMD's less power constrained laptop cores in throughput tests.
I think this could be a fine battle with a strong apple lead at least until amd releases a 5nm design, but i'm afraid that apple, will be apple and keep doing thin/light/constrained devices, with maybe a super expensive mac pro that can't be justified by anyone who doesn't burn cash in their grill.
Windows on ARM is already working on M1 Mac through parallels. The big issue now is licensing and I assume some niceties in terms of official drivers/fine-tuning support.
According to recent videos, Windows in a VM on M1 Mac today has same or better performance than Windows on ARM running natively on Surface Pro X.
Licensing isn't an issue: a standard Windows 10 Pro license is processor-independent and includes a single-seat virtualization license. If Apple allowed it, you could even use the same license for virtualization and BootCamp like you can on amd64 macs.
I'm not sure ARM images are available, except for big OEMs. Also IIRC the boot is not restricted to Apple software (or at least can be unlocked), so I'm not sure that Apple is actually disallowing anything. Getting native drivers is another issue...
or for ARM to come out with a system on a chip that's documented enough for Windows and Linux to make a port
Linux (i.e. Android) already has been ported to countless ARM SoCs, the problem being that all of these SoCs are subtly different even if they're well-documented.
Around x64 cpus you have plenty of peripherals that are subtly different too. Arguably the situation is "worse" for ARM, but it is getting worse every year on the x64 front too, esp since MS decided to release extremely short lived OSes quite fast, without really any fixed HW support policies. In the end I've recently had built-in ethernet on a recent Intel not recognized by a few years old Linux distro, so at the end of the day it does not matter much if you can't use a computer because of an essential "peripheral" or because of an interrupt controller; and if you have recent enough software it will be supported in both cases.
You can buy a couple different Qualcomm-based Windows ARM laptops today, like the Surface Pro X. There's even a ~5 year old Asus (or maybe Acer) laptop using a Snapdragon processor that you can find for dirt cheap on ebay.
ARM is not the gatekeeper or reason why windows on ARM is lagging--the ARM ISA is fully documented and available to license holders (of which Microsoft is certainly one). Microsoft is the reason windows on ARM is a bit of a hot mess. They let the windows ARM platform languish for years and are scrambling like mad to catch up now (both in the Azure server space, and in the consumer space).
> for ARM to come out with a system on a chip that's documented enough for Windows and Linux to make a port.
glances at the pile of ARM boards on his desk
don't care about windows but... linux ? ... you know that 99% of android devices are linux and run arm, right ? and that debian has been available for arm, for, what, 2.5 decades ? what do you think raspberry pi's run ?
Sad, 10 years of nothing from intel but single digit uplifts and increasing power. I think apple sort of proves that there is plenty of single threaded perf uplift available with the right designs (aka one not designed for a server). Even AMD traditionally trailing is easily surpassing them now too, the AMD core that is easily matching or beating them isn't even close to the best desktop part they have (and rightfully so, a desktop with 8 cores... lets see phones with 8 cores have been a thing for how many years again?).
Couldn't happen to a nicer company, but for the health of the industry I hope they have something to keep up with the Arm hordes. What we need is a good decade of a 3+ way back and forth between an arm vendor or three and amd and intel to really push the industry again.
BTW: Just imagine what apple could do with a 300W power budget!
Yet they are still the king of single threaded performance (barely). I'm not so quick to rule out Intel, even if I have no love for them. While the M1 was extremely impressive from a power efficiency standpoint, I think the jury is far from out on whether Apple will be able to take those gains and scale them to serious computers.
Where are you getting that they’re the king of single thread? That was true (barely) last gen of AMD, and is not true anymore. AMD is better at single and multi core as well as power efficiency. From the article:
> Our results clearly show that Intel’s performance, while substantial, still trails its main competitor, AMD. In a core-for-core comparison, Intel is slightly slower and a lot more inefficient.
They certainly are not king of single threaded performance. Both Zen 3 and Apple's M1 comfortably best Intel in single core performance, using much less power to achieve that. Intel lost the single threaded crown in mid 2020.
I assure you my M1 Macbook Air is a serious computer. I know what you mean by serious, but honestly the only thing lacking from the M1 chip is connectivity. All three computers Apple released have the same limitation; two PCIe lanes for external connections.
Gaming performance regressions leave me a bit stunned. That's impressively bad, even for Intel. What are they going to do now, price this below the remaining Comet Lake stock and hope that Zen 3 has low availability until they have Alder Lake ready?
Among other things: intel has an internal fab that they do not want to let go. This makes them several generations behind on the process. Even copying AMD arch would not help.
263 comments
[ 3.3 ms ] story [ 218 ms ] threadEdit: Temperature peaks on some workloads up to 100c. Jeez
Nobody uses it because it wasn't available and, when available, wasn't much faster overall. Now it is.
Visicalc didn't make a big impact in mainframes.
Anything more is an advantage.
What else does that leave?
Most scientific and engineering computing applications use double precision floats.
I'm not sure, but we haven't seen it being widely deployed just yet.
AVX-512 is a pleasure from the programmer's perspective, much nicer than writing software for GPUs or for earlier SIMD instruction sets like AVX2/SSE
Versus a GPU, the AVX-512 machine model is just much simpler, and there are no drivers or proprietary compilers
Versus earlier SIMD instruction sets, AVX-512 provides all the instructions you want (permutations, floating point conversions, etc.) and they're all masked so edge cases are easy. AVX-512 Foundation is a really nice instruction set and it has a bright future
It might eventually get normalized but atm it’s a bit of a mess.
If you don't cut down the support table to eliminate irrelevant niche products, it's most of the way to being a Gray code.
There is also the big unknown of what kind of support AMD will bring to the table.
I think Intel is banking on people using OneAPI more than writing AVX-512 code using intrinsics or compiler magic.
At that point compatibility might be not as much of an issue if you ignore potentially drastic differences in performance.
Cooling this for real work will require a custom watercooling set up...and it isn't even as fast as a Ryzen 8 core CPU.
Truly a rocket. All it needs is a nozzle and some reaction mass. Maybe they should pivot?
"Intel exhaust outside". Do some computations and get thrust for free!
Still, I can't get my 5950X to ever come anywhere close to that, even running AVX2 Prime95 with max overclocking tops out at 190-ish IIRC.
Ice Lake (which Rocket Lake is a backport of) similarly doesn't downclock much: https://travisdowns.github.io/blog/2020/08/19/icl-avx512-fre...
The 100+C temperatures suggest it probably should downclock a bit, at least on 14nm.
PS My work desktop died on Wednesday afternoon during the SpaceX SN10 flight. I rushed to my local computer store to replace it. Out of the entire store, only 2 motherboards models with Intel's LGA1200 are in stock. All the other motherboards on the shelf were for AMD.
I picked the i5 10400f as it's under deep discount. I asked around and everyone agrees that Intel is competing only on price point right now.
"The new generation Rocket Lake is the combination of two different backported technologies. Intel took the Sunny Cove core from 10nm Ice Lake, and re-built it on 14nm, calling it now Cypress Cove. Intel also took the Xe graphics from 10nm Tiger Lake and re-built those on 14nm, but these are still called Xe graphics."
So, yes, it appears to be a backport.
https://semiaccurate.com/2020/10/29/intels-palpable-desperat...
Just filter out Charlie Demerjian's grudge against Intel.
Companies don’t backport because a core is designed for a process. It’s efficiency and architecture is based on the transistors that it will use from the design start, caches are sized based on the process too, and a lot of the architectural gains are due to the added transistor counts. If you backport a CPU you lose the efficiency of these new transistors so the energy use goes up and the clocks likely go down as well. Cores take up 2x the size, or at least a lot more than the older core did on the older process, so costs go up too. You can either cut out bits of the core and lose performance or eat it on area and therefor cost. Backporting a design not made for portability isn’t a lose/lose proposition, it is a lose/lose/lose/lose proposition. But Intel is desperate so…
This review makes it clear that unless Intel magically gets a process improvement they're entirely screwed.
This is precisely the same kind of design issue that hit with the latter end of the P4 era, where thermal limitations were so severe that they were crippling the chips. I'd have to do some digging, but I absolutely remember reading an interesting set of benchmarks that made it clear that you were 99% likely to thermally throttle those P4s even with proper thermal paste and an improved high-end heatsink/fan.
AMD will make gains for sure at this rate, but TSMC can't serve the whole market. Intel owns fabs, and these rocket lake chips will sell to the mainstream just fine.
I expect intel's fabs to be a big asset in these COVID times with sold out computers.
Pretty much every Rocket-lake score is faster than Skylake across the board.
I'd definitely buy a Rocket Lake over any old Skylake based off of these reviews. Rocket Lake seems a wee-bit inferior to Zen3, but AMD Zen3 is pretty much sold out, so if you need a computer, Intel is still there.
Except for gaming, where rocket-lake is across the board slower. Which is important since gaming was Intel's last holdout prior to the arrival of Zen3. So if you're gaming this is worse, and if you're not gaming why would you go with this over a 3900x at basically the same price? Even verse skylake if you're doing productivity stuff a 10920x seems likely to be consistently faster thanks to 50% more cores (and way way more pcie lanes if you're into that)
The only reason to get this seems to be if the unique combination of avx512, no ECC, and only up to 8 cores is somehow ideal for your use case.
Their fixed costs & massive R&D requirements have to be paid for by shrinking revenue: Projected 1st quarter 2021 revenue is down about 12% over the same quarter in 2020. This isn't COVID-related: AMD's projected revenue for the same period jumped 80% year over year.
Intel is far from okay. They are far from having a safe path due to production bottlenecks by competitors: TSMC's 7nm capacity-- which is what AMD uses-- is on track to increase 18% in just the next quarter alone. I don't know how sustainable that production growth is, but it doesn't have to last long when they're bringing new fabs online and Samsung's fabs may be taking on some of Apple's M1 production, further freeing up TSMC capacity.
I don't think I'm saying anything particularly controversial or surprising or new when I say that Intel probably has about 2 years to figure out their own fab problems before they're well out of things for the next decade or so, having to claw their way back the way AMD has done.
It's out of stock everywhere..
Of course, Intel's CPUs have less PCIE lanes, no ECC support, lessor upgrade path, generates more heat, etc.
Finally my old desktop was Intel and I am working on a lot of kernel/VM stuff. I don't want to switch architecture halfway. I did that before and it was painful.
What? Aside from edge-cases resulting from slightly divergent microcode implementations you should not be dealing with that. They are both x86-64. (Forgive me if I am missing an elephant in the room.)
> Of course, Intel's CPUs have [...] lessor upgrade path
One big + for AMD is their socket system, and keeping it the same for multiple generations via a roadmap so people can plan. Contrast that with intel, who comes up with a new socket every generation forcing you to buy a new motherboard and likely new ram if you want to upgrade. AMD's Socket AM4 was released in 2016 and works with Zen+, Zen 2, Zen 3 and Excavator microarchitectures. If you upgraded every 2 years, that's 3 motherboards and memory sets you would have had to purchase if you went with Intel vs a single motherboard/memory and upgrading only the cpu with AMD. That's quite a premium.
What happened? I’ve never seen a computer die during use.
>My work desktop died [..]
Are these related, perhaps? I found the juxtaposition quite humorous.
They need their new process node and whatever new architecture they have in the pipe to land -soon-.
Looks like my 9900k might be the last Intel processor I buy for a long time unless things drastically change.
Not soon but yesterday
Around 2012 I bought a 4-core Core i5 and after some basic upgrades (max RAM to 32GB and a good SSD) it had more-than-acceptable performance for my work until mid-2018. Even when I dabbled with Machine Learning I borrowed a 1070 from my office and was plenty to run my tasks.
Around end of 2018 I started working more in different offices, so I needed a more powerful laptop. No strong offering from AMD yet, so I ended up with a System76 core i7, 6-core 32GB RAM.
2020 came and with it I got back into working at home. So I got excited about upgrading my workstation. I bought all the components to get a Ryzen 5, a 3700X I think. What a shitshow: had to go through three different pairs of RAM to find one that was actually compatible with the motherboard. Then I had to find a way to upgrade the BIOS, and I could only do it by getting an Athlon CPU. BIOS upgraded, then the original 3700 died. I ended up turning the whole system into a very expensive media center just with the Athlon and continued working with the laptop.
A couple of weeks ago my laptop's battery puffed up, so I decided to go look again into a nice desktop. An 12-core Ryzen 5900X is coming out at 900€ and a basic motherboard is 300€. 10-core Intel i9-10900 is 300€ and a motherboard for it is 180€. Guess which one I ended up taking and that I managed to get running without any issue whatsoever?
Here is to hope that in 6-8 years from now I finally manage to make the switch to AMD.
(Edit: why the downvotes, HN? It's an honest account of my experience.)
The "odd" thing is that this Intel system I put together ended up being the very same brand. I am regretting it now a bit because I realized that the board can only do 4k@30Hz through the HDMI port and the i9 can do 4k@60, but given that my laptop had a similar limitation and I got used to run with my monitor at 1440p, I think I will hold on to it until I feel like it is time to get a dedicated GPU.
A 'basic' motherboard two weeks ago was the same as it is today, a low-end B550, and can be purchased for under 100€.
The 10900 trades blows with the 3900x, particularly for multi-threaded, workstation-type workloads where the AMD part pulls ahead with its two extra cores. And the 3900x can be had around the same price as a 10900.
So your "honest account" comes off as a little disingenuous, which may explain the downvotes.
(Yes, the high-end Zen 3 parts are pricey and have very limited availability right now, but that is afflicting the entire industry - see also consoles, graphics cards and many other things. With their own fabs, intel are uniquely placed to supply right now, but their CPUs are more competing with the last gen of AMD's if you're talking about a workstation)
Still, point taken. If I didn't get burned by the whole ordeal with the 3700x I think I would've taken a second look at the AMD 3000.
The threadripper boards are starting at $400 for the current generations. I regret spending what I did on the first generation boards, as they dropped support for them on the third generation. I'm waiting to see a zen3 series threadripper before I'll replace my old 1950x. The even older 3930k i7 was able to keep up on single threaded processes... which surprised me. Funny it really took 10 years before I got that wow factor again. The number of cores that I'd consider 'normal' has certainly gone up. Think the 3930k was around $600 in 2011, the 1950x around $800 in 2017, and the 5950x at $800, so that price point has not changed much. The first two were picked up at launch date, the last was a paper launch that took till January to find.
Relatedly, I am surprised by the commenters who believe that the trends in Intel pricing were somehow involuntary. The outperformance of AMD CPUs was clear even prior to the pandemic, and I recall talk at that time about Intel using price pressure to resist AMD's expansion of market share and gross margins. Now that silicon supply is getting tight, market share is all the more relevant because whatever demand AMD can't meet gets caught by Intel. It's not all about benchmarks (although IMO Intel is hanging in there with Rocket Lake). You actually have to get a customer to buy your product in order to generate revenue.
Given the choice, Intel would rather resist AMD's expansion of market share and gross margins without lowering their prices.
In this case Intel is undercutting AMD and setting a low price that forces AMD to keep its own pricing lower. This is disadvantageous for AMD because its ability to satisfy spikes in demand is capped; additional orders for unexpected demand have to flow through the TSMC pipeline, which is also used by many other companies. In contrast, Intel can meet demand because it owns its production processes; Intel churns out more product and absorbs the unexpected demand. Per-unit margin matters less for Intel because its share of the unexpected/marginal demand is high.
Your honest account seems more like a series of errors you've invited on yourself. Your motherboard prices are insane. And this litany of spontaneously breaking components make you sound like you're just yeeting the stuff around. Maybe that's all unfair, but it's how it comes across.
Now you're comparing the prices of the i9-10900 to the 5900X, when the 5800X's performance more than matches it. The 5800X is about 10% cheaper than the 10900 here.
If you don't understand the platform and chipsets, use a picker service or even buy a bundle. There are lots of component sellers flogging matched components in tested configurations. You won't get top-spec motherboards but there's no reason to spunk €300 on a desktop motherboard.
Like I actually did, I have been doing since I can't even remember to build previous workstations, NAS, the dappnode under my desk, and my home server that runs the services I self-host?
Looking around now, it seems that MSI X570 is available at a lower price, but it wasn't when I first checked.So, yeah, bad timing. But this patronizing is unwarranted.
> The 5800X is about 10% cheaper than the 10900 here.
In Germany it's about 15% more expensive, and my calculation was based on price-per-core than absolute performance, not to mention that the 10900 has a 65W TDP vs 105W for the Ryzens.
You're technically supposed to buy memory from the motherboard's QVL for exactly that reason, but I don't know anyone who does that anymore.
TDP is lie these days .The 65W intel chips will use over 200W in most motherboards.[1] The 10 core you're using will use even more. The 5900X would use 130W max.
Watch your temps, your passive thermal solution may not be up to the task and your CPUs may be throttling or dying prematurely.
[1] https://images.anandtech.com/graphs/graph16343/120661.png
I was indeed weary of keeping only the fanless cooler, so I ended up installing the stock one. Coupled with the fact that it is a big case and it has a lot of fans, I assume that would be enough.
Anyway, what you are saying about the misleading TDP is interesting. Perhaps I will get a 3900X or a 5800X system and if it works out I will return the Intel.
Basically a large group of companies are forming a group to create an agreed upon fingerprinting system for 'content' (read: everything) to destroy privacy in the name of 'fighting disinformation'.
Is there a privacy-respecting solution that you prefer?
Don't have a good enough social credit because you shared an opinion somebody doesn't like one too many times? Intel and Microsoft will just remotely brick all your devices.
> Is there a privacy-respecting solution that you prefer?
Yeah, it's called being media literate, using my brain and being vigilant. I don't need Pat Gelsinger, Satya Nadella, Sundar Pichai, Shantanu Narayen, Tim Cook, Mark Zuckerberg, Jack Dorsey or any other power hungry jackass telling me what to think.
If it's all opt-in (and otherwise anonymous), doesn't that solve the privacy issue?
I'm definitely not advocating for the OP solution, which is why I asked if anyone knew of alternatives.
It starts off with opt in, then they do an update to now opt out, then then do an update to make it harder to opt-out and next thing we know this is something everyone has to live with.
But the main problem is the people most vulnerable to fake media are already bamboozled by bad Photoshop jobs. I don't see how any signing scheme will help.
Edit: don't get me wrong, _not_ doing that for children is evil. But, the results thus far do speak for themselves... And uncle Bob will not take being sent back to middle school all that well.
You have this idea that society is a monolith you can keep people in and out of based on the results of an algorthm, and that's only true for the most authoritarian and oppressive regimes.
That's the world you're trying to create. Social Credit scores as maintained by facebook, etc.
It's truly repulsive.
Almost everyone get's swimming lessons and tests. And yet, and this might really blow your mind, you can still actually go swimming without actually having had passed the test. No SWAT team will barge down your door and take you away if you go swimming.
Then whats the purpose of the test? Tests are scores used to determine something. You're talking about forcing people to take a test to do what, exactly? Or is it just to indoctrinate them with whatever the test-maker (because it will change every election) comes up with? A thing you must study for but with no reprecussions? Ya right.
Once you create a certification for it, it will become a private passport for access to goods or services, or it will be meaningless. In both cases its either a waste or evil.
I'm not sure why you keep going down this dark path which I didn't even touch on remotely. You just completely imagine it.
I have definitely fallen victim to that one myself, and have pushed that argument when considering Apple's walled garden in iOS, and how much that has assisted me personally in not having to provide tech support for members of my family.
Unfortunately, between these arguments, and the continued push of individual/consumer responsibility and "vote with your wallet", the vast majority of people do not focus their attention on how technology platforms work, instead focusing their interest and energy in other areas, and will "vote with their wallet" for whatever is the simplest solution at the time.
First order effects are generally not considered when it's not something you care that much about. Second and third order effects are definitely not in mind.
The only people who care enough to give up all privacy on the internet in exchange are deeply misunderstanding what normal people care about. Uncle bob would much rather the meme he made about weed not trace back to him or get to his employer with his fingerprint.
whether some video was filmed in xy doesn't matter unless you're in some sort of political crisis. Bob isn't. You're using a fictional bob to represent your own desires and present them as 'everyman issues'.
Bob doesn't care, and if he did he'd already use authentication. This is your problem, not bob's, but you're using a fake bob as some kind of 'personal shield' for an opinion leaning towards sacrificing freedom and privacy for very little in return.
What real bob wants most is probably less of you interfering in what he can see and read. He's not your kid and your intrusion on his freedom to read and speak are aggression against him.
I don't have an answer per se, but if your answer is some sort of content blockchain so everything is traceable, then it doesn't matter that I don't have the answer. That's so repulsive as to not require an alternate solution. It's like giving yourself a lobotomy because you have a headache. It's the wrong answer full stop.
The possibility of this being opt-in is so unlikely it probably not worth discussing.
They also fall for non-deepfaked media as well, and spearphishing, and so on. The greatest threat vector has always been the wetware.
The big problem now is semantic forensics - taking media out of context and spinning a new narrative.
https://www.theverge.com/2019/3/5/18251736/deepfake-propagan...
I guess I agree that the 'trusted computing' stuff it seems like they're trying to do is a little scary, but the tech isn't really there yet, at least not on the desktop (look at the fiasco that is Intel SGX) and it's happening with or without whatever this CAI thing is.
I guess a world where your iPhone's camera sends signed frames to the processor's secure enclave which processes them and signs them with a key signed by Apple is... a little different from today? They do basically this for Face ID today.
https://contentauthenticity.org/our-members
I agree that the project sounds like garbage.
The problem comes when it's built into the hardware so you are forced to sign all documents.
Is that just because they couldn't do it at 10nm? I thought the whole point of die shrinking was to get higher frequency, lower heat etc?
It sure seems like the hope is to limp along on 14(+++++) for a while longer and leap straight to a kinks-worked-out 7?
Super frustrating that desktop users can't expect modern connectivity. It should be on the chip, like mobile. Notably, one can take any old TB4 cable, connect two modern Intel laptops, and get a 40Gbps networking connection between them. Alas, only a couple rare desktops will be able to participate & do this, because for whatever reason this on-chip capability that Intel has was omitted on desktop. I was so excited to be entering an era of enhanced connectivity!! Not yet, I guess.
Locked processors shouldn't be a thing anymore, ECC should be supported, motherboard chipset segmentation stopped, and it would be nice if the new processors wouldn't be set to 300W energy usage. With the right price Intel's processors would have a chance then even though they are slower. But like this - who would want to buy this? You are right to be flabbergasted, they should do things like integrated thunderbolt support to better the situation.
How is it that mobile phones get technology support two years before laptops and desktops?
It seems there IS a market for high end androids, but not for PCs, which really is an indictment of mainly Windows but also desktop Linux.
I think it is more about the market size. Only a small part of population needs or wants PCs at all, so the base for marketing the high end is limited. But almost everybody needs a phone, so the base for marketing of pricey phones is many times bigger.
It's about more than the throughput. It's about usability. TB4 and USB4 both have some great usability leaps over USB3.
"will be outsold" has a different meaning, (which is likely also true for this processor, but not what you meant)
Is it plausible that they wouldn’t be able to get 10nm to work, but would be able to make 7nm work? Or put differently, would abandoning 10nm be tantamount to abandoning the effort to develop a smaller process?
Intel somewhere along the line chose the wrong set of solutions to achieve their 10nm process.
IMO these mistakes will not necessarily mean that some other node will be inherently disadvantaged other than the fact that they may now find themselves scrambling for tech lead. They will still have learned a lot from 10nm and hopefully should be able to feed that into future density improvements.
Complicating it further, Intel made the same mistake over and over again-- they've been trying to get things back on track since roughly Sandy Bridge and have replaced their head designer every chip since that point when things didn't go well. Instead of fixing their design, they throw it out and start completely over. Again. and. Again.
It would require a miracle at the top to fix their decision process.
But when you're "close" to getting your 10nm fab working, it's hard to start a new design for 14nm. If their 7nm had shown up ready to go, that would have saved them, but it looks troubled too. Maybe they'll be able to work out the kinks faster on 7nm.
Maybe they'll design something out of character for Intel that works for the realities of their 10nm process; from what I can tell, their 10nm process is mobile only because it doesn't clock well, they could turn that into a less negative by doing a design they normally couldn't do because of the need for high clockspeeds. It may look copycat of M1, but something that's a lot wider might be viable with a 3Ghz clock target, but isn't viable for 5Ghz that Intel needs to hit on desktop chips.
https://fuse.wikichip.org/wp-content/uploads/2018/02/iedm-20...
That being said they really need to step up their game or else they'll be bleeding market share for years to come.
Given the 7nm shortage at TSMC, Intel might still beat it in price / availability.
The only thing I feel like I'll be missing in the 10700K compared to the 11700K is AVX512. It's been derided for throttling, but it does look kind of amazing for certain workloads.
One thing I find curious about the Zen 2 and Zen 3 product lines is that the processor I really want is OEM only: first the 3900, now the 5900. I like the idea of twelve cores at an easier-to-cool 65W. I bought an OEM Athlon back in the day, but I don't see a lot of OEM processors for sale these days. Maybe just more fallout from the chip shortage.
Realistically now probably mid 2022 we’ll see something, but probably not a full lineup until 2023.
At some point Intel will get their act together and only be one step behind rather than... christ, are they three steps behind? They just have way too much money.
But it does look like with the management purges, re-hires of old guard, and other changes it'll be a while before they right ship.
There's probably pride to overcome, since a lot of process choices probably are "stake my career" things and now those careers may be coming to an end.
The interesting thing may be that Intel now not only must close a gap in process technology, they may also need to change instruction set and make an ARM offering.
They're not really three steps behind. At some point, node measurements became more useless as Intel and TSMC differ on what part of the thing they're measuring. Intel's 14 nm is roughly the same as TSMCs 10, and i10 ~= t7, i7 ~= t5. TSMC is shipping their 5nm chips, and intel is halfway shipping their 10nm chips, so that's somewhere between 1.5 and 2 steps behind.
With the code of processes and half-nodes / node+ tiers which easily cost as much as full node shrinks before, well, financially it may be three steps.
Plus the fact you don't chase what's in the marketplace if you're behind, you have to chase where the ball is rolling to.
RKL might turn out to have been a stop-gap measure that was only semi-needed. A lot of the hyperventilating about how much trouble Intel is in here (which is, frankly, moderately justified) ignores the fact that the desktop segment just isn't that important. I mean, it's important to me. I've been waiting for a new desktop for a while (and, despite being ex-Intel, may go Team Red). But the money is in laptops and servers. Desktop is clearly not high in the queue.
It's at the point where I'll either go Team Red (and probably buy a TGL NUC for AVX-512 experimentation) or buy an 8-core TGL "desktop replacement laptop". The power consumption is ridiculous. A lot of my argument for wanting a desktop is not having laptop-style power management wobbles while trying to do fine-grain performance benchmarking, but what kind of wobbles am I going to have on a system like RKL? 291W? 104C? Yeesh.
Supposedly the issues are fixed with 10nm in 10++ or whatever it's called (SuperFin?). We're not going to learn about that until September or whenever ADL tips up.
I'm running DDR4-3200 CL22 ECC overclocked to 3666 CL16 on an ASRock board with a 5800X.
Which ASRock board did you go with?
I've used non-ECC RAM for my entire life so far and bit flipping has never caused an obvious error for me, so if I'm going to pay more for ECC RAM I'd need a good reason to do so.
They are demonstrably broken though, and I agree it should be ECC everywhere
The particular laptop that I bought is the New Dell XPS 13, and it has an integrated 4K screen that randomly flickers, fails to start after sleep, dims and has a plethora of other issues related to power management bugs in the drivers that Dell have told me are a work-in-progress with the Xe. In the last month or so, there has been a new graphics driver or BIOS nearly every week, and insofar the issues are now nearly all resolved.
As for the CPU, the i7-1185G7 runs really hot on idle but thankfully it doesn't get much hotter under extreme load. It feels absurdly fast, but the battery life is quite lackluster as you'd expect due to the heat produced under idle.
"Yeah go with Intel if you don't need top performance because the CPUs offer more[0], but slower cores, albeit with much higher power draw and a decent-enough integrated GPU at a cheaper price point."
Are you sure we're not talking about early 2010s era AMD APUs?
[0]: There are as many caveats here as there were 10 years ago for AMD.
I'm seeing the i9-10900K for $467, but in terms of dollars/performance per core I think it's still behind the ryzen which costs $82 more.
https://www.techspot.com/review/2132-amd-ryzen-5900x/
I bought a RX550 a year or so ago so my father could run a 4K screen. It cost around 100 euros at the time. This would be ideal for my use case, but now I can't find one. Are those even useful for mining?
On the nvidia side, all you can find is some GT 710. Not sure those can run 4K screens (didn't check). Even the GT 1030 is hard to find - a few months ago they were everywhere.
The 8-core 5800X is available at MSRP, though (at least, I've seen Newegg stock it).
So OP assuming I still want to buy an AMD because of its ECC support, (and can get my hands on a GPU), can you point me somewhere so that I can read about the last point you made. (EDIT: haven't researched but https://www.amd.com/en/processors/ryzen-with-graphics has integrated?)
Can you also list any other factors that I should weight in my decision, specifically against AMD?
I do wonder how good the secure enclave offerings of both platforms are. So far I am left believing that Intel's is actually mature and AMDs is not.
Regarding the last point, you can learn more about it here:
https://wiki.archlinux.org/index.php/Intel_GVT-g
Or, on their official GitHub Wiki:
https://github.com/intel/gvt-linux/wiki/GVTg_Setup_Guide
AMD's similar MxGPU is only available on Radeon Pro and their Workstation graphics:
https://www.amd.com/en/graphics/workstation-virtual-graphics
Regarding other factors to consider, I really don't have anything else. If ECC is important, then as you mentioned Intel doesn't offer ECC on these chipsets.
Additionally, AMD's advanced security features like memory guard is only available on the PRO processors, however there are versions of the 4650GE Pro and 4750GE Pro for a descent price point:
https://www.amd.com/en/technologies/pro-security
More advanced capabilities like SEV are only supported by EPYC from what I can tell, which is what various cloud providers are now using for confidential computing:
https://www.amd.com/en/processors/amd-secure-encrypted-virtu...
Intel offers SGX on most of their consumer chips, but it has been plagued with issues and exploits in the past:
https://en.wikipedia.org/wiki/Software_Guard_Extensions
Intel GVT-g is not supported on Gen11 and Gen12 gen Intel Graphics. It's only supported on Gen9 and earlier.
About Intel discrete graphics:
https://github.com/intel/gvt-linux/issues/160
> Internally we have some prototype but there is no formal plan to support GVT-g on discrete GPU like DG1
and about integrated later ones:
https://github.com/intel/gvt-linux/issues/126
> until now there is no plan on it.
Hopefully that clarifies it a bit...
As for issues and exploits, I would wager its the same reason as Windows having more exploits than Linux...
Thanks again! I wonder if you use lobste.rs? (Not much better than here but less population means less noise...)
https://geizhals.de/?cat=gra16_512&v=l&hloc=at&hloc=de&hloc=...
What is the level of capabilities we can expect from this? How many simultaneous VMs can use the Intel GPU?
https://en.wikichip.org/wiki/intel/uhd_graphics/630
https://www.techradar.com/news/intel-core-i7-11700k-may-outp...
> The answer from our side of the fence is that Rocket Lake has some regressions in core-to-core performance and its memory latency profile.
Considering how little fab capacity there is for new designs right now (and Apple has probably booked it all out for months/years to come), I can't see this situation changing much. AMD is the closest but they also have insane supply shortages right now.
If you want to run MS Windows then obviously I can't recommend it.
I think this could be a fine battle with a strong apple lead at least until amd releases a 5nm design, but i'm afraid that apple, will be apple and keep doing thin/light/constrained devices, with maybe a super expensive mac pro that can't be justified by anyone who doesn't burn cash in their grill.
According to recent videos, Windows in a VM on M1 Mac today has same or better performance than Windows on ARM running natively on Surface Pro X.
https://www.macrumors.com/2020/12/22/m1-mac-windows-parallel...
https://twitter.com/imbushuo/status/1332484912549687297?ref_...
If someone wants to try it on an m1 Mac, I'd be interested where it fails.
Linux (i.e. Android) already has been ported to countless ARM SoCs, the problem being that all of these SoCs are subtly different even if they're well-documented.
ARM is not the gatekeeper or reason why windows on ARM is lagging--the ARM ISA is fully documented and available to license holders (of which Microsoft is certainly one). Microsoft is the reason windows on ARM is a bit of a hot mess. They let the windows ARM platform languish for years and are scrambling like mad to catch up now (both in the Azure server space, and in the consumer space).
glances at the pile of ARM boards on his desk
don't care about windows but... linux ? ... you know that 99% of android devices are linux and run arm, right ? and that debian has been available for arm, for, what, 2.5 decades ? what do you think raspberry pi's run ?
Couldn't happen to a nicer company, but for the health of the industry I hope they have something to keep up with the Arm hordes. What we need is a good decade of a 3+ way back and forth between an arm vendor or three and amd and intel to really push the industry again.
BTW: Just imagine what apple could do with a 300W power budget!
We'll likely see by the end of 2022 (probably with less than 300w):
https://www.theverge.com/2020/12/7/22158264/apple-32-core-pr...
> Our results clearly show that Intel’s performance, while substantial, still trails its main competitor, AMD. In a core-for-core comparison, Intel is slightly slower and a lot more inefficient.