AnandTech reviews AMD’s latest and greatest.
AMD has scored wins across almost all of our benchmark suite. In anything embarrassingly parallel it rules the roost by a large margin (except for our one AVX-512 benchmark). Single threaded performance trails the high-frequency mainstream parts, but it is still very close. Even in memory sensitive workloads, an issue for the previous generation Threadripper parts, the new chiplet design has pushed performance to the next level. These new Threadripper processors win on core count, on high IPC, on high frequency, and on fast memory.
[…]If you had told me three years ago that AMD were going to be ruling the roost in the HEDT market with high-performance 32-core processors on a leading-edge manufacturing node, I would have told you to lay off the heavy stuff. But here we are, and AMD isn’t done yet, teasing a 64-core version for next year. This is a crazy time we live in, and I’m glad to be a part of it.
I need one of these for translating, posting to OSAlert, and playing a few non-demanding games, right?
I am a little bit tempted, but the Ryzen 3900X already made my full clean compile times go down to 7 minutes. Going up to 24 cores might reduce it to 3 or 4 minutes. But even infinite cores couldn’t get it below 2 minutes since single threaded link time starts to dominate at that point. So probably not worth it.
Are you still using the BFD linker? Gold or LLD can be multhreaded, especially LLD.
Happy to see this kit benchmarked, finally!
For highly parallel loads, AMD scored big. Not only are these chips outperforming intel, but at a significant cost savings too. This is great news for anyone in the market for highly a parallel CPU. Until intel is able to deliver on a smaller die size, it’s chips are at a technological disadvantage. For data center and build servers, AMD looks like the best choice today.
For single threaded/low parallelism tasks AMD’s results are kind of disappointing though. I was hoping they would be able to beat or at least match intel given that intel is still on 14 nm.
In the consumer space with more emphasis on single threaded tasks, it still looks like intel’s chips from last year are not only faster, but also cheaper than AMD’s ryzen processors. So all and all I’d say the way things have been evolving, these AMD ryzen chips are probably headed towards the datacenter where they will be appreciated. AMD needs a killer application in the consumer space for such high core counts to become useful for normal consumers.
Alfman,
Which tasks in consumer space are significantly faster on an Intel CPU in such an amount that they warrant the extra cost of such a CPU over an AMD Ryzen? Even for gaming on a realistic level (i.e. high or ultra-high settings on QHD or UHD, not low FHD) the top-end consumer Intel CPUs aren’t significantly better overall than the newest Ryzen CPUs. Consider me curious.
Gargyle,
Which Ryzen chips are you talking about? The ones in the article are quite a bit more expensive than the intel’s 8 core consumer chips, which would be better for most games given that they tend to be sequential. So although I can concede you don’t need the fastest cpus to have a good gaming experience, it’s still hard to say that these particular ryzen cpus make much sense for gaming given that they’re not only more expensive, but they’ll perform worse with non-heavily multitasked workloads. Of course some people may benefit from massive numbers of cores/threads AMD supports, but that’s generally not the case with most games.
The problem here is that you are comparing AMD’s HEDT products against Intel’s consumer grade CPU.
If you look at the best gaming CPUs for holidays 2019 on anandtech (cf. https://www.anandtech.com/show/9793/best-cpus) you’ll see that all budget categories under $500 are dominated by AMD Ryzen.
Gargyle,
Yes, that’s a fair point, however my point was that these CPUs aren’t for normal consumers. I’ll reiterate that “AMD needs a killer application in the consumer space for such high core counts to become useful for normal consumers.”
I agree that for highly parallel applications, AMD’s processors have an impressive lead this for this generation.
Yeah, budget-wise AMD has always been the cheaper option. Intel is able to charge more due to it’s market dominance. Competition is extremely important to keep prices in check. Just looking at max performance though, intel still dominates at executing sequential software.
Personally, for my last build, I did consider CPUs with large core counts, but higher single threaded performance was more important to me than more cores with less single threaded performance. I could have gone AMD if I didn’t care as much about sequential performance. Here’s the thing, more cores would be great for my servers, but when I look at the CPU monitor on my desktop computers it’s usually only a couple cores that are maxed out at 100% and the remaining cores are under utilized anyways. Trading off sequential performance for more cores would be a net negative. 8 cores is sufficient for me given that the bigger bottleneck tends to be single threaded performance in day to day usage.
Of course, that’s just me, naturally people will have different workloads and budgets.
As an aside…if I could get a 4 core processor with 50% more single threaded performance, that would be very compelling even though I’d have half as many cores. This obviously won’t happen because increasing sequential performance is hard. This is the opposite of where the industry’s going because it’s physically easier to add more cores than it is to increase their speed.
I think it’s an open question whether desktop software as a whole will ever take advantage of such high core counts or not; so far developers with computationally intensive workloads are leaning towards GPU technology, which scales better than CPUs. This is why I keep going back to the idea that the best market for these high core CPUs is in data centers. It will be interesting to see what happens!
Ah so your gripe is a general one, not one specifically targetting AMD, right?
I agree with that, pushing multithreaded hardware down our throats while software much rather has a very quick singlethreaded CPU.
I guess we can’t have it all, right?
Gargyle,
I think physics will ultimately have to determine the future of computing, But the tech industry has invested so much time, resources, and money into what it is today that they’re rather slow to embrace changes. In past decades it was pretty easy just to point towards moores law and simply upgrade the hardware, but now that those easy gains are mostly behind us, we may finally be forced to adapt to more scalable software paradigms.
Or, there’s another plausible alternative that saddens me a bit, but it already seems to be happening in some respects: software evolution just stalls out and we may end up with a long lull in major innovation. Rather than developing new products, incumbents will focus on software as a service models where they stay afloat by renting out software that doesn’t really change rather than investing in new software. The fact that so many leading tech companies are chasing SAAS models leads me to believe that internally (not publicly) they’re betting against real innovation and finding ways to keep selling the products they already have. So it’s within the realm of possibility that this is what the next hundred years has in store for us: rental models that are technically sustainable but not very exciting for consumers.
Can we get some Chuck Norris-esque memes for Jim Keller? He built the “Zen” architecture the current success of AMD is based on, as well as some important architectures of the past. If Bruce Schneier gets to have Chuck Norris-esque memes, why not Mr Keller?
That cool and all, but I don’t see workstation or gaming laptops with AMD CPUs and GPUs, I can buy some workstation laptop with 8 core Intel’s CPU, but after several ZEN arch revisions we had only 1 laptop model with non ultra-low voltage processor, and that was normal desktop processor. For some reason AMD totally ignore DTR laptop market.
BlackV,
At 280W TDP, that laptop would be a furnace. Typical high end laptop batteries might be around 90WH, unless you stayed plugged in, the battery life would be crap on any workload that benefited from so many cores. I’d still love to see it get built though just for kicks
It’s more practical for heavy processing to be delegated to a conventional stationary computer and the data streamed to a portable thin client. I’ll grant you this isn’t as fun though, haha
That’s not a parallel computer. ,…. this is a parallel computer – https://www.cerebras.net/introducing-the-cerebras-cs-1-the-industrys-fastest-artificial-intelligence-computer/
lapx432,
Yeah, I predict it will become more and more evident over time that FPGA/GPGPU/vector processors will eventually win out for parallel computation. Scaling CPUs is reaching practical limits due to limited applications, difficulties of programming for NUMA architectures and whatnot. However scalable vector processing has only just begun!
I definitely /want/ one for similar levels of work – transpiling JS, uploading and downloading stuff, occasionally playing 5 or 20 year old games, etc.
“I need one of these for translating, posting to OSAlert, and playing a few non-demanding games, right?”
Soon you will need because the next version of your favorite operating system will be based on the latest web framework and need all that power just to render the cursor.