Which projects use AVX-512?

[zombie67]

Which projects use AVX-512? I think it may be only one or two projects, but I can't recall any details.

I could swear we talked about this in the past, but I can't find anything using search. I tried searching on both "AVX-512" and "AVX 512". No results at all. Maybe search broke?

[scole of TSBT]

The new Acoustics opti apps have an AVX512 version https://github.com/sirzooro/Acoustics-at-home/releases/tag/opti1.0

[zombie67]

The new Acoustics opti apps have an AVX512 version https://github.com/sirzooro/Acoustics-at-home/releases/tag/opti1.0

Thanks! Looking at the message board thread, there is not much improvement of AVX-512 over AVX2. In this case, it probably better just to have a couple more cores running AVX2.


Original-Linux 51,988
Original-Windows 46,707
SSE2 30,681
SSE4.1 28,680
AVX 24,659
AVX2 23,882
AVX512 21,569

[scole of TSBT]

I'd take 10% anytime

[zombie67]

I'd take 10% anytime

Agreed, all things being equal.

But I was considering a 6 core Skylake-X i7-7800X with AVX-512 vs. an 8 core i9-9900K Coffee Lake with AVX2. I think the greater number of cores with AVX2 will out-perform the lesser number of cores with AVX-512. At least with this particular project. Am I wrong?

[scole of TSBT]

Get a i9-9900X and you'll get more cores plus avx512 :)

[zombie67]

Get a i9-9900X and you'll get more cores plus avx512 :)

Heh! Would love to, but a bit out of my budget... :D

[zombie67]

Well crap. Now that I am looking, I see the i9-9820X is on sale for $850. Same number of cores as the i9-9900X, but a slightly slower clock, and a bit less L3. The clock speed I think I can make up with OC. Will any of the projects really take advantage of more L3?

Skylake-X specs (https://en.wikipedia.org/wiki/Skylake_(microarchitecture)#High-end_desktop_processors)

[Bryan]

The only one I specifically know that relies on the amount of L3 is MIPs on WCG. The project scientist posted last year that you shouldn't run more than 2 MIPs at the same time because efficiency goes to hell in a hand basket on most CPUs. He gave a formula of L3/???? that would tell you how many you can run on a given host without paying a huge penalty.

[VietOZ]

Rake is also can use avx-512 I believe. Same dude wrote the app for Acoustics.

[zombie67]

It looks like the new MT genefer app at prime grid (https://www.primegrid.com/forum_thread.php?id=8168) uses AVX-512, and shows a lot of performance.

Edit: Maybe not yet. I misread what the developer was saying, I think. Sounds like he will be adding AVX-512, and expects big improvements over AVX2.

[zombie67]

FWIW, my new build with a i9-9820X (10/20 C/T), using the win 64 bit AVX-512 app does a task just under 1300 seconds with HT on. With HT off, it takes about 1150 seconds. I thought with only 2 512 processors, that would be the throttle-neck, so turning off HT would at least cut the time in half. Guess I was wrong. Anyway, this is all without any OC. I will try that soon.

[John P. Myers]

Many projects benefit from L3 because the CPU will load the entire .exe into L3 if it fits which greatly reduces latency compared to RAM. But since we typically use all threads on a CPU simultaneously, there's no room for that many instances. The i7-5775C is a very underrated CPU because so few people took advantage of the massive L4 cache.