Question 10900k real world comparison to 3900k . Maybe, if I can buy one.

[Markfw]

The 10900K was available for a few minutes today at Newegg (at $529)

Now that its sort of there, I put myself in on auto-notify

 

[Yeroon]

With the distributed computing stuff mark does, the intel might have an edge in some things. Anything able to make use of avx512 ?

I've seen it mentioned a few times here, but 10900K doesnt support anything past avx2. Its still rehashed skylake round 4.
You'd need the 10900X hedt for versions of avx-512.

 

[Markfw]

I've seen it mentioned a few times here, but 10900K doesnt support anything past avx2. Its still rehashed skylake round 4.
You'd need the 10900X hedt for versions of avx-512.

That won't help what I try to do, or this thread.

I just think that other than a few FPS on gaming, its hot and sucks a lot of power. I just ordered my second EPYC 7742. They take 225 watts for 64 cores. Thats at 24/7 @ 100% usage.

Who cares about AVX512. Surely not me.

 

[amrnuke]

I've seen it mentioned a few times here, but 10900K doesnt support anything past avx2. Its still rehashed skylake round 4.
You'd need the 10900X hedt for versions of avx-512.

The only distributed computing scenarios with AVX512 that I'm aware of are purpose-written corporate/private projects not open to the public. Would make no sense to pay the premium if nothing on BOINC or F@H can use it.

 

[DrMrLordX]

Why can't F@H use AVX512?

 

[Keysplayr]

Is it worth actually buying a Comet Lake CPU right now? Isn't this supposedly the last 14nm milk for Intel? 10nm or 7nm (whatever they call it internally) is around the corner or is it a long way off? I keep visiting microcenter and newegg sites often lately which tells me I have the upgrade itch. But, I don't mind waiting for Intels die shrink gen to see how that goes, or at the very least Zen3 from AMD.

 

[jpiniero]

Is it worth actually buying a Comet Lake CPU right now?

For gaming yes. But it's hard to acquire in DIY so it may not even be an option.

Isn't this supposedly the last 14nm milk for Intel?

Nope, one more next year, called Rocket Lake.

10nm or 7nm (whatever they call it internally) is around the corner or is it a long way off?

For desktop it's another two years for something that's not 14 nm.

 

[DrMrLordX]

Is it worth actually buying a Comet Lake CPU right now?

Not really. At least wait until AMD launches their products this year, either Matisse Refresh or Vermeer. Then compare performance. Availability on the 10900k is really poor as well.

 

[Markfw]

BTW, I already just got another EPYC 7742, so if it becomes available, I may have to wait a bit before I get it. This is just a test anyways,

 

[amrnuke]

Why can't F@H use AVX512?

It can. The question is whether they've chosen to do so. As I recall, almost all work units do not utilize it.

 

[Yeroon]

That won't help what I try to do, or this thread.

I just think that other than a few FPS on gaming, its hot and sucks a lot of power. I just ordered my second EPYC 7742. They take 225 watts for 64 cores. Thats at 24/7 @ 100% usage.

Who cares about AVX512. Surely not me.

The only distributed computing scenarios with AVX512 that I'm aware of are purpose-written corporate/private projects not open to the public. Would make no sense to pay the premium if nothing on BOINC or F@H can use it.

I was merely pointing out to other users who seemed to think the 10900K had avx512, and that there might be a benefit to that over a zen2 cpu.

I think most DC type applications would sooner move to opencl type work, as the gpu power available would be far more than a HEDT/server only instruction set as of now.

 

[myocardia]

I was merely pointing out to other users who seemed to think the 10900K had avx512, and that there might be a benefit to that over a zen2 cpu.

I think most DC type applications would sooner move to opencl type work, as the gpu power available would be far more than a HEDT/server only instruction set as of now.

Too late. The largest DC application, Folding @ Home, has been using Cuda for years now. I'm not sure whether or not it can be run with OpenCL, but I would imagine that it can.

 

[myocardia]

It can. The question is whether they've chosen to do so. As I recall, almost all work units do not utilize it.

It honestly wouldn't matter if it could use it. Only an extremely small percentage of the processors doing all of the different types of distributed computing, at least those owned by individuals and not by companies, are able to run AVX 512 code. That's the reason that more or less all of them the software uses AVX 2. Every Intel CPU since the first gen Haswells can run AVX 2. It would be counterintuitive, and even counterproductive for them to waste their time/resources to write code that maybe 0.5-1% of the CPUs running their code can benefit from.

edit: Or am I being Captain Obvious to point that out?

 

[Yeroon]

Too late. The largest DC application, Folding @ Home, has been using Cuda for years now. I'm not sure whether or not it can be run with OpenCL, but I would imagine that it can.

Folding at home with gpu is OpenCL only right now, using both SP and DP precision as needed (a22 core). Cpu core is avx-256. Most other projects are still cpu only, though occasionally a gpu project will run for a while. Nearly all cpu projects that I know of utilize avx2 when available.

I'm of the opinion the money is much better spent towards epyc, as Mark has already done. Anyone who buys a 10900K for something outside of gaming either has a specific workload that happens to work better on intel, perf/watt be damned, or they have their head in the sand. (Competitive overclocking gets an exception)

 

[Markfw]

Folding at home with gpu is OpenCL only right now, using both SP and DP precision as needed (a22 core). Cpu core is avx-256. Most other projects are still cpu only, though occasionally a gpu project will run for a while. Nearly all cpu projects that I know of utilize avx2 when available.

I'm of the opinion the money is much better spent towards epyc, as Mark has already done. Anyone who buys a 10900K for something outside of gaming either has a specific workload that happens to work better on intel, perf/watt be damned, or they have their head in the sand. (Competitive overclocking gets an exception)

Yes, I do all my F@H on GPU, and WCG and Rosetta (BOINC) on CPU. The most efficient CPU is the EPYC series. 7002 series to be exact. This will be my 5th system, and my 3rd Rome system. Naples is barely better than Ryzen.

 

[myocardia]

Folding at home with gpu is OpenCL only right now, using both SP and DP precision as needed (a22 core). Cpu core is avx-256. Most other projects are still cpu only, though occasionally a gpu project will run for a while. Nearly all cpu projects that I know of utilize avx2 when available.

This page led me to believe that: https://foldingathome.org/faqs/gpu3-common/frequently-asked-questions/what-gpus-are-supported/for-nvidia/ Note that I am not saying that you are wrong.

edit: Deleted the 2nd quoted paragraph, to which I was not responding.

 

[Yeroon]

This page led me to believe that: https://foldingathome.org/faqs/gpu3-common/frequently-asked-questions/what-gpus-are-supported/for-nvidia/ Note that I am not saying that you are wrong.

Requirements - Folding@home

foldingathome.org

Opencl since fahcore_17, current gpu core is fahcore_22

 

[Markfw]

Well, I will say this... PPD is almost directly related to cuda cores. (within a family, like 1000 or 2000 series)

 

[myocardia]

Mark, do you happen to know why most people use nVidia cards for F@H, when AMD cards have always dominated, when it came to OpenCL performance?

edit: I mean most of the people who fold for Team Anandtech, not most of the people worldwide, BTW.

 

[jpiniero]

Was in stock again for like 5 minutes a few hours ago. Newegg does have the 10700K in stock (albeit at $409)

 

[Markfw]

Mark, do you happen to know why most people use nVidia cards for F@H, when AMD cards have always dominated, when it came to OpenCL performance?

In F@H, Nvidia cards are almost double the PPD/$ of AMD cards. I think its due to all the calculations are not double precision, THERE AMD cards KILL Nvidia.

 

[myocardia]

In F@H, Nvidia cards are almost double the PPD/$ of AMD cards. I think its due to all the calculations are not double precision, THERE AMD cards KILL Nvidia.

Ahh, that makes perfect sense. Thanks for the clarification.

 

[Yeroon]

I believe the gpu core is based around openMM, and the code is optimized far better for nvidia in general, even when cuda is not used. Cuda has been the dominant api in that field, so it makes sense that if majority of your cards are already nvidia, thats what you optimize for. Its also a self feeding loop, people buy the card that nets the most perf/$.

https://github.com/openmm/openmm/issues/2471

 

[Ken g6]

For everyone, please let me know feedback on
ASUS TUF GAMING Z490-PLUS (WI-FI) LGA 1200

Its $200 just like my 3900x board, and the name is the close ASUS Prime X470-Pro AM4 AMD

I was looking at a build a guy is considering in Computer Building. I looked at the board you mention, but he mentioned the Asus Prime Z490 A. It looks significantly better in VRM and, supposedly, memory performance, for only $30 more.

 

[Markfw]

I was looking at a build a guy is considering in Computer Building. I looked at the board you mention, but he mentioned the Asus Prime Z490 A. It looks significantly better in VRM and, supposedly, memory performance, for only $30 more.

Thanks, If I ever can find the CPU for sale, I will go that route.