Folding Forum

Gleep wrote:Unrelated to FAH, the current high core count offerings of AMD and Intel make it clear that Windows has some significant issues handling 30+ logical processors.

I agree.

Nevertheless, FAH has always focused on the *@home market and there are very few folks who will spend $1000+ for a CPU for a home computer. First is the basic issue of science. Adding an expensive GPU to a computer adds more computing power than adding a threadripper. Second, there's the issue of popularity. A lot more Donors see a gamer quality GPU as more useful than one with a limited video capability but with the ability to run more programs concurrently. People are much more likely to spend $500+ to $2000+ for a high end GPU than a high-end CPU. Doing so also earns a lot more FAH points reflecting FAH's preference for that improved science.

Of course I have no idea if 45 threads is better overall for the science or if it risks the slot stalling due to WU shortage or whatever. I'm surprised to see a CPU above 1M PPD, although most of the points are from the bonus of returning WUs so quickly.

FAH will not stall. You'll get an assignment, though it will reduce the number of allocated threads to some number that works.

Most people have compensated for FAH's shortcoming by manually splitting up the processors by creating multiple CPU slots. e.g.- two CPU:16 slots or even three CPU:12 slots are more likely to be fully utilized that one with a setting of CPU:32+

Oh, and I've heard that the 32-thread limitation is a Windows issue. Does you Linux system allow more (not that it's going to be a real benefit.)

The software used to construct FAHCore_a7 is from gromacs.org. I doubt they're going to change that selection inasmuch as the limitations are not a problem for most Donors.

GROMACS uses Domain Decomposition to allocate portions of the calculation to each SMP thread. This method works well with the CPUs on most home computers. That software is not used to construct the FAHCore's for GPUs.

Windows can handle high thread/core counts, but the way they license it gets in the way a lot. You need to be running with a server license for Windows to go beyond 32 threads, and specific programming and compiler flags need to have been used.

bruce wrote: Nevertheless, FAH has always focused on the *@home market and there are very few folks who will spend $1000+ for a CPU for a home computer. First is the basic issue of science. Adding an expensive GPU to a computer adds more computing power than adding a threadripper. Second, there's the issue of popularity. A lot more Donors see a gamer quality GPU as more useful than one with a limited video capability but with the ability to run more programs concurrently. People are much more likely to spend $500+ to $2000+ for a high end GPU than a high-end CPU. Doing so also earns a lot more FAH points reflecting FAH's preference for that improved science.

I agree completely. My setup is not typical, I post to reason thru FAH settings for optimal performance with the hardware I have and to let others know what the general performance is. A single top tier GPU does more FAH work in a day than a threadripper and cheaper.

I don't expect FAH to make any changes to accommodate this type of hardware.

bruce wrote: Oh, and I've heard that the 32-thread limitation is a Windows issue. Does you Linux system allow more (not that it's going to be a real benefit.)

The software used to construct FAHCore_a7 is from gromacs.org. I doubt they're going to change that selection inasmuch as the limitations are not a problem for most Donors.

GROMACS uses Domain Decomposition to allocate portions of the calculation to each SMP thread. This method works well with the CPUs on most home computers. That software is not used to construct the FAHCore's for GPUs.

It most definitely is a windows limitation. I'm dual booting windows/linux and in linux the FAH client auto config selected 46 threads (FAHCore_a7 dropped to 45), I found the PPD difference between the two OSes to be substantial, but that's not news to anyone around here.

Gleep wrote:I installed linux and let it auto configure and run over the weekend. From a PPD perspective the difference is massive. I have 2 GPUs now so it set the CPU slot to use 46 threads (which the logs show it drop to 45, I assume because of the prime thread count weirdness). The PPD of the CPU slot is 1-1.1 million. Under Windows I was getting 600-650k, with some slot configs dropping to 500k (the four slot config).

Of course I have no idea if 45 threads is better overall for the science or if it risks the slot stalling due to WU shortage or whatever. I'm surprised to see a CPU above 1M PPD, although most of the points are from the bonus of returning WUs so quickly.

Unrelated to FAH, the current high core count offerings of AMD and Intel make it clear that Windows has some significant issues handling 30+ logical processors.

1m PPD on a cpu slot hasn't been seen in years and even then it was a quad cpu set up, how times have changed!

I am running FAH in a docker on UnRAID. I have 20 of my 24 threads dedicated to FAH. There are two GPUs as well, but I have plenty of threads to handle that. I am only seeing about 45K PPD, but I am reading that the threadripper CPUs can hit near 100K PPD. Is there something I am doing wrong? Should I just do this in a VM instead?

A couple of things to note:

1) If you have not successfully completed 10 WUs after configuring a passkey, do that first. You don't get bonus points until then.

2) "I have heard that" ... is a dangerous preamble, particularly if they fail to specify whether they're talking about Linux or Windows. Wait for actual reports by Donors associated with YOUR OS.

3) Most hardware CPU threads share floating point resources with at least one oher thread ... and it is worse in a VM than with native hardware. FAH's CPU performance depends greatly on having dedicated floating point hardware.

4) See also the issues of Domain Decomposition mentioned on the first page of this topic.

A Ryzen 3950X (16/32 cores) for $750. looks good to me. I would wait for Ubuntu 18.04.4 though, to get the latest Linux kernel.
But what is the benefit to science? I like the idea of running Gromacs, but what does it do differently than the GPU app?

See Gromacs.org.

Gromacs is one of the software develpment projects that has been used by FAH to calculate protein folding since the very early days of FAH --- back when typical CPUs had 1 or 2 or 4 cores. Over the years, as an Open project, it as progressed in many areas. Meanwhile, Stanford chartered a parallel development effort called OpenMM which has also progressed in many areas and which focused mainly on useing GPUs although technically either package can run on either hardware platform. There are internal differences but I'm not knowledgable enough to explain them.

Other software has also been used as the basis of historic FAHCores but currently only these two are active.

I would recommend using a task manager or htop, and see CPU load.
If all cores are at a 100% load, try dropping a few threads, until 1 of the cores show just under 100%. Then bump it up by 1 thread. You'd be surprised that sometimes running a 24/20 configuration runs the same PPD or higher than a 24/24 config (maxed out cores).

Hi,

i have some similar problems here. have some Dual Socket EPYC 32core Servers here but i can not fully use the 128 Threads with FAH under Microsoft Server 2019.
I hope it will be fixed with a newer version.

Regards

EPYC_Server wrote:i have some Dual Socket EPYC 32core Servers here but i can not fully use the 128 Threads with FAH under Microsoft Server 2019.
I hope it will be fixed with a newer version.

As a work around, you can configure multiple CPU slots. Eight 16 core CPU slots should use everything.