Folding@Home bigadv EOL 2nd Anniversary Challenge on 12:00 UTC January 23rd: The race is over.

Page 6 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

motqalden

Member
Jun 13, 2009
82
111
106
While i agree with your general consensus on "efficiency" I think the contest part of the equation throws this out the window as a consideration and leaves me with simply "can i get more points overall or not" I would not run CPU normally on this project i can agree on that part of it.
 

crashtech

Lifer
Jan 4, 2013
10,521
2,111
146
When using really low end cards (geforce 710, r250x) I don't think you need to worry about it as much. I see no change in PPD on either of those low end cards if I leave a dedicated CPU for them, or not.

edit - on my ancient A10-7870k, I have the igpu + the r250x running F@H and 3 of 4 CPU threads folding. The host CPU is around 81% - Honestly I might even be able to use 4 CPU threads and still get a increase in PPD.
Right, many caveats here. I was assuming (incorrectly) that an average folding GPU would be something like a 1070 or better. Also in the other direction, slower platforms and CPUs might need to have more resources conserved for the GPU than others.
 

Assimilator1

Elite Member
Nov 4, 1999
24,120
507
126
Ok so I'm totally lost now, should I have a spare physical core for F@H GPU or not? lol :p (rigs in sig).
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,481
14,434
136
Ok so I'm totally lost now, should I have a spare physical core for F@H GPU or not? lol :p (rigs in sig).
I would say reserve 1 on the 3600 and 2 on the 4930k for the GPU., so, 5 and 10 respectively for CPU.
 

Assimilator1

Elite Member
Nov 4, 1999
24,120
507
126
The 4930k is a 6 core CPU, shouldn't I do 10 threads for that too?

I'll give that a shot with the Ryzen anyway, current combined ppd for it is 522k, (108k for the CPU, 414k for the GPU on project 17424).
[edit] If I change that mid way through a WU will it abort it?
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,481
14,434
136
The 4930k is a 6 core CPU, shouldn't I do 10 threads for that too?

I'll give that a shot with the Ryzen anyway, current combined ppd for it is 522k, (108k for the CPU, 414k for the GPU on project 17424).
[edit] If I change that mid way through a WU will it abort it?
Isn't that what I said ? as far as changing mid-stream, I have never had one abort by just changing number of CPU's
 
  • Like
Reactions: voodoo5_6k

Endgame124

Senior member
Feb 11, 2008
954
669
136
Right, many caveats here. I was assuming (incorrectly) that an average folding GPU would be something like a 1070 or better. Also in the other direction, slower platforms and CPUs might need to have more resources conserved for the GPU than others.
From personal testing, a 1660 super had approximately the same ppd with 1 cpu open on a a10-5800k (appx core 2quad speed) vs a 2700x with one core open. This was pre cuda though, so I can’t say this is still the case. My 1080ti was definitely slower on the A10-5800k - I had to leave 2 cores open to keep the same ppd on the 1080ti.
 

Assimilator1

Elite Member
Nov 4, 1999
24,120
507
126
Isn't that what I said ? as far as changing mid-stream, I have never had one abort by just changing number of CPU's
Oh sorry I thought you meant 10 for the Ryzen, in which case why did you say 5 for the Ryzen? Its 6C12T...., I guess that was your mistake ;), unless I'm missing something?

Good to know on the latter point.
(current ppd total ~517k ppd, GPU 410k, CPU 107k project 16927, 12T CPU).
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,481
14,434
136
Oh sorry I thought you meant 10 for the Ryzen, in which case why did you say 5 for the Ryzen? Its 6C12T...., I guess that was your mistake ;), unless I'm missing something?

Good to know on the latter point.
(current ppd total ~517k ppd, GPU 410k, CPU 107k project 16927, 12T CPU).
well, its experimentation. The 3600 might need 2 virtual cores, maybe just one. But yes, most likely BOTH need 2T GPU, 10T CPU. I was thinking 6t, not 12 for the Ryzen, Duhhhhh
 

Icecold

Golden Member
Nov 15, 2004
1,090
1,008
146
Just added 56 threads of Haswell Xeon for ~450k PPD.

Edit - I mean.. uhh.. if anybody from [H] is reading this right now.. nobody here has anything in reserve.. you can probably let up now and enjoy a hard earned victory.. Basically what I'm getting at is:


points.png
 
Last edited:

voodoo5_6k

Senior member
Jan 14, 2021
395
443
116
I don't use any CPU for folding, as its very inefficient. But the same applies to other things. So my lowest cpu core count are 3900x's with 24 threads. I use 22 threads for WCG or Rosetta, and 2 for F@H along with the video card. One physical core is always been good.
well, its experimentation. The 3600 might need 2 virtual cores, maybe just one. But yes, most likely BOTH need 2T GPU, 10T CPU. [...]
Interesting discussion and thanks for the estimation/information!

I've tested it on my system and this worked out pretty good, gave the system a nice little overall boost in PPD... :sunglasses: Just in time for this close race :sweat:

Difference between 2T GPU/everything else CPU and 50%T GPU/50%T CPU was 1-2s in TPF for the current GPU WU (3m:19s - 3m:21s estimated TPF). I'll keep checking this with the next few GPU WUs.
 
Last edited:

Assimilator1

Elite Member
Nov 4, 1999
24,120
507
126
Icecold
Lol, good one :D

(current ppd total ~517k ppd, GPU 410k, CPU 107k project 16927, 12T CPU).
I don't know how much difference different projects makes to points (and I failed to note the GPU project for some reason!).
But my main rig is currently recording a total of ppd of ~631k, GPU 490k (project 17324) and CPU 141k (project 16949). If that difference (+22%) largely holds through different projects, that's quite a jump! :) (and I'm sure it was said not long ago that leaving the CPU slot to auto was the best config!).

My 2nd rig's current total ppd is 265k, GPU 163k (project 17426), CPU 103k (project 16948), with 12T on CPU. Switching to 10T now on that rig.

Well that's odd, I switched from 12T to 10T, then looked at task manager, F@H is taking ~83% of the CPU as expected, but when I switched back to 12T it stayed at ~83% (I looked for a minute or so), what's that about?
 
Last edited:

Endgame124

Senior member
Feb 11, 2008
954
669
136
Well that's odd, I switched from 12T to 10T, then looked at task manager, F@H is taking ~83% of the CPU as expected, but when I switched back to 12T it stayed at ~83% (I looked for a minute or so), what's that about?
CPU changes don't take effect until the next WU starts as far as I understand.
 
  • Like
Reactions: Assimilator1

Ken g6

Programming Moderator, Elite Member
Moderator
Dec 11, 1999
16,219
3,798
75
Stats are looking a little strange lately. @ericlp passed me, then appeared to stop, and I passed him again. Some of my results, especially CPU work, didn't appear to get credited either.
 

StefanR5R

Elite Member
Dec 10, 2016
5,459
7,718
136
CPU changes don't take effect until the next WU starts as far as I understand.
Unless you delete and recreate the slot, but then you lose the work.
IME it is very often possible to change the thread count of a CPU slot with a workunit in progress. FahCore stops and restarts automatically, and may even restart at non-zero progress.

What is GRO_A8?
Looks like a newer CPU core that utilizes AVX2/FMA3 - https://foldingforum.org/viewtopic.php?f=24&t=36051#p342216
I haven't checked out this forum link, but I can already tell you: FahCore_a8 sux! ;-)

FahCore_a7 scales easily to 128 threads.
FahCore_a8 only scales to 64 threads. When loaded into a 128-threaded slot, it logs this…
ERROR:128 OpenMP threads were requested. Since the non-bonded force buffer reduction is prohibitively slow with more than 64 threads, we do not allow this. Use 64 or less OpenMP threads.​
…and starts singlethreaded. :facepalm:

Edit: dual EPYC 7452, TDP/PPT set to 180 W:
FahCore_a8 in a 64-threaded slot (other 64 threads unused): 170 kPPD
FahCore_a7 in a 128-threaded slot: 2.8 M PPD
 
Last edited:

Icecold

Golden Member
Nov 15, 2004
1,090
1,008
146
I haven't checked out this forum link, but I can already tell you: FahCore_a8 sux! ;-)
That has been my experience as well, outside of that I do not have any machines that are 64 threads or higher. I was excited to get an A8 task thinking the PPD would be much better due to AVX2/FMA3 and it being a newer core, but the PPD was significantly lower than A7 tasks on both a Ryzen 3950x and Haswell Xeon.
 

crashtech

Lifer
Jan 4, 2013
10,521
2,111
146
...IME it is very often possible to change the thread count of a CPU slot with a workunit in progress. FahCore stops and restarts automatically, and may even restart at non-zero progress...
Agree absolutely, but if one is doing testing on utilization, especially outside of a contest environment, deleting the slot always works. It IS wasteful of CPU cycles, for sure.