16th Annual Folding@Home Holiday Season Race: The race is over and the Moonshots win!

Page 9 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

StefanR5R

Elite Member
Dec 10, 2016
4,682
5,767
136
Thinking out loud...
When the PG thing is done, I may have to resort to desperate measures like CPU folding. Research will be needed:
  • Has the GRO_A8 core been improved to work with higher threadcounts yet?
  • If not, can the selection of a "cause" (folding slot option, the disease to research) be used to enforce work reception for GRO_A7 only?
    Note to self, the latest post in "Announcements - Folding Consortium" which explicitly mentions GRO_A7 is about a COVID project.
  • If not, is it perhaps possible to blacklist workservers which carry GRO_A8 work?
  • It is possible to "abort-and-report" GRO_A8 work shortly after it was received, manually or scripted. But maybe foldingathome.org has mechanisms in place which penalize clients with frequent failures of this kind.
 

Pokey

Platinum Member
Oct 20, 1999
2,730
398
126
Speaking of cause preference made me look at my advanced tabs:
They only promise to give you the work when available, no guarantees.

For what it's worth I took a look at my current WUs and although I have Covid-19 spec'd on all the rigs, I didn't have a single Covid WU. Most were cancer, one Alz, and a few "unspecified". So I suspect specifying All in preferences probably won't hurt you. You are going to get what they have available no matter the flag.
 

Endgame124

Senior member
Feb 11, 2008
877
549
136
Thinking out loud...
When the PG thing is done, I may have to resort to desperate measures like CPU folding. Research will be needed:
  • Has the GRO_A8 core been improved to work with higher threadcounts yet?
  • If not, can the selection of a "cause" (folding slot option, the disease to research) be used to enforce work reception for GRO_A7 only?
    Note to self, the latest post in "Announcements - Folding Consortium" which explicitly mentions GRO_A7 is about a COVID project.
  • If not, is it perhaps possible to blacklist workservers which carry GRO_A8 work?
  • It is possible to "abort-and-report" GRO_A8 work shortly after it was received, manually or scripted. But maybe foldingathome.org has mechanisms in place which penalize clients with frequent failures of this kind.
Can you create multiple 32 CPU slots for multiple parallel WU?
 

Icecold

Golden Member
Nov 15, 2004
1,070
983
146
Can you create multiple 32 CPU slots for multiple parallel WU?
I'm pretty sure you can but with the quick return bonus you get more points if you can max out your thread count. I will likely try it out on the 256 thread EPYC after the Primegrid race is over.
 

Endgame124

Senior member
Feb 11, 2008
877
549
136
I'm pretty sure you can but with the quick return bonus you get more points if you can max out your thread count. I will likely try it out on the 256 thread EPYC after the Primegrid race is over.
I believe last year F@H said the max CPU for A8 WUs was 32 CPU, so that might be the best you can do for CPU per task
 

StefanR5R

Elite Member
Dec 10, 2016
4,682
5,767
136
With GRO_A7, a single 128-threaded slot on dual 7453 or single 88-threaded slot on dual E5-2696 v4 was much superior in PPD to any combination of split CPU slots.

The reason for this is of course Folding@Home's over-emphasis on the Quick Return Bonus, as @Icecold pointed out.

However, even before GRO_A8 was added, you sometimes were unable to get work which would be compatible with such high thread counts. And in case of the 88-thread computer, the additional trouble is that the thread count of a CPU slot should be a figure which is a composite of small factors. (I.e. a power of two, maybe with an occasional 3 or 5 factored in.) That's because the computational domain can't be easily divided into subdomains which are a multiple of higher factors such as 7 or 11. Hence, when I wanted to use 88-threaded computers for longer periods of Folding, I generally ended up dividing this into a 64-threaded and a 24-threaded slot. In case of a 128-threaded computer, there were periods during which I had to dumb it down to 2 64-threaded slots in order to receive work, for considerably lower total PPD than 1x128. But dual E5-2690 v4 with 56 threads generally could run with a single 56-threaded slot all the time, back then, even though their thread count is not ideal with 7*2^3. (Again, all this was before the disastrous introduction of GRO_A8.)
 
Last edited:

biodoc

Diamond Member
Dec 29, 2005
6,064
1,928
136
@StefanR5R , I fired up my old dual socket Ivy Bridge (40 threads) and it is using all 40 threads on an GRO_A8 WU. It's showing 454K PPD.

Code:
19:08:20:WU00:FS00:0xa8:        CPU: Intel(R) Xeon(R) CPU E5-2690 v2 @ 3.00GHz
19:08:20:WU00:FS00:0xa8:     CPU ID: GenuineIntel Family 6 Model 62 Stepping 4
19:08:20:WU00:FS00:0xa8:       CPUs: 40
19:08:20:WU00:FS00:0xa8:     Memory: 94.33GiB
19:08:20:WU00:FS00:0xa8:Free Memory: 92.51GiB
19:08:20:WU00:FS00:0xa8:    Threads: POSIX_THREADS
19:08:20:WU00:FS00:0xa8: OS Version: 5.4
19:08:20:WU00:FS00:0xa8:Has Battery: false
19:08:20:WU00:FS00:0xa8: On Battery: false
19:08:20:WU00:FS00:0xa8: UTC Offset: -5
19:08:20:WU00:FS00:0xa8:        PID: 2236
19:08:20:WU00:FS00:0xa8:        CWD: /var/lib/fahclient/work
19:08:20:WU00:FS00:0xa8:********************************************************************************
19:08:20:WU00:FS00:0xa8:Project: 18302 (Run 46, Clone 0, Gen 0)
19:08:20:WU00:FS00:0xa8:Unit: 0x00000000000000000000000000000000
19:08:20:WU00:FS00:0xa8:Digital signatures verified
19:08:20:WU00:FS00:0xa8:Calling: mdrun -c frame0.gro -s frame0.tpr -x frame0.xtc -cpi state.cpt -cpt 15 -nt 40 -ntmpi 1
The rankings and data from here seem suspect but a 32 core threadripper is at the top.
 

biodoc

Diamond Member
Dec 29, 2005
6,064
1,928
136
BTW, it looks like my Marauders are going down in production, and the Moonshot gang is going up ! One week until we can all get back to just this.
I shut down the rented 3090 yesterday morning so i'm running on my own hardware which is 19.5 M PPD which means the daily production gap will increase again for the Marauders.

1639303857180.png
 
Last edited:

biodoc

Diamond Member
Dec 29, 2005
6,064
1,928
136
Well, so far the Moonshot Gang is only 10% behind. Sure, it's certainly going to get worse and worse during the next days until Friday inclusive. But then we shall strike back, big time! >:-D
I appreciate your enthusiasm, Stefan. Are you ok with leaving the teams as is for the remainder of the month? It's definitely an uphill battle.
 

Pokey

Platinum Member
Oct 20, 1999
2,730
398
126
Well, so far the Moonshot Gang is only 10% behind. Sure, it's certainly going to get worse and worse during the next days until Friday inclusive. But then we shall strike back, big time! >:-D
I await with bated breath............................. :p
 

StefanR5R

Elite Member
Dec 10, 2016
4,682
5,767
136
Well, so far the Moonshot Gang is only 10% behind. Sure, it's certainly going to get worse and worse during the next days until Friday inclusive. But then we shall strike back, big time! >:-D
I appreciate your enthusiasm, Stefan. Are you ok with leaving the teams as is for the remainder of the month? It's definitely an uphill battle.
I await with bated breath............................. :p
Yeah, I know, these are big words from somebody who has got just some slightly dated Pascal GPUs.¹ To be honest, I am betting a lot on @Icecold's Folding comeback too.

@biodoc, I for one am OK with leaving the teams as they are. But a) I haven't done any math, b) maybe I am the only one of the Marauders who is that optimistic. :-)


________________
¹) OK, and some rather nice CPUs. But it remains to be seen what these are worth at Folding@home.
 
Last edited:
  • Like
Reactions: Icecold

ASK THE COMMUNITY