18806 and 18820 Dumped WUs

[arisu]

Of you fold on all 12 threads, your GPU will choke and will be folding very slowly

He has it set to 11 threads but only 10 are being used. On v8, he has to set it to 12 threads if he wants to use 11 threads. If it is in the same resource group as the GPU, running a GPU WU will automatically subtract one thread and it will go to 11.

[muziqaz]

Of you fold on all 12 threads, your GPU will choke and will be folding very slowly

He has it set to 11 threads but only 10 are being used. On v8, he has to set it to 12 threads if he wants to use 11 threads. If it is in the same resource group as the GPU, running a GPU WU will automatically subtract one thread and it will go to 11.

You can override whatever client sets automatically in the beginning, which is 11/12
Client does not take one core from you if you set it to 12 out of 12, regardless of the fahcores.
On V8 he has to set it to 11 if he wants 11 to be used.

[arisu]

Of you fold on all 12 threads, your GPU will choke and will be folding very slowly

He has it set to 11 threads but only 10 are being used. On v8, he has to set it to 12 threads if he wants to use 11 threads. If it is in the same resource group as the GPU, running a GPU WU will automatically subtract one thread and it will go to 11.

You can override whatever client sets automatically in the beginning, which is 11/12
Client does not take one core from you if you set it to 12 out of 12, regardless of the fahcores.
On V8 he has to set it to 11 if he wants 11 to be used.

That's strange. I have mine set to 8/8 cores, slider all the way to the right, and it's using only 7 cores for the CPU and one for the GPU. When there is no GPU WU, it automatically increases itself to use all 8 cores (and then decrease to 7 if a GPU WU starts).

Code: Select all

$ lufah config cpus
8
$ lufah get units.0.cpus
1
$ lufah get units.1.cpus
7
$ lufah get units.0.gpus
[
  "gpu:03:00:00"
]
$ lufah get units.1.gpus
[]

It is definitely running with 7 and isn't just a lufah or API bug:

Code: Select all

$ ps -U fah-client -o cmd= | grep -Eo -- "-np.*"
-np 7

[muziqaz]

That might be biproduct of lufah. Natively controlled fah-client starts folding on 32 threads out of 32 (slider to the max), even when GPU is already folding. However, this is the case when CPU has its own resource group and GPU has its own group.
Ok, if GPU and CPU are sharing the same Resource Group (Default), CPU WU does end up reducing to -nt 31, while the slider is still at 32.

I think for simplicity reasons it is best to recommend people manually set the slider to max thread count minus number of folding GPUs on the system

[muziqaz]

The problem with fahcore_a9 might be that it doesn't take lightly the dynamic changes in core count mid folding

fahcore_a8 doesn't care. It will add extra threads, but it will not use them until new WU is downloaded and started.
a8 core drops threads dynamically mid current WU, but increase in threads only applies when new WU is started

[arisu]

At least on my system, a8 will happily change the number of threads. I could decrease it to one thread and it would use only one, then back up to 8 and it would use its maximum (7), even with the same WU.

Anyway in OP's case he does have to set it to 12 to get 11 threads. I think that is the case with most people since creating resource groups is more advanced.

[muziqaz]

there have not been any changes to a8 since our initial testing.
You start folding on a8 with 32 threads, it uses 32.
You change it to 31 while folding, it will change and use 31 straight away
You change back to 32 threads on the same WU which you had set to 32t, and then dropped to 31, it might pick back up to 32t and use them. However, if you start new WU with 31t, and increase to 32t while that WU is being folded, a8 will not fold on all 32t until new WU is downloaded and started.
It can only fold at thread count at which that WU started with, or less. But not more.

The reason we tested it(a8) so thoroughly is because a7 core had issues with certain thread counts (prime numbers)

[arisu]

That makes sense. I was starting at the maximum and found I could decrease and then increase it back.

[arisu]

there have not been any changes to a8 since our initial testing.
You start folding on a8 with 32 threads, it uses 32.
You change it to 31 while folding, it will change and use 31 straight away
You change back to 32 threads on the same WU which you had set to 32t, and then dropped to 31, it might pick back up to 32t and use them. However, if you start new WU with 31t, and increase to 32t while that WU is being folded, a8 will not fold on all 32t until new WU is downloaded and started.
It can only fold at thread count at which that WU started with, or less. But not more.

The reason we tested it(a8) so thoroughly is because a7 core had issues with certain thread counts (prime numbers)

Actually it must have changed. I just tested this with both a8 and a9 and it was able to start with 4 threads and increase to 8 threads. I tested this both with fah-client and by manually running the cores themselves. It used all the threads it was given, no matter how many it started with.

I even made the client think I had a 24 core system mid-WU (hooked sysconf), restarted the client, and the core started up 24 threads without issue (of course performance was terrible because they were all sharing just 8 cores).

If the cores are not meant to do that, then this must be a regression.

[Manndolin]

Update:

In the last few days I've tried two things:
1. Set CPU usage to 12/12 WITH the GPU enabled.
Result: Dropped WU from project 18820.
2. Set CPU usage to 12/2 WITHOUT the GPU enabled.
Result: Eventually loaded and completed a WU from 18820.

Although I do not know how to turn this into an actionable solution, I do now know that it will run the unit if it has access to as many CPUs as listed. The only problem is that I would much prefer to contribute with my GPU as well, especially since these are the only CPU jobs that seem to crash when my GPU is also running.

I may email the Flatiron Institute or who runs both projects or Cauldron Development who authored them and inform them of the issue.

[Manndolin]

12:13:54:I1:WU187:Requesting WU assignment for user __________ team 0
12:13:55:I1:WU187:Received WU assignment oN7wU5gAbON9Fmhg9ZIEKSZ-PF-SHdk4k__rqrDEqso
12:13:55:I1:WU187:Downloading WU
12:13:55:I1:WU187:Received WU P18820 R356 C0 G248
12:13:55:I3:WU187:Started FahCore on PID 13584
12:13:55:I1:WU187:*********************** Log Started 2025-03-26T12:13:55Z ***********************
12:13:55:I1:WU187:************************** Gromacs Folding@home Core ***************************
12:13:55:I1:WU187: Core: Gromacs
12:13:55:I1:WU187: Type: 0xa9
12:13:55:I1:WU187: Version: 0.0.12
12:13:55:I1:WU187: Author: Joseph Coffland <[email protected]>
12:13:55:I1:WU187: Copyright: 2022 foldingathome.org
12:13:55:I1:WU187: Homepage: https://foldingathome.org/
12:13:55:I1:WU187: Date: Nov 15 2022
12:13:55:I1:WU187: Time: 13:31:08
12:13:55:I1:WU187: Compiler: Visual C++
12:13:55:I1:WU187: Options: /TP /std:c++17 /nologo /EHa /wd4297 /wd4103 /O2 /Zc:throwingNew /MT
12:13:55:I1:WU187: Platform: win32 10
12:13:55:I1:WU187: Bits: 64
12:13:55:I1:WU187: Mode: Release
12:13:55:I1:WU187: SIMD: avx2_256
12:13:55:I1:WU187: OpenMP: ON
12:13:55:I1:WU187: CUDA: OFF
12:13:55:I1:WU187: OpenCL: OFF
12:13:55:I1:WU187: Args: -dir oN7wU5gAbON9Fmhg9ZIEKSZ-PF-SHdk4k__rqrDEqso -suffix 01
12:13:55:I1:WU187: -version 8.4.9 -lifeline 19376 -np 12
12:13:55:I1:WU187:************************************ libFAH ************************************
12:13:55:I1:WU187: Date: Nov 15 2022
12:13:55:I1:WU187: Time: 13:30:33
12:13:55:I1:WU187: Compiler: Visual C++
12:13:55:I1:WU187: Options: /TP /std:c++14 /nologo /EHa /wd4297 /wd4103 /O2 /Zc:throwingNew /MT
12:13:55:I1:WU187: Platform: win32 10
12:13:55:I1:WU187: Bits: 64
12:13:55:I1:WU187: Mode: Release
12:13:55:I1:WU187:************************************ CBang *************************************
12:13:55:I1:WU187: Date: Nov 15 2022
12:13:55:I1:WU187: Time: 13:29:57
12:13:55:I1:WU187: Compiler: Visual C++
12:13:55:I1:WU187: Options: /TP /std:c++14 /nologo /EHa /wd4297 /wd4103 /O2 /Zc:throwingNew /MT
12:13:55:I1:WU187: Platform: win32 10
12:13:55:I1:WU187: Bits: 64
12:13:55:I1:WU187: Mode: Release
12:13:55:I1:WU187:************************************ System ************************************
12:13:55:I1:WU187: CPU: 12th Gen Intel(R) Core(TM) i5-12400F
12:13:55:I1:WU187: CPU ID: GenuineIntel Family 6 Model 151 Stepping 5
12:13:55:I1:WU187: CPUs: 12
12:13:55:I1:WU187: Memory: 15.82GiB
12:13:55:I1:WU187:Free Memory: 7.54GiB
12:13:55:I1:WU187: Threads: WINDOWS_THREADS
12:13:55:I1:WU187: OS Version: 6.2
12:13:55:I1:WU187:Has Battery: false
12:13:55:I1:WU187: On Battery: false
12:13:55:I1:WU187: UTC Offset: -4
12:13:55:I1:WU187: PID: 13584
12:13:55:I1:WU187: CWD: C:\ProgramData\FAHClient\work
12:13:55:I1:WU187: Exec: C:\ProgramData\FAHClient\cores\gromacs-core-a9\windows-10-64bit\cpu-avx2_256-release\fahcore-a9-windows-10-64bit-cpu-avx2_256-release-0.0.12\FahCore_a9.exe
12:13:55:I1:WU187:********************************************************************************
12:13:55:I1:WU187:Project: 18820 (Run 356, Clone 0, Gen 248)
12:13:55:I1:WU187:Reading tar file core.xml
12:13:55:I1:WU187:Reading tar file frame248.tpr
12:13:55:I1:WU187:Digital signatures verified
12:13:55:I1:WU187:Calling: mdrun -c frame248.gro -s frame248.tpr -x frame248.xtc -cpt 5 -nt 12 -ntmpi 1 -update cpu -nb cpu -bonded cpu -pme cpu -pmefft cpu
12:13:55:I1:WU187:Steps: first=620000000 total=622500000
12:13:57:I1:WU187:Completed 1 out of 2500000 steps (0%)
12:14:48:I1:WU187:Completed 25000 out of 2500000 steps (1%)
12:14:59:W :WU187:Visualization frame 1 unchanged, skipping
12:15:36:I1:WU187:Completed 50000 out of 2500000 steps (2%)
....ETC ETC
13:30:31:W :WU187:Visualization frame 97 unchanged, skipping
13:31:07:I1:WU187:Completed 2450000 out of 2500000 steps (98%)
13:31:54:I1:WU187:Completed 2475000 out of 2500000 steps (99%)
13:32:05:W :WU187:Visualization frame 99 unchanged, skipping
13:32:41:I1:WU187:Completed 2500000 out of 2500000 steps (100%)
13:32:41:I1:WU187:Saving result file ..\logfile_01.txt
13:32:41:I1:WU187:Saving result file frame248.gro
13:32:41:I1:WU187:Saving result file frame248.xtc
13:32:41:I1:WU187:Saving result file md.log
13:32:41:I1:WU187:Saving result file science.log
13:32:41:I1:WU187:Saving result file state.cpt
13:32:41:I1:WU187:Folding@home Core Shutdown: FINISHED_UNIT
13:32:42:I1:WU187:Core returned FINISHED_UNIT (100)
13:32:42:I1:WU187:Uploading WU results
13:32:42:I1:WU187:Credited

[muziqaz]

Set the slider to for CPU to 10, and leave the GPU ticked. Save. Happy folding
Or, heck, set it to 8.