Page 1 of 1

Cores with -1 not detected correctly?

Posted: Fri Jul 03, 2020 7:55 am
by SDI-89
Hello together,
In this log i can see that 64 cores are detected. But in the Slot with the -1 configuration only 32 cores are used.
Can anyone tell me why? And what do I have to do for best performance?

Code: Select all

07:47:02:WU01:FS00:0xa7:*********************** Log Started 2020-07-03T07:47:02Z ***********************
07:47:03:WU01:FS00:0xa7:************************** Gromacs Folding@home Core ***************************
07:47:03:WU01:FS00:0xa7:       Type: 0xa7
07:47:03:WU01:FS00:0xa7:       Core: Gromacs
07:47:03:WU01:FS00:0xa7:       Args: -dir 01 -suffix 01 -version 706 -lifeline 2224 -checkpoint 15 -np
07:47:03:WU01:FS00:0xa7:             16
07:47:03:WU01:FS00:0xa7:************************************ CBang *************************************
07:47:03:WU01:FS00:0xa7:       Date: Oct 26 2019
07:47:03:WU01:FS00:0xa7:       Time: 01:38:25
07:47:03:WU01:FS00:0xa7:   Revision: c46a1a011a24143739ac7218c5a435f66777f62f
07:47:03:WU01:FS00:0xa7:     Branch: master
07:47:03:WU01:FS00:0xa7:   Compiler: Visual C++ 2008
07:47:03:WU01:FS00:0xa7:    Options: /TP /nologo /EHa /wd4297 /wd4103 /Ox /MT
07:47:03:WU01:FS00:0xa7:   Platform: win32 10
07:47:03:WU01:FS00:0xa7:       Bits: 64
07:47:03:WU01:FS00:0xa7:       Mode: Release
07:47:03:WU01:FS00:0xa7:************************************ System ************************************
07:47:03:WU01:FS00:0xa7:        CPU: Intel(R) Xeon(R) Gold 6244 CPU @ 3.60GHz
07:47:03:WU01:FS00:0xa7:     CPU ID: GenuineIntel Family 6 Model 85 Stepping 7
07:47:03:WU01:FS00:0xa7:       CPUs: 64
07:47:03:WU01:FS00:0xa7:     Memory: 1.50TiB
07:47:03:WU01:FS00:0xa7:Free Memory: 2.45GiB
07:47:03:WU01:FS00:0xa7:    Threads: WINDOWS_THREADS
07:47:03:WU01:FS00:0xa7: OS Version: 6.2
07:47:03:WU01:FS00:0xa7:Has Battery: false
07:47:03:WU01:FS00:0xa7: On Battery: false
07:47:03:WU01:FS00:0xa7: UTC Offset: 2
07:47:03:WU01:FS00:0xa7:        PID: 2844
07:47:03:WU01:FS00:0xa7:        CWD: C:\Users\Administrator\AppData\Roaming\FAHClient\work
07:47:03:WU01:FS00:0xa7:******************************** Build - libFAH ********************************
07:47:03:WU01:FS00:0xa7:    Version: 0.0.18
07:47:03:WU01:FS00:0xa7:     Author: Joseph Coffland <joseph@cauldrondevelopment.com>
07:47:03:WU01:FS00:0xa7:  Copyright: 2019 foldingathome.org
07:47:03:WU01:FS00:0xa7:   Homepage: https://foldingathome.org/
07:47:03:WU01:FS00:0xa7:       Date: Oct 26 2019
07:47:03:WU01:FS00:0xa7:       Time: 01:52:30
07:47:03:WU01:FS00:0xa7:   Revision: c1e3513b1bc0c16013668f2173ee969e5995b38e
07:47:03:WU01:FS00:0xa7:     Branch: master
07:47:03:WU01:FS00:0xa7:   Compiler: Visual C++ 2008
07:47:03:WU01:FS00:0xa7:    Options: /TP /nologo /EHa /wd4297 /wd4103 /Ox /MT
07:47:03:WU01:FS00:0xa7:   Platform: win32 10
07:47:03:WU01:FS00:0xa7:       Bits: 64
07:47:03:WU01:FS00:0xa7:       Mode: Release
07:47:03:WU01:FS00:0xa7:************************************ Build *************************************
07:47:03:WU01:FS00:0xa7:       SIMD: avx_256
07:47:03:WU01:FS00:0xa7:********************************************************************************
07:47:03:WU01:FS00:0xa7:Project: 16806 (Run 8, Clone 673, Gen 13)
07:47:03:WU01:FS00:0xa7:Unit: 0x0000001282ed0b915eb41bf043828239
07:47:03:WU01:FS00:0xa7:Digital signatures verified
07:47:03:WU01:FS00:0xa7:Calling: mdrun -s frame13.tpr -o frame13.trr -cpi state.cpt -cpt 15 -nt 16
07:47:03:WU01:FS00:0xa7:Steps: first=6500000 total=500000
07:47:03:WU01:FS00:0xa7:Completed 109782 out of 500000 steps (21%)
07:47:08:WU01:FS00:0xa7:Completed 110000 out of 500000 steps (22%)
07:48:44:WU01:FS00:0xa7:Completed 115000 out of 500000 steps (23%)
07:50:20:WU01:FS00:0xa7:Completed 120000 out of 500000 steps (24%)
07:51:05:Saving configuration to config.xml

Re: Cores with -1 not detected correctly?

Posted: Fri Jul 03, 2020 8:19 am
by ajm
The standard configuration of CPU cores/threads is starting to date quite a bit. It was designed for a time when CPUs had only very few cores and the result with the latest CPUs is less than optimal. In your case, according to your log, the slot is actually using 16 threads:

Code: Select all

07:47:03:WU01:FS00:0xa7:Calling: mdrun -s frame13.tpr -o frame13.trr -cpi state.cpt -cpt 15 -nt 16
The -nt indicates how many threads (or CPUs as FAH calls them) are used.

You can change that in Advanced Control. Configure -> Slots -> double click "cpu" -> enter the number of threads/CPUs you want to use in the CPUs section (instead of "-1") -> OK -> Save. If you choose a number higher than the number presently used by a WU, it will become effective only for the next WU.

Now, it is not advisable yet (to my knowledge, but it is changing quite rapidly these days) to enter a number higher than 32, as only few WUs can really use them. It is better to split the threads among several slots, eg maybe 1x32, 1x24, 1x6, or 2x24 and 1X12. This in order to keep a few cores free for GPU management (count 1 thread per "normal" PGU and rather 2, or one core, for high-end GPUs). You have to avoid prime numbers and some of their multiples. 5 or 7 won't work at all, for example, and FAH will automatically reduce them. And some WUs don't accept multiples of primes like 10 or 15. The safest is to chose numbers divisible by 2 and 3.

Re: Cores with -1 not detected correctly?

Posted: Fri Jul 03, 2020 8:21 am
by JimboPalmer
My understanding is that while there is no limit in Linux, most versions of Windows have a 32 core limit per process. You can define two CPU slots of 32 each. (or 4 slots of 16 threads each)

Re: Cores with -1 not detected correctly?

Posted: Fri Jul 03, 2020 8:35 am
by SDI-89
Thank you for the answer.
I changed it to 1x32 1x24 1x8.

There is no GPU installed that can be used :)

Re: Cores with -1 not detected correctly?

Posted: Fri Jul 03, 2020 4:02 pm
by Neil-B
To be honest 2x32 would be better for the science (and therefore points) .. My 32/56 slot is pretty much always busy (at 32) - occasionally it has been "downstepped" by the AS to 12 when there has been a lack of WUs for larger core slots (maybe 10 times in total over multiple thousands of WUs) - although I have always run client-type as advanced (or more recently beta having joined the beta team) .. I use a 32/56 slot and a 24/56 slot and no GPU and have found that the lack of any "headroom" hasn't impacted my throughput so 2x32 should work well for you if you want to give it a go :)

Re: Cores with -1 not detected correctly?

Posted: Fri Jul 03, 2020 9:32 pm
by _r2w_ben
Xeon Gold 6244 is 8C/16T so this must be a 4S server. Either 2x32 or 4x16 should give the best results depending on Windows scheduling and how much time is spent syncing threads on different sockets.