Issue with multi-socket Intel wildcat in Windows Server 2019

Moderators: Site Moderators, FAHC Science Team

Issue with multi-socket Intel wildcat in Windows Server 2019

Postby fishy_boy256 » Sat May 08, 2021 4:22 am

I am brand new to folding, I have a lot of excess computing power lying around and I thought I would try it out. Folding works great on my gaming computer so I tried a bigger one. I've got a 2016 Intel Wildcat rack server capable of supporting two Xeon E5-2699 CPUs, each with 18 cores (so, 36 total and 64 usable threads). All 36 cores are accounted for in System Information and task manager reports their usage.
There is no GPU and there is a total of 64 GB of RAM installed (all of which the FAH client recognizes).

When folding at Full, only the 32 threads of one socket (node 0) get used. The remaining 32 threads sit idle.
Currently, the used cores are warming up to about 75 degrees while node 1 cores sit at 40 degrees, which is a little bit concerning. I want to use all 64 threads to fold at Low, hopefully reducing the temperature without affecting productivity (or possibly even increasing it).
(^^I am using Speccy to monitor their temperature)

I have tried making two CPU folding slots each with 32 cores but the node 1 cores still report 0 usage. I have tried making 2 slots with 16 each with no effect. I have tried making one CPU folding slot with 64 cores but it just auto sets it back to 32 and only uses node 0 cores.

Interestingly the RAM should be split between the two CPUs (1 CPU only has access to half), FAH reports 32 cores but all the RAM.
I had thought "Well, it does say 32 cores (which is the correct number of usable cores), but there should be 64 threads. My gaming PC has an intel i9-9900K (8 cores, 16 threads) and FAH reports "16 cores" so it must be measuring threads and not just cores".

I would like to split 2 WUs between the two CPUs so that way they don't need to cross and interact but in order to get that started, I need to be able to use node 1.

Can someone point me in the right direction to set this up?
I was also wondering how many ppd (or WUs or whatever stat this should be measured in) a computer of this size should be producing.

Here are the first ~100 lines of the log (as I've seen requested many times)
Code: Select all
*********************** Log Started 2021-05-07T22:44:28Z ***********************
22:44:28:******************************* libFAH ********************************
22:44:28:         Date: Oct 20 2020
22:44:28:         Time: 13:36:55
22:44:28:     Revision: 5ca109d295a6245e2a2f590b3d0085ad5e567aeb
22:44:28:       Branch: master
22:44:28:     Compiler: Visual C++ 2015
22:44:28:      Options: /TP /nologo /EHa /wd4297 /wd4103 /O2 /Zc:throwingNew /MT
22:44:28:     Platform: win32 10
22:44:28:         Bits: 32
22:44:28:         Mode: Release
22:44:28:****************************** FAHClient ******************************
22:44:28:      Version: 7.6.21
22:44:28:       Author: Joseph Coffland <joseph@cauldrondevelopment.com>
22:44:28:    Copyright: 2020 foldingathome.org
22:44:28:     Homepage: https://foldingathome.org/
22:44:28:         Date: Oct 20 2020
22:44:28:         Time: 13:41:04
22:44:28:     Revision: 6efbf0e138e22d3963e6a291f78dcb9c6422a278
22:44:28:       Branch: master
22:44:28:     Compiler: Visual C++ 2015
22:44:28:      Options: /TP /nologo /EHa /wd4297 /wd4103 /O2 /Zc:throwingNew /MT
22:44:28:     Platform: win32 10
22:44:28:         Bits: 32
22:44:28:         Mode: Release
22:44:28:         Args: --open-web-control
22:44:28:******************************** CBang ********************************
22:44:28:         Date: Oct 20 2020
22:44:28:         Time: 11:36:18
22:44:28:     Revision: 7e4ce85225d7eaeb775e87c31740181ca603de60
22:44:28:       Branch: master
22:44:28:     Compiler: Visual C++ 2015
22:44:28:      Options: /TP /nologo /EHa /wd4297 /wd4103 /O2 /Zc:throwingNew /MT
22:44:28:     Platform: win32 10
22:44:28:         Bits: 32
22:44:28:         Mode: Release
22:44:28:******************************* System ********************************
22:44:28:          CPU: Intel(R) Xeon(R) CPU E5-2699 v3 @ 2.30GHz
22:44:28:       CPU ID: GenuineIntel Family 6 Model 63 Stepping 2
22:44:28:         CPUs: 32
22:44:28:       Memory: 63.91GiB
22:44:28:  Free Memory: 61.46GiB
22:44:28:      Threads: WINDOWS_THREADS
22:44:28:   OS Version: 6.2
22:44:28:  Has Battery: false
22:44:28:   On Battery: false
22:44:28:   UTC Offset: -7
22:44:28:          PID: 2660
22:44:28:          CWD: C:\ProgramData\FAHClient
22:44:28:Win32 Service: false
22:44:28:           OS: Windows Server 2019 Standard Evaluation
22:44:28:      OS Arch: AMD64
22:44:28:         GPUs: 0
22:44:28:         CUDA: Not detected: Failed to open dynamic library 'nvcuda.dll': The
22:44:28:               specified module could not be found.
22:44:28:
22:44:28:       OpenCL: Not detected: clGetPlatformIDs() returned -1001
22:44:28:***********************************************************************
22:44:28:<config>
22:44:28:  <!-- Folding Slots -->
22:44:28:</config>
22:44:28:Trying to access database...
22:44:28:Successfully acquired database lock
22:44:28:FS00:Initialized folding slot 00: cpu:31
22:45:29:Saving configuration to config.xml
22:45:29:<config>
22:45:29:  <!-- Folding Slots -->
22:45:29:  <slot id='0' type='CPU'/>
22:45:29:</config>
22:45:29:Set client configured
22:45:29:WU00:FS00:Connecting to assign1.foldingathome.org:80
22:45:30:WU00:FS00:Connecting to assign1.foldingathome.org:80
22:45:30:WU00:FS00:Assigned to work server 129.32.209.205
22:45:30:WU00:FS00:Requesting new work unit for slot 00: cpu:31 from 129.32.209.205
22:45:30:WU00:FS00:Connecting to 129.32.209.205:8080
22:45:30:WU00:FS00:Downloading 1.47MiB
22:45:30:WU00:FS00:Download complete
22:45:31:WU00:FS00:Received Unit: id:00 state:DOWNLOAD error:NO_ERROR project:16969 run:8 clone:332 gen:0 core:0xa8 unit:0x0000014c000000000000424900000008
22:45:31:WU00:FS00:Downloading core from http://cores.foldingathome.org/win/64bit-av
fishy_boy256
 
Posts: 3
Joined: Sat May 08, 2021 3:35 am

Re: Issue with multi-socket Intel wildcat in Windows Server

Postby JimboPalmer » Sat May 08, 2021 6:41 am

Welcome to Folding@Home!

https://ark.intel.com/content/www/us/en ... 0-ghz.html

Two of these should have 72 threads available.

They would in Linux. There are enterprise levels of Windows that support more than 32. I suspect not the version you are running.

https://www.anandtech.com/show/15483/am ... x-review/3
This page of a thread ripped review is about Windows limits.

As you imagine, this is not an issue many folders have. Linux is one solution, a different version of Windows may be a solution.

Or may be a real expert will have a better idea.
Tsar of all the Rushers
I tried to remain childlike, all I achieved was childish.
A friend to those who want no friends
JimboPalmer
 
Posts: 2335
Joined: Mon Feb 16, 2009 5:12 am
Location: Greenwood MS USA

Re: Issue with multi-socket Intel wildcat in Windows Server

Postby fishy_boy256 » Sat May 08, 2021 8:53 pm

Thanks for the tip about Windows limits. Just to clarify I am running Windows Server 2019 not Windows 10 Home or pro (the article cited doesn't talk about it a lot). I looked into windows server and it is possible that I don't have the correct number of licenses for the number of cores I have but The OS has yet to complain. It still reports all 64 cores and both processors, If I run another intensive program, (a stress test) the cores of both nodes jump to high usage.

I was running Linux before but switched to Windows server when I started folding because I wanted to run VMs and other software that is more compatible with Windows server. I would prefer not to switch back but if that is what must be done so be it.
There are 18 cores per processor but 1 core is reserved for system services and 1 for NUMA applications in both CPUs so only 32 of the 36 threads are usable. I'm sure those cores could be motivated to do work too but in comparison to losing a whole node it seems like a small issue.

I've seen other people having issues with multi-socket systems but none for a computer of this size or on this OS. Once again wondering if anyone has folding experience on a similar machine and how they configured it to run both nodes.
fishy_boy256
 
Posts: 3
Joined: Sat May 08, 2021 3:35 am

Re: Issue with multi-socket Intel wildcat in Windows Server

Postby Joe_H » Sat May 08, 2021 10:21 pm

There are multiple issues with running F@h on Windows with large multi-processor systems. The first as mentioned is having the correct licenses in place as mentioned already. The second is having the right options selected when creating the executable. Without that, in general the process will be limited to 32 CPU threads at a max depending on the system and OS version and license type.

I don't have access to detailed information on what options need to be in place for the executables. There is a list of options used when FAHClient was compiled show in the log, and when a folding core is running, its options are shown in that log section showing the core starting up.

There may also be limits based on whether the code it 32-bit or 64-bit. FAHClient is a 32-bit application, does not need the extra registers and address space to handle its function of downloading, starting, stopping, and uploading WUs. Whether that will cause it to only detect 32 CPU threads I am not certain, it has been a while since I did any reading in that area. The folding cores are 64-bit processes running in their own space and potentially should not have any issues.

What some have done on Windows is create a second CPU slot, configure it to use 32 threads, and run 2 WUs at the same time.

However most persons doing folding on larger systems have loaded Linux and just avoided the issues with Windows entirely.
Image

iMac 2.8 i7 12 GB smp8, Mac Pro 2.8 quad 12 GB smp6
MacBook Pro 2.9 i7 8 GB smp3
Joe_H
Site Admin
 
Posts: 7054
Joined: Tue Apr 21, 2009 5:41 pm
Location: W. MA

Re: Issue with multi-socket Intel wildcat in Windows Server

Postby fishy_boy256 » Sun May 09, 2021 6:20 pm

How can I make sure my options are the correct ones when setting up the executable?
fishy_boy256
 
Posts: 3
Joined: Sat May 08, 2021 3:35 am

Re: Issue with multi-socket Intel wildcat in Windows Server

Postby Joe_H » Sun May 09, 2021 6:42 pm

fishy_boy256 wrote:How can I make sure my options are the correct ones when setting up the executable?


The options I mentioned would be the ones used by the F@h developers when they compiled the code, not something you can change. Both FAHClient and the folding cores are closed source, so you would not be able to recompile.
Joe_H
Site Admin
 
Posts: 7054
Joined: Tue Apr 21, 2009 5:41 pm
Location: W. MA


Return to CPU Projects - released FAHCores _a7 & _a8 (a4 retired)

Who is online

Users browsing this forum: No registered users and 2 guests

cron