[Win 10][AMD GPUs] New WUs don't start on locked session

[therealsummoner]

Hello,

When the client gets a new WU, viewer tells it starts running, shows 0.00%, but then its stuck at this point, until I unlock the PC and log on again.
I use Remote Desktop and also the FaH Viewer remotely. I had this problem also with v7.5.1 and now with two AMD GPUs in this PC and both do the same waiting at 0.00% until i log on. I can see the 0.00% reported in the Viewer remotely.

This is one of the GPUs; the WUs are getting to this point and then everything sits there without even timeouting:

Code: Select all

06:58:17:WU03:FS02:Requesting new work unit for slot 02: READY gpu:1:Tahiti XT [Radeon R9 200/HD 7900/8970] from 128.252.203.10
06:58:17:WU03:FS02:Connecting to 128.252.203.10:8080
06:59:09:WU03:FS02:Downloading 29.59MiB
06:59:15:WU03:FS02:Download 4.22%
06:59:21:WU03:FS02:Download 24.08%
06:59:27:WU03:FS02:Download 55.34%
06:59:33:WU03:FS02:Download 68.01%
06:59:39:WU03:FS02:Download 91.03%
06:59:40:WU03:FS02:Download complete
06:59:40:WU03:FS02:Received Unit: id:03 state:DOWNLOAD error:NO_ERROR project:11761 run:0 clone:626 gen:45 core:0x22 unit:0x0000004f80fccb0a5e6d7d21e013ed3f
06:59:40:WU03:FS02:Starting
06:59:40:WU03:FS02:Running FahCore: "C:\Program Files (x86)\FAHClient/FAHCoreWrapper.exe" C:\Users\*****\AppData\Roaming\FAHClient\cores/cores.foldingathome.org/v7/win/64bit/Core_22.fah/FahCore_22.exe -dir 03 -suffix 01 -version 706 -lifeline 4792 -checkpoint 15 -gpu-vendor amd -opencl-platform 0 -opencl-device 0 -gpu 0
06:59:40:WU03:FS02:Started FahCore on PID 12008
06:59:40:WU03:FS02:Core PID:4064
06:59:40:WU03:FS02:FahCore 0x22 started
06:59:41:WU03:FS02:0x22:*********************** Log Started 2020-04-23T06:59:41Z ***********************
06:59:41:WU03:FS02:0x22:*************************** Core22 Folding@home Core ***************************
06:59:41:WU03:FS02:0x22:       Type: 0x22
06:59:41:WU03:FS02:0x22:       Core: Core22
06:59:41:WU03:FS02:0x22:    Website: https://foldingathome.org/
06:59:41:WU03:FS02:0x22:  Copyright: (c) 2009-2018 foldingathome.org
06:59:41:WU03:FS02:0x22:     Author: John Chodera <[email protected]> and Rafal Wiewiora
06:59:41:WU03:FS02:0x22:             <[email protected]>
06:59:41:WU03:FS02:0x22:       Args: -dir 03 -suffix 01 -version 706 -lifeline 12008 -checkpoint 15
06:59:41:WU03:FS02:0x22:             -gpu-vendor amd -opencl-platform 0 -opencl-device 0 -gpu 0
06:59:41:WU03:FS02:0x22:     Config: <none>
06:59:41:WU03:FS02:0x22:************************************ Build *************************************
06:59:41:WU03:FS02:0x22:    Version: 0.0.2
06:59:41:WU03:FS02:0x22:       Date: Dec 6 2019
06:59:41:WU03:FS02:0x22:       Time: 21:30:31
06:59:41:WU03:FS02:0x22: Repository: Git
06:59:41:WU03:FS02:0x22:   Revision: abeb39247cc72df5af0f63723edafadb23d5dfbe
06:59:41:WU03:FS02:0x22:     Branch: HEAD
06:59:41:WU03:FS02:0x22:   Compiler: Visual C++ 2008
06:59:41:WU03:FS02:0x22:    Options: /TP /nologo /EHa /wd4297 /wd4103 /Ox /MT
06:59:41:WU03:FS02:0x22:   Platform: win32 10
06:59:41:WU03:FS02:0x22:       Bits: 64
06:59:41:WU03:FS02:0x22:       Mode: Release
06:59:41:WU03:FS02:0x22:************************************ System ************************************
06:59:41:WU03:FS02:0x22:        CPU: AMD Phenom(tm) II X6 1100T Processor
06:59:41:WU03:FS02:0x22:     CPU ID: AuthenticAMD Family 16 Model 10 Stepping 0
06:59:41:WU03:FS02:0x22:       CPUs: 6
06:59:41:WU03:FS02:0x22:     Memory: 8.00GiB
06:59:41:WU03:FS02:0x22:Free Memory: 4.67GiB
06:59:41:WU03:FS02:0x22:    Threads: WINDOWS_THREADS
06:59:41:WU03:FS02:0x22: OS Version: 6.2
06:59:41:WU03:FS02:0x22:Has Battery: false
06:59:41:WU03:FS02:0x22: On Battery: false
06:59:41:WU03:FS02:0x22: UTC Offset: 2
06:59:41:WU03:FS02:0x22:        PID: 4064
06:59:41:WU03:FS02:0x22:        CWD: C:\Users\tamas\AppData\Roaming\FAHClient\work
06:59:41:WU03:FS02:0x22:         OS: Windows 10 Pro
06:59:41:WU03:FS02:0x22:    OS Arch: AMD64
06:59:41:WU03:FS02:0x22:********************************************************************************
06:59:41:WU03:FS02:0x22:Project: 11761 (Run 0, Clone 626, Gen 45)
06:59:41:WU03:FS02:0x22:Unit: 0x0000004f80fccb0a5e6d7d21e013ed3f
06:59:41:WU03:FS02:0x22:Reading tar file core.xml
06:59:41:WU03:FS02:0x22:Reading tar file integrator.xml
06:59:41:WU03:FS02:0x22:Reading tar file state.xml
06:59:41:WU03:FS02:0x22:Reading tar file system.xml
06:59:42:WU03:FS02:0x22:Digital signatures verified
06:59:42:WU03:FS02:0x22:Folding@home GPU Core22 Folding@home Core
06:59:42:WU03:FS02:0x22:Version 0.0.2

After unlocking the station (logon remotely or even locally when i'm at the PC) it starts everything to work normally and then i can logout (lock station) and leave them running:

Code: Select all

07:18:33:WU03:FS02:0x22:Completed 0 out of 2000000 steps (0%)
07:18:33:WU03:FS02:0x22:Temperature control disabled. Requirements: single Nvidia GPU, tmax must be < 110 and twait >= 900
07:21:37:WU03:FS02:0x22:Completed 20000 out of 2000000 steps (1%)
07:24:25:WU03:FS02:0x22:Completed 40000 out of 2000000 steps (2%)
07:27:26:WU03:FS02:0x22:Completed 60000 out of 2000000 steps (3%)
07:30:17:WU03:FS02:0x22:Completed 80000 out of 2000000 steps (4%)

GPU drivers i tried were 20.3.1 and 20.4.1. The GPU clocks are already corrected by me so, that both completes WUs correctly, because both cards are factory overclocked thus needed to take back on the clocks a bit.

Do you have any suggestion i could try here?

Thanx in advance!

[therealsummoner]

Update:
Something similar happens with the station locked and pausing a WU remotely from within the Viewer, then waiting some minutes, then starting it again.
The only difference i see in this case is, that the Viewer continues counting the work progress, but if i log on back to the PC (unlocking station), the GPU load is 0 in this time interval (running GPU-Z for both cards in the background to see this), so there is no actual computing occurring.
Locally made (unlocked session) works everything as intended.

[PantherX]

Welcome to the F@H Forum therealsummoner,

Please note that using RDP will kill any GPU folding. Also, I think that locking the system might cause the GPU to stop folding as the GPU can only be used if the user is logged in.

[therealsummoner]

Well this is strange, because my other PC with two nVidia GPUs is working without problem, left locked all the time.
The PC with AMD GPUs are also working, once i logged on via RDP and logged off.
All these i can follow on GPU-Z history or HWinfo.
Then maybe it is a thing which should not work at all, but somehow it works for me on nVidia and partially as wrote before on AMD as well?

[PantherX]

It could be a driver implementation that varies between Nvidia and AMD when it comes to locked screens. I don't know put that's an interesting observation.

If you log GPU-Z to a text file, see what happens to the frequencies and power state. That might shed some light in your case.

[therealsummoner]

Can be true, because when connecting to the station, the clocks are jumping up on at least one GPU, ofc because RDP should be HW accelerated and uses GPU.
I'll try maybe some fixed clocks or similar things.

[Jfb_tyne]

Hi, I'm having this exact same problem:

[*]Progress on started WU when left unattended[/*]
[*]Client uploads completed then downloads new WU[/*]
[*]New WU fails to start, stuck at 0.00% until I log in (either locally or RDP)[/*]
[*]No error messages in log[/*]

System is Win 10, i5, AMD RX470

I've tried various Windows power settings, uninstalling and reinstalling client and drivers. Also I read that disabling ULPS might help, but that was on old threads and it didn't.

Let me know if fixing clocks improves the situation, thanks.

[therealsummoner]

Well i cannot apply fixed clocks on neither the 290x nor the 7970.
Maybe it is possible to do this on newer cards only(?)

[therealsummoner]

Ok there is a not so elegant method, but it works.
Instead of disconnecting from the RDP session, you double-click a batch file, which then leaves the session opened and disconnects you.

Create a batch file with this:

Code: Select all

for /f "tokens=3-4" %%a in ('query session %username%') do @if "%%b"=="Active" set RDP_SESSION=%%a
tscon %RDP_SESSION% /dest:console

Then create a shortcut for the batch file and in the shortcut properties check the "run as administrator" option (Advanced button).
So then instead of closing the session, you just double-click this shortcut and its all done.
Cheers!

[Jfb_tyne]

Thanks!

I tried fixing the clocks of State 0 (300 MHz) to State 1 (466 MHz) and that didn't work.

[therealsummoner]

The P-state 0 is normally not adjustable. The only way to step the GPU more up to higher states is to force it with some tools, or mod the Bios, which i just wouldn't like.
This hack above i found is still a "nicer" workaround imho.

[ajm]

Well this is strange, because my other PC with two nVidia GPUs is working without problem, left locked all the time.
The PC with AMD GPUs are also working, once i logged on via RDP and logged off.
All these i can follow on GPU-Z history or HWinfo.
Then maybe it is a thing which should not work at all, but somehow it works for me on nVidia and partially as wrote before on AMD as well?

Just a word to confirm that Remote Desktop Connection doesn't affect the GPU WUs on my setup either. I had noticed that before already, but I was unsure as several members of the forum stated that RDC would kill GPU folding and my remote controlled machine was then on Linux for some testing. Today, I revert it to windows (10 Pro) and I can confirm that RDC doesn't hurt folding. I'm seeing now through RDC the CPU and both GPUs (one AMD and one Nvidia) of that machine happily folding while I'm installing FAH in Ubuntu running there on top of WSL 2 from my main PC (also Win 10 Pro).

[therealsummoner]

Well this is strange, because my other PC with two nVidia GPUs is working without problem, left locked all the time.
The PC with AMD GPUs are also working, once i logged on via RDP and logged off.
All these i can follow on GPU-Z history or HWinfo.
Then maybe it is a thing which should not work at all, but somehow it works for me on nVidia and partially as wrote before on AMD as well?

Just a word to confirm that Remote Desktop Connection doesn't affect the GPU WUs on my setup either. I had noticed that before already, but I was unsure as several members of the forum stated that RDC would kill GPU folding and my remote controlled machine was then on Linux for some testing. Today, I revert it to windows (10 Pro) and I can confirm that RDC doesn't hurt folding. I'm seeing now through RDC the CPU and both GPUs (one AMD and one Nvidia) of that machine happily folding while I'm installing FAH in Ubuntu running there on top of WSL 2 from my main PC (also Win 10 Pro).

I only have this on a pure AMD GPU PC. I also tried a mixed one, but i was not happy with the mixed nVidia and Amd drivers/tools, there was always some problem there. Now on the pure nVidia PC (after a fresh Win10pro install) everything is flawless.

[ajm]

Yes, I can confirm that too.
But it was nice to have both on one machine for my tests. And it did work well on Ubuntu 18.04, eventually. I'll try tomorrow to rebuild that setup on an Ubuntu running on top of WSL 2. Wish me luck...

[therealsummoner]

Well i just witnessed that my method is causing the main GPU which have a display connected to drop its clock to 300MHz (P-State 0) and as it looks, it stays there for some reason :O This can be healed only via a reboot.
Strange happening!