It seems that a lot of GPU problems revolve around specific versions of drivers. Though NVidia has their own support structure, you can often learn from information reported by others who fold.
neognomic wrote:Well, I had version 17 on windoz all this AM and based on your response just started the FAH-GPU on Linux (driver 375.26) but servers are still pushing version 17
No, there's no "PUSHING" involved. You already have 17 and the project did not FORCE the update. I doubt that the log shows a new core being downloaded.
What happens if you PAUSE any GPU that's running and you rename (or delete if you trust it can be recovered from the trash)
FAHCore_21.exe before resuming. Is a new core downloaded? If so, what version?
03:21:45:WU01:FS02:0x21:ERROR:126: Bad platformId size.
This says that the GPU was not properly configured. Problems of runs containing that error have nothing to do with the version of the core.
09:44:56:WARNING:WU02:FS01:FahCore returned: CORE_OUTDATED (110 = 0x6e)
09:44:57:WU02:FS01:Downloading core from http://fahwebx.stanford.edu/cores/Win32/AMD64/NVIDIA/Fermi/Core_21.fah
09:44:57:WU02:FS01:Connecting to fahwebx.stanford.edu:80
09:44:57:WU02:FS01:FahCore 21: Downloading 3.47MiB
09:44:59:WU02:FS01:FahCore 21: Download complete
09:44:59:WU02:FS01:Valid core signature
09:44:59:WU02:FS01:Unpacked 11.80MiB to cores/fahwebx.stanford.edu/cores/Win32/AMD64/NVIDIA/Fermi/Core_21.fah/FahCore_21.exe
09:44:59:WU02:FS01:Starting
09:44:59:WU02:FS01:Running FahCore: "J:\Program Files (x86)\FAHClient/FAHCoreWrapper.exe" J:/Users/rsett/AppData/Roaming/FAHClient/cores/fahwebx.stanford.edu/cores/Win32/AMD64/NVIDIA/Fermi/Core_21.fah/FahCore_21.exe -dir 02 -suffix 01 -version 704 -lifeline 7444 -checkpoint 15 -gpu 0 -gpu-vendor nvidia
09:44:59:WU02:FS01:Started FahCore on PID 11012
09:45:00:WU02:FS01:Core PID:10204
09:45:00:WU02:FS01:FahCore 0x21 started
09:45:01:WU01:FS01:Upload 39.50%
09:45:01:WU02:FS01:0x21:*********************** Log Started 2017-02-02T09:45:00Z ***********************
09:45:01:WU02:FS01:0x21:Project: 9415 (Run 54, Clone 0, Gen 7)
09:45:01:WU02:FS01:0x21:Unit: 0x00000007ab436c9d585e06c93e654ab6
09:45:01:WU02:FS01:0x21:CPU: 0x00000000000000000000000000000000
09:45:01:WU02:FS01:0x21:Machine: 1
09:45:01:WU02:FS01:0x21:Reading tar file core.xml
09:45:01:WU02:FS01:0x21:Reading tar file integrator.xml
09:45:01:WU02:FS01:0x21:Reading tar file state.xml
09:45:01:WU02:FS01:0x21:Reading tar file system.xml
09:45:01:WU02:FS01:0x21:Digital signatures verified
09:45:01:WU02:FS01:0x21:Folding@home GPU Core21 Folding@home Core
09:45:01:WU02:FS01:0x21:Version 0.0.18
I had a hand full of version 17 work units that surprisingly folded just fine on driver version 376.33 just before I got version 18 for the first time. Since then, I've folded everything without failures. I think most people should be past the transition period by now, environmental issues aside. The question I have is if Nvidia is going to tear out the workaround right away in the next release, and if they'll bother mentioning so in the release notes when they do.
neognomic wrote:Well, I had version 17 on windoz all this AM and based on your response just started the FAH-GPU on Linux (driver 375.26) but servers are still pushing version 17
No, there's no "PUSHING" involved. You already have 17 and the project did not FORCE the update. I doubt that the log shows a new core being downloaded.
What happens if you PAUSE any GPU that's running and you rename (or delete if you trust it can be recovered from the trash)
FAHCore_21.exe before resuming. Is a new core downloaded? If so, what version?
03:21:45:WU01:FS02:0x21:ERROR:126: Bad platformId size.
This says that the GPU was not properly configured. Problems of runs containing that error have nothing to do with the version of the core.
Post the top of the log.
No.
I have spent too much time here and on this issue. ...sadly, to no avail.
The solution is simple
Packages (1) foldingathome-7.4.4-3
Total Removed Size: 9.11 MiB
:: Do you want to remove these packages? [Y/n] Y
:: Processing package changes...
(1/1) removing foldingathome [########################] 100%
It's 11:06:29 CST (UTC-0600) on Sun Feb 05, week 06 in 2017.
Thanks anyway.
Supermicro X8DTH-iF: Dual X5687; 24GB; GTX1060-6G; Corsair RM850X; ... Linux 4.9,x; MS Windows 8.1.
Linux Registered User 212962; Still Folding with Team #12501 ...
I started a work unit on my GTX 1060 with 376.48, TPF was 2:11. Just installed 378.49 and now the TPF averages 2:10 to 2:12. Seems on par to me. I'm keeping it.
SombraGuerrero wrote:So you're getting 0x21 WUs and folding cleanly? If so, awesome! This is exactly why we've needed more input from Linux users in this thread.
Sorry for the late reply. But Yes, i get 0x21 WUs and folding without errors.
No problem. Back before the core was updated, I had observed a couple 0x21 projects that were unaffected by the bug, but my computer only took those once in a blue moon. Now, it will be interesting to see how things turn out once Nvidia takes out the workaround app profile.
The nvidia driver 378.66 interface opencl32.dll is version 2.0 but that does not mean the runtime is fully working for 2.0 yet.
Currently folding uses OpenCL 1.1 on NVidia and OpenCL 1.2 on AMD.
I heard the next FahCore will use CUDA for Nvidia GPUs again.
I also see no indication in the release notes that the app profile workaround has been removed yet, so I have to wonder if this set is truly optimal again.
Just tried installing 378.66 and it doesn't work at all for me on Win 10 Home 64-bit with GeForce GTX 1070. Several work units downloaded and failed immediately. Have now reverted to 378.49 which has been working for five days with no problems.
UPDATE: For information, in case others have this issue. Today (22 May) tried 382.05 and the same issue of immediately failing WUs arose. However, as bruce has posted elsewhere, this time I stopped folding and deleted FahCore_21.exe from the ‘C:\Users\<username>\AppData\Roaming\FAHClient\cores\fahwebx.stanford.edu\cores\Win32\AMD64\NVIDIA\Fermi\Core_21.fah’ folder which causes a new FahCore_21.exe to be downloaded when a new WU starts. I am now successfully folding a Core 21 WU for Project 9175 using 382.05.
Last edited by STFC9F22 on Mon May 22, 2017 1:14 pm, edited 1 time in total.
376.48 was working great, then recently stalled gpu's, now testing 378.66 so far so good. Hopefully Nvidia will update, 'open cl' soon, like before the next decade
I've been running 378.66 for 36 hours and while it is stable, I'm still getting ~30% lower PPD than I did with 372.70. This is on 5 machines with 960,970,980's.