Thread 'Has BOINC 7.12.1 broken NVIDIA detection?'

Message boards : GPUs : Has BOINC 7.12.1 broken NVIDIA detection?
Message board moderation

To post messages, you must log in.

AuthorMessage
slre

Send message
Joined: 16 Jul 14
Posts: 7
United Kingdom
Message 87760 - Posted: 22 Aug 2018, 21:05:30 UTC

Just installed BOINC client version 7.12.1 for windows_x86_64.
BOINC can no longer find my two NVIDIA GPUs; both are now reporting as 'missing'.
Restarting both BOINC and windows is ineffective.
Have updated NVIDIA driver to 398.82: still no luck.

System is windows 10 pro build 1803

Seems to be associated with the new BOINC version, but is this just me or a general problem?
ID: 87760 · Report as offensive
Bernie Vine
Volunteer moderator
Avatar

Send message
Joined: 10 Dec 12
Posts: 323
Message 87761 - Posted: 22 Aug 2018, 21:35:37 UTC - in response to Message 87760.  
Last modified: 22 Aug 2018, 22:27:33 UTC

Did you get the "updated" drivers from the Nvidia website?

"Windows" drivers often do not work correctly .

I have 5 PC's running BOINC 7.12.1 and all have working Nvidia cards
ID: 87761 · Report as offensive
Richard Haselgrove
Volunteer tester
Help desk expert

Send message
Joined: 5 Oct 06
Posts: 5129
United Kingdom
Message 87762 - Posted: 22 Aug 2018, 21:44:28 UTC - in response to Message 87760.  

BOINC itself is still detecting NVidia just fine, but as Bernie says it's most likely the incomplete drivers from a Windows 10 update that's causing the problem.
ID: 87762 · Report as offensive
mmonnin

Send message
Joined: 1 Jul 16
Posts: 146
United States
Message 87766 - Posted: 23 Aug 2018, 14:40:11 UTC - in response to Message 87762.  

BOINC itself is still detecting NVidia just fine, but as Bernie says it's most likely the incomplete drivers from a Windows 10 update that's causing the problem.


CUDA should still be installed though.

If this is a 2nd client then GPU detection broke on 2nd clients in 7.10.
ID: 87766 · Report as offensive
Richard Haselgrove
Volunteer tester
Help desk expert

Send message
Joined: 5 Oct 06
Posts: 5129
United Kingdom
Message 87767 - Posted: 23 Aug 2018, 15:25:47 UTC - in response to Message 87766.  

If this is a 2nd client then GPU detection broke on 2nd clients in 7.10.
Since I was responsible for releasing v7.10, I'd be interested in learning more about that. I don't think I've seen it reported before. (I do have a machine with a second client installed for testing, and with dual NVidia cards, so I'll take a look myself. What should I be looking for?)
ID: 87767 · Report as offensive
mmonnin

Send message
Joined: 1 Jul 16
Posts: 146
United States
Message 87769 - Posted: 23 Aug 2018, 16:50:59 UTC - in response to Message 87767.  

If this is a 2nd client then GPU detection broke on 2nd clients in 7.10.
Since I was responsible for releasing v7.10, I'd be interested in learning more about that. I don't think I've seen it reported before. (I do have a machine with a second client installed for testing, and with dual NVidia cards, so I'll take a look myself. What should I be looking for?)


http://boinc.berkeley.edu/dev/forum_thread.php?id=12454&postid=86596#86596

It was an AMD RX580 card but I assume it would apply to both manufactures. I went from 7.6 to 7.10 and the 2nd client saw no GPUs. The 1st instance in the default program data folder saw the GPU. No other changes, just BOINC. Went back to 7.8 and the 2nd client sees the AMD card. I use 7.9 in Ubuntu since its in the repository and they see GPUs on 2nd clients. Installs were on top of the old version. ProgramData folders stayed put during the installs. Upon starting the 2nd client E@H said there were no usable GPU for the tasks that had already been downloaded with 7.6 so nothing ran.
ID: 87769 · Report as offensive
Richard Haselgrove
Volunteer tester
Help desk expert

Send message
Joined: 5 Oct 06
Posts: 5129
United Kingdom
Message 87770 - Posted: 23 Aug 2018, 16:58:09 UTC - in response to Message 87769.  
Last modified: 23 Aug 2018, 17:23:26 UTC

OK, thanks. My test machine is also Win7, so it'll be a fair comparison of NV / AMD.

Edit - and the same problem occurs under v7.12.1 - why?

I'm seeing

23-Aug-2018 18:14:07: CreateProcess failed: 'The system cannot find the file specified. (0x2)'
in the stderrdae.txt file in the second data directory. There was some work in this area prior to v7.10.2 - there was a complaint about problems starting BOINC at the command line without specifying the name as boinc.exe

But I have boinc.exe in my batch file - that's not (exactly) it. But it gives us a place to look.
ID: 87770 · Report as offensive
mmonnin

Send message
Joined: 1 Jul 16
Posts: 146
United States
Message 87771 - Posted: 23 Aug 2018, 17:30:40 UTC - in response to Message 87770.  

OK, thanks. My test machine is also Win7, so it'll be a fair comparison of NV / AMD.

Edit - and the same problem occurs under v7.12.1 - why?

I'm seeing

23-Aug-2018 18:14:07: CreateProcess failed: 'The system cannot find the file specified. (0x2)'
in the stderrdae.txt file in the second data directory. There was some work in this area prior to v7.10.2 - there was a complaint about problems starting BOINC at the command line without specifying the name as boinc.exe

But I have boinc.exe in my batch file - that's not (exactly) it. But it gives us a place to look.


Unless something is changed functionality and bugs should carry over from one version to the next, no?
ID: 87771 · Report as offensive
Richard Haselgrove
Volunteer tester
Help desk expert

Send message
Joined: 5 Oct 06
Posts: 5129
United Kingdom
Message 87772 - Posted: 23 Aug 2018, 17:33:18 UTC - in response to Message 87771.  
Last modified: 23 Aug 2018, 17:43:18 UTC

Indeed. Ironically, it was my own report (but picked up from an email on boinc_alpha) that led to the last change - #2029. Still poking at it.

And I think I've found a workround. As I think I've posted before, I start my second client instance with a batch file:

D:\BOINC\boinc.exe --allow_multiple_clients --redirectio --detach_console --gui_rpc_port 31418 --dir D:\BOINCdata2
That was placed on my desktop.

Note that my BOINC program folder is on hard drive D:, whereas my desktop is on SSD C:

Putting the batch file directly into D:\BOINC\, and starting it from a shortcut on the desktop, restores GPU detection here - could you test, please?

It'll be a working directory problem. Now, how to code that in C++?
ID: 87772 · Report as offensive
slre

Send message
Joined: 16 Jul 14
Posts: 7
United Kingdom
Message 87773 - Posted: 23 Aug 2018, 17:38:59 UTC - in response to Message 87761.  

Did you get the "updated" drivers from the Nvidia website?


Yes; driver update was direct from geforce.com.
ID: 87773 · Report as offensive
slre

Send message
Joined: 16 Jul 14
Posts: 7
United Kingdom
Message 87774 - Posted: 23 Aug 2018, 17:59:48 UTC - in response to Message 87773.  

Did you get the "updated" drivers from the Nvidia website?


Yes; driver update was direct from geforce.com.

... and a clean install.
ID: 87774 · Report as offensive
slre

Send message
Joined: 16 Jul 14
Posts: 7
United Kingdom
Message 87775 - Posted: 23 Aug 2018, 18:40:53 UTC

An update on this ....

Turns out that the problem was not BOINC, Windows or Nvidia.
I run the BOINC client under BoincTasks, a (normally) very nice third party tool that makes it easy to monitor several BOINC computers (I run 2-4 so this is useful). (If anyone is interested, you can get it at https://efmer.com/boinctasks-how-to-start/.

But .... I have also updated BoincTasks; same day as updating BOINC.

Interestingly, if I start the BOINC client from BoincTasks, BOINC does not find my GPUs. But if I start the local instance from BOINC Manager, the BOINC client _does_ find my GPUs. This appears reproducible on my machine.

So for anyone else seeing a similar problem recently; if you're running BoincTasks and have recently updated to v 1.75, maybe that's the issue. In the mean time, try starting BOINC with BOINC's own manager; it may behave better. And starting BoincTasks once the client is runing causes no trouble as long as you don;t stop and restart the client from inside BoincTasks.
ID: 87775 · Report as offensive
Richard Haselgrove
Volunteer tester
Help desk expert

Send message
Joined: 5 Oct 06
Posts: 5129
United Kingdom
Message 87776 - Posted: 23 Aug 2018, 18:53:51 UTC - in response to Message 87775.  

In that case, it could be a side effect of the same problem that mmonnin and I were just discussing. If BoincTasks doesn't set the working directory explicitly, it could trigger the same problem as our second instance batch file launchers. That makes it more important to fix the client, because BoincTasks is a popular tool.
ID: 87776 · Report as offensive
mmonnin

Send message
Joined: 1 Jul 16
Posts: 146
United States
Message 87777 - Posted: 23 Aug 2018, 20:22:04 UTC

I start my additional clients with a batch file. I think there are over 10 clients on this PC for Goofy NCI work so I have the same line with their own port and directory. I wasn't aware BOINCTasks could start clients but I know I am at least 1 version behind on BOINCTasks.
This is from my Guide on multiple clients so it should be the same what's on my PC.

I start the batch from my 2nd ProgramData\BOINC2 folder and the GPU client I think is BOINC6

"c:\Program Files\BOINC\boinc.exe" --allow_multiple_clients --redirectio --detach_console --gui_rpc_port 31418 --dir C:\ProgramData\BOINC2 
exit


I've started the clients with 7.6, 7.8 and 7.10 using the same batch file. Directories are complete without relative paths. Should it be something different?

I'll have to try it again when I get home.
ID: 87777 · Report as offensive
Richard Haselgrove
Volunteer tester
Help desk expert

Send message
Joined: 5 Oct 06
Posts: 5129
United Kingdom
Message 87780 - Posted: 24 Aug 2018, 16:54:51 UTC
Last modified: 24 Aug 2018, 17:19:36 UTC

OK, after other commitments this morning, I've been able to devise a definitive test.

Starting the second client with the batch file on my desktop (i.e. in a different location from the BOINC program files), I get these displays with process explorer.

With BOINC v7.8.3:


That was the one that wouldn't detect GPUs unless you started it with boinc.exe on the command line, instead of just boinc. Otherwise, it worked.

With BOINC v7.12.1:


Notice that the command line starts with just "boinc.exe". That's hard-wired now, but in the process we've lost the previous "D:\BOINC\..." - that's what's causing the problems in this thread. Instead of a getting a fix, we just shifted the deckchairs around a bit. That's enough for me to submit a technical report - I'll edit in the issue number when I've written it.

Submitted: #2657
ID: 87780 · Report as offensive

Message boards : GPUs : Has BOINC 7.12.1 broken NVIDIA detection?

Copyright © 2024 University of California.
Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.