Welcome to MilkyWay@home

Nvidia GPU compute client only doing 1/5 of the required work

Message boards : Number crunching : Nvidia GPU compute client only doing 1/5 of the required work
Message board moderation

To post messages, you must log in.

AuthorMessage
Jesse Viviano

Send message
Joined: 4 Feb 11
Posts: 86
Credit: 60,913,150
RAC: 0
Message 66261 - Posted: 2 Apr 2017, 17:52:16 UTC

I recently had to shut off GPU work for this project in my preferences because the Nvidia GPU client was only doing the final work unit in the now-standard 5 work unit bundle, causing my work to be declared as validate errors. The client was working fine until April 1 (No, this is no April fools joke.) My GPU is still generating valid work for other projects. Is anyone else having a similar problem?
ID: 66261 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Jesse Viviano

Send message
Joined: 4 Feb 11
Posts: 86
Credit: 60,913,150
RAC: 0
Message 66262 - Posted: 2 Apr 2017, 17:58:41 UTC

I also noticed that I am using the SSE4.1 path when my CPU, a Xeon E5-2690 v3, is capable of using AVX. In the work units where the GPU client failed, those who did the cleanup work after my computer messed up were using an AVX path. Could something be wrong with the SSE4.1 code path?
ID: 66262 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Brickhead
Avatar

Send message
Joined: 20 Mar 08
Posts: 108
Credit: 2,607,924,860
RAC: 0
Message 66263 - Posted: 2 Apr 2017, 19:51:51 UTC

I'd venture a guess at Windows 10 auto-updating drivers. Have a look at this thread:

https://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=4101
ID: 66263 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Jesse Viviano

Send message
Joined: 4 Feb 11
Posts: 86
Credit: 60,913,150
RAC: 0
Message 66266 - Posted: 3 Apr 2017, 3:52:17 UTC

I reinstalled my video driver using a clean install, and that seems to have fixed the problem.
ID: 66266 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Jesse Viviano

Send message
Joined: 4 Feb 11
Posts: 86
Credit: 60,913,150
RAC: 0
Message 66267 - Posted: 3 Apr 2017, 4:40:10 UTC
Last modified: 3 Apr 2017, 4:44:46 UTC

The problem is back. I have had to disable GPU work for MilkyWay@home for now.
ID: 66267 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 8 May 09
Posts: 3315
Credit: 519,939,839
RAC: 22,823
Message 66268 - Posted: 3 Apr 2017, 11:07:13 UTC - in response to Message 66267.  

The problem is back. I have had to disable GPU work for MilkyWay@home for now.


Did Windows reinstall it's own drivers again?
ID: 66268 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Jesse Viviano

Send message
Joined: 4 Feb 11
Posts: 86
Credit: 60,913,150
RAC: 0
Message 66269 - Posted: 3 Apr 2017, 12:14:45 UTC - in response to Message 66268.  

There is no sign that Windows 10 reinstalled any driver. Since other projects' GPU work still is being turned in and successfully validating, I would think that the problem is either a bug exists in the driver, or a new optimization in the driver exposed a previously undetected bug in the GPGPU software client (the latter has happened with Folding@home).
ID: 66269 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Brickhead
Avatar

Send message
Joined: 20 Mar 08
Posts: 108
Credit: 2,607,924,860
RAC: 0
Message 66270 - Posted: 3 Apr 2017, 13:37:02 UTC

However it got there, you are using a brand new nvidia driver that is known to have caused failures in MW.

http://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=4101&postid=66257
ID: 66270 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 8 May 09
Posts: 3315
Credit: 519,939,839
RAC: 22,823
Message 66273 - Posted: 4 Apr 2017, 17:46:22 UTC - in response to Message 66270.  

However it got there, you are using a brand new nvidia driver that is known to have caused failures in MW.

http://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=4101&postid=66257


I'm using the same driver here and it seems to be working just fine for my Nvidia 750Ti gpu's. Now I JUST put them on today but so far no bad workunits for either gpu. I am knocking on wood!!!
ID: 66273 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Jesse Viviano

Send message
Joined: 4 Feb 11
Posts: 86
Credit: 60,913,150
RAC: 0
Message 66285 - Posted: 9 Apr 2017, 20:22:26 UTC

I just filed a bug report with Nvidia over this issue.
ID: 66285 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Jesse Viviano

Send message
Joined: 4 Feb 11
Posts: 86
Credit: 60,913,150
RAC: 0
Message 66374 - Posted: 5 May 2017, 7:33:05 UTC

Either a new driver or the new application version fixed the bug.
ID: 66374 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Message boards : Number crunching : Nvidia GPU compute client only doing 1/5 of the required work

©2024 Astroinformatics Group