Problem with memory speed on GTX 9xx when running compute software

Graphics / Linux Linux
1

I run Folding@Home compute software on several computers and an assortment of Nvidia graphics cards. Lately on some of the newer released work units there is a problem of the 9xx cards getting errors and not being able to complete the work units. This was only happening on my 3 - GTX 980 Classified cards and a GTX 970 SC I am also running a GTX 770 2 - GTX 680’s and a GTX 580 which are not affected by the problem.

I noticed that when running compute software that the 980’s and 970 default to P2 state which limits the memory to 6000Mhz, so I got curious as to why they were limiting to 6000Mhz when the older cards were not limited, I started playing around with the GPU memory speed on the Windows rigs using Nvidia Inspector and when I slowed the memory down the errors stopped and the cards are able to run the work units with no problem. But for some reason there is no way to adjust the P2 state memory speed in Linux that I can find.

Is it possible to enable it in coolbits interface or is there a nvidia-setting command that can be used to set the memory speed. I am able to set voltage with a nvidia-setting command so I thought perhaps I could post here and somebody could help me out. If there currently is not a way to set memory speed, how would a person go about getting NVidia to enable it in there Linux drivers.

2

I’ll second that.

I’ll add that even though the only editable performance level in X-Server Settings is shown as Level 3, an adjustment to the Graphics Clock Offset also operates when the card switches to Level 2. If not a bug, this is at least a significant inconsistency.

3

Same here: more possibilities for the Linux driver to adjust the characteristics of the hardware is very much appreciated and needed to compensate for error conditions.

4

i fold on 8 GTX980’s and 2 Titan X’s and i really would appreciate if you enabled this. especially the Titan X’s.