GTX 980 clock speed

[billford]

Not a problem (afaict), just an oddity than piqued my curiosity.

I have two (nominally) identical GTX 980s in more-or-less identical PCs, same OS (Mint 17.1), same drivers (346.35), same overclock (125MHz), same preferred mode (Auto => Adaptive).

Happened to notice the other day that they were running at different clock speeds… paid a bit more attention and it seems to be that with a P9201 they run at 1440MHz, with a P9411 they run at 1466MHz. (They don't get any others)

Is this

a) normal
b) down to the drivers (a consequence of the Adaptive mode?) or
c) something determined by the project?


edit- forget that, one of them is now running a P9411 at 1440MHz so it must be something else

[Kjetil]

It is normal. My 980/970 do the same: this pc 2x980 evga sc gpu 0 1441 and gpu 1 1429. PS running win7

[billford]

I thought it most probably was normal, but confirmation is good- thanks

[Rel25917]

Probably adjusting their clock speed to maintain their target power level and or temp.(assuming that maxwell runs the same as keplar)

[Sailer]

I have noticed the same speed variance on my GTX970 and GTX980. The bigger problem that I notice is that except for the rare times they pick up a core 17, their folding PPD is abysmal. I'm not sure if this is really a driver problem or there has been a change in the programing from Stanford. Its a question that troubles me; why do older core 17s work, but newer ones don't?

[Joe_H]

I have noticed the same speed variance on my GTX970 and GTX980. The bigger problem that I notice is that except for the rare times they pick up a core 17, their folding PPD is abysmal. I'm not sure if this is really a driver problem or there has been a change in the programing from Stanford. Its a question that troubles me; why do older core 17s work, but newer ones don't?

The newer core relied on an OpenCL function that nVidia did not properly support in the Maxwell drivers, so the core had a workaround added so Maxwell's could fold at all. But the workaround is not as fast as using the OpenCL function. Supposedly this has been fixed in the latest drivers from nVidia, version 346 for Linux and 347 for Windows.

[wilding2004]

I have noticed the same speed variance on my GTX970 and GTX980. The bigger problem that I notice is that except for the rare times they pick up a core 17, their folding PPD is abysmal. I'm not sure if this is really a driver problem or there has been a change in the programing from Stanford. Its a question that troubles me; why do older core 17s work, but newer ones don't?

The newer core relied on an OpenCL function that nVidia did not properly support in the Maxwell drivers, so the core had a workaround added so Maxwell's could fold at all. But the workaround is not as fast as using the OpenCL function. Supposedly this has been fixed in the latest drivers from nVidia, version 346 for Linux and 347 for Windows.

Yep, the newest drivers work well for me. Both my 960's and 970's now produce roughly the same PPD with core_17 or core_18. Obviously there are some difference between projects, but if you average out the PPD over all WU's, there is very little difference. The 970's appear to do a little better on core_17, while the 960's appear to do better with core_18. Again only small differences.

[Sailer]

I have the 347.52 driver on my GTX980 and installed the 347.88 driver on my GTX970 yesterday. So far, the GTX970 has only picked up Core 15s and getting about 64k PPD, which is less PPD than when I used a GTX770. The GTX980 will get about 315k PPD when it gets a 17, but drops to about 65k PPD with a core 15. As it is, the top performers for me are GTX780 Ti cards which consistently get about 300k PPD. As far as I can tell, the GTX980 has never picked up a core 18, but that could merely be luck of the draw.

[wilding2004]

I have the 347.52 driver on my GTX980 and installed the 347.88 driver on my GTX970 yesterday. So far, the GTX970 has only picked up Core 15s and getting about 64k PPD, which is less PPD than when I used a GTX770. The GTX980 will get about 315k PPD when it gets a 17, but drops to about 65k PPD with a core 15. As it is, the top performers for me are GTX780 Ti cards which consistently get about 300k PPD. As far as I can tell, the GTX980 has never picked up a core 18, but that could merely be luck of the draw.

Yeah, core_15 is always going to look bad against 17 and 18, as it doesn't qualify for QRB. I've had a couple of core_18 WU's produce over 300kPPD on the 970's