GPU Load Drop - Normal?

[fxyefx]

I just recently got into some GPU folding for the first time. I'm running 7.1.52 on an AMD 7850 using the GPU.txt whitelist, just the GPU slot going, no SMP. The monitor for the GPU load is showing a periodic drop from regular 100% load to 30-70% about every 25-30 seconds. Is this just normal load for folding? Does anyone else get this pattern, or is my card doing some kind of throttling/my setup being wacky? Thanks!

[iceman1992]

Do you use your computer for anything while folding? My GTX260 GPU usage also drops periodically when I'm browsing, probably the hardware acceleration or something.

[fxyefx]

Nice idea, but doesn't seem to be the case for me. I'm often browsing while folding and wouldn't be surprised to see that hit the GPU load if I run into a bunch of Flash or WebGL. I still observed the pattern after closing my browser, though, and even after a fresh reboot with nothing opened before the FAH manager.

[bruce]

What are the time-intervals between the load-drops?

The computations are never quite uniform. There will be checkpoint events which have to wait for the disk to store data. There will be end-of-step events when the type of processing being done changes briefly. There will be end-of-kernel events when the FahCore will pass data to/from main RAM and VRAM. There may be other such events when processing changes and something has to wait on something else.

I wouldn't worry about it. There's a wide range of "normal" depending both on the particular Project and the particular type of GPU doing the processing.

[fxyefx]

What are the time-intervals between the load-drops?

They occur once every 27-28 seconds, like clockwork. Now that you mention it, they actually seem to line up with the completion of each 0.1% in work unit progress! (project 11293)

The computations are never quite uniform. There will be checkpoint events which have to wait for the disk to store data. There will be end-of-step events when the type of processing being done changes briefly. There will be end-of-kernel events when the FahCore will pass data to/from main RAM and VRAM. There may be other such events when processing changes and something has to wait on something else.
I wouldn't worry about it. There's a wide range of "normal" depending both on the particular Project and the particular type of GPU doing the processing.

Thanks for the explanation! Sounds like it's almost certainly something to do with the nature of the work load, rather than my card getting ready to explode. My mind is now at ease.