Since the problem was reproduced with the nviDia sample code as well, there must be something fishy…
Can some1 from NVIDIA comment on this please?
btw,
We recently participated in a roadshow where we showcased our financial algos on a multi-TESLA10-GPU platform. We were launching our kernels again and again and reporting speedups on the screen. The app ran for more than 2 hours without showing any bugs or slow-downs. Things were normal. But we did not use any driver API or contexts…
I actually just upgraded to 180.22 on Linux (released a month ago, newest I know of), and this problem doesn’t seem to be present anymore. I will do more intensive testing overnight. Also, others have mentioned monkeying around with pushing and popping contexts… I will be adding this and will post here if the issue reappears.
I can confirm this appears to be fixed in 181.20 on Windows XP - but that’s nothing new - someone above already stated XP was unaffected - this was a potentially a Vista issue.
If/when I get the time I’ll boot back into Vista and give it a shot.
Well, this behavior was quite evident on Linux under driver 180.06. That’s the reason I went looking for a thread like this and posted in it. :) It might be an unrelated bug, but the results were the same. As I said earlier, 180.22 fixed whatever the issue was, it seems.
This topic seems to have been resolved, at least somewhat, but I wanted to add that I have been experiencing this problem on our Mac Pro w/ Leopard. I have a kernel that normally runs in about 40 ms, but after a period of time, it shoots up to 100 ms or so. I log in w/ SSH and run everything in a console, and no one else was using the computer. What I discovered is that after 15 minutes or so, an OpenGL screen saver was coming on, which I could see slowing things down like what I saw. If I VNCed into the computer, it disabled the screen saver, and all was good again (of course I went through days of remotely rebooting every 30 minutes before discovering this). So, now I have disabled the screen saver, and it seems to be working again. Hopefully this helps other people who might be experiencing slow-downs like this!