Forum > GPU crunching
Latest drivers (NVidia and ATI)
Fredericx51:
I still have ocasionell system-freeze, still use 266.58 and run LUNATICs, used the v0.37 installer,
host is an Q6600, stock(+),
this host.
Also run Einstein and a few CPU only, Projects. Since at 'hangs' i.s.o. BSOD, no useble 'log' or other
message.
Happens, once or twice a week, maybe heavy outside-temps, could be a reason, which I doubt.
I use cold air, atleast at night, which changes from 2-4C to 20C in 2 till 6 hours?!
I've seen the Q6600 with a 104C temp reading, but a sysinfo which reported 94C, still too high.
Time to reseat the Cooler. (A push/pull Coolermaster.)
Pepi:
New drivers are out: 275.33!
Initial benchmark show at least 20% increase in speed on my 560Ti!
It is very good news!
Jason G:
Looking good, with minor speedups & improved 'feel' during full crunch with test builds here too, on two different systems:
-> Win7x64 Pro/sp1 , GTX 480
-> WinXPx86 Pro/sp3 , GTX 260
I'm unclear as to whether nVidia may have robustified the new drivers to abuse from applications yet (might have to try break them ;) ) . Here's hoping those problems slip quietly into history, even though our test app avoids triggering the issues, it would be a lot less painful for those who crunch other Cuda-enabled Boinc projects, or prefer to run stock.
Jason
Richard Haselgrove:
Message from your messenger pidgeon: is there any way we can help to work out if the new drivers do help with the down-clock problem? If that can be established, I can start doing the rounds again....
Jason G:
--- Quote from: Richard Haselgrove on 02 Jun 2011, 10:04:57 am ---Message from your messenger pidgeon: is there any way we can help to work out if the new drivers do help with the down-clock problem? If that can be established, I can start doing the rounds again....
--- End quote ---
Hmm, running stock 6.10 or x32f in bench runs, repeatedly killing the processes via task manager should trigger it on effected systems if the driver remains sensitive to the bodgy cuda app exits. I'm reasonably satisfied that x38b is viable for Beta soon to prove the V7 code, so I'll try a round of knabench abuse here with that & earlier known more abusive variants.
Feel free to try likewise if your 470 ever was sensitive (repeatably). Snoozing/unsnoozing Boinc with any pre-x37 series app, or another project Cuda app should also trigger 'sticky downclock' if still susceptible to it under the new drivers. x37-x38 series are somewhat hardened against it, though should still be possible through task manager abuse ( not recommended for live Boinc installations of course)
[Edit:] x32f still triggers sticky downclock on killing with task manager here... So don't do that unless you really want.
Rebooting & abandoning hope that drivers can make up for evil code ::)
Jason
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version