Nvidia 295.73 Driver

No issues with 295.20 on Linux here (S@H and E@H). None with previous versions, too. I'm running two computers with GPUs; one doesn't have a monitor connected (and no X is running), on other I'm using DVI and start X only on demand. Works also if I unplug the monitor. CUDA works like clockwork all the time and is unaffected by that. Looks like the Linux driver is properly made - GPU can be utilized just as a co-processor and not also display device.
Of course, I'm using the "nvidia-smi -pm 1" (persistence mode) setting to keep kernel driver loaded even with no X running.
____________

I tried the 295.73 version on two of my rigs - one with GTX480 and the other running GTX260. As soon the system restarted after the installation, I noticed a sudden drop in GPU temperatures, even though the GPU usage was being indicated at 99%. Later I found out that the processing time for CUDA MB had increased from around 10 minutes to about 45 minutes. Seems the newer version isnt doing a good job during the crunching. Reverted back to 285.58 and the rig is back to its normal crunching time.

Just an observation, not sure if it is isolated case or others out there have observed the same.
______________

I tried the 295.73 version on two of my rigs - one with GTX480 and the other running GTX260. As soon the system restarted after the installation, I noticed a sudden drop in GPU temperatures, even though the GPU usage was being indicated at 99%. Later I found out that the processing time for CUDA MB had increased from around 10 minutes to about 45 minutes. Seems the newer version isnt doing a good job during the crunching.

Win7 64-bit with this recently downloaded driver seems to yield about a 20% "computation error" on Cuda-Fermi jobs.

Anyone else seeing this?
FWIW this is on two systems. One with a GTX-560ti and one with a GTX-465.
My WinXP-64 seems OK with this version seems basically OK -- 2xGTX-450s SLI.

Processing times are longer. GPUs (EVGA precision monitoring & OC) are showing noticeably cooler.

Tempted to revert to a clean install of 285.62. Thoughts?

Don't have the monitors set to sleep. Win7 power setting set to Max Perf. ASUS AI Suite set to "rocket" mode. None of that changed, just the new nVidia driver.
Missed the release of 290.53... Usually don't install from the site, normally save to RAID and install all of the Win7 PCs from there.
Thanks, though Arkayn.

Couldn't stand it... Went back to 285.62 on one PC. 10 Cuda-Fermi jobs complete. No computation errors. Estimated job times dropped from ~22 minutes to ~6:10. GPU utilisation is back up. The temps are pushing the old fan profile to blow air again.

When I get a few minutes the rest of the boxes are switching back, too.

295.73 definately caused issues for me on daul gtx 570s. When power save settings turned off the displays, computation errors occurred. Reverted to 285.62 problem went away. Went back to 295.73, same issue again. Changed display power off to never and screen saver to none, turned the monitors off manually and ran clean over night for sedti and milkyway.

Was a bear to figure out because webroot started causing issues with boinc/milkyway on the same day I added the second 570, updated windows and updated the nividia drivers. Webroot was wrecking Windows responses and locking Boinc. Both run fine without the other. At first I thought it was new card. Only figured it out by going back to a restore point. For a little while, all was cool and after a little while lock ups started again. Traced it back to webroot updating itself.

Nothing like having 2 new problems and 4 new variables (I should know better than do mass upgrades, but each individual step seemed ok until boinc issues hit).

I had a lot of computational errors with 295.73 driver but only in one Win7(32) host.
No problems wih 295.73 driver on XP hosts.
I have installed Lunatics v0.39 everywhere and now I don't see errors in the Win7(32) too.
____________