I searched this forum and no direct answer was found, so I made this new thread. Just that GTX295 is still "recomended card".
I have this card and I have tried few times to run GPUGRID on it. Didn't run under Win7 64bit pro (newest drivers), nor under Ubuntu 12.04 x64 (drivers from repository 295.40). Let's just forget windows for this moment, I have now Ubuntu cruncher and need to enable it. All tasks failed with "process exited with code 247 (0xf7, -9)", all are CUDA 4.2 tasks.
Thanks for advice

you need to get it using only cuda 3.1. on windows install the 285 drivers, or on Linux install the 296 drivers.

the 296 windows drivers have a terribly cuda bug relating to the monitor sleeping so 285 is recommended

---

XtremeSystems.org - #1 Team in GPUGrid

Thanks, that worked. Linux v296 drivers have Cuda42, after installing driver v285 GPUGRID is finally crunching. I have two more questions:
Does this mean, that after new year, when all Cuda31 apps will be discontinued, also all GTX2xx cards will be unusable? Or is there a possibility, that Cuda42 (or Cuda5) will be able to run on this cards, with newer drivers or whatever?

As far as I understand the problem is related to bugs - so it's not ruled out they'll get fixed by nVidia in future drivers and/or CUDA versions. But I wouldn't count on it.. these cards are getting old, from the viewpoint of nVidia there are no further sales to be gained by better supporting them.

MrS

---

Scanning for our furry friends since Jan 2002

We have suggested to the project admins that they leave the short queue running cuda 31. The beta and long queues are likely to become cuda 42 only.

At this point in time the project don't feel there is anything to be gained by switching to cuda 50 so will be staying with cuda 42.

---

BOINC blog

Leaving short queue as Cuda 31 seems to me as good idea. Older cards can still contribute. May I suggest a new thread containing frequently occured task error numbers. When I had this issue, searching up for error 247 would be good enought for me, I would install drivers v285 right away.

I have a couple of GTX 275 running on x64 Linux with 295.40 drivers.

Long NOELIA units take about 44-47 hours to finish.

Make sure that you have nouveau driver blacklisted in the /etc/modprobe.d/ and add the following two lines to /etc/init.d/boinc-client file into start() function:

sleep 15
nvidia-smi

This will ensure boinc only starts after nvidia drivers had been properly initialized.

Sorry, I spoke too soon. There must be a couple of cuda31 long units from an old queue that I've got after New Year.

After those two units finished on Jan 3 all new cuda42 units were failing even after I upgraded to 304 and then to 310 drivers. Short ACEMD2 cuda31 units work fine.

310 driver seem to be buggy - it crashed twice when I was starting YouTube videos - so I switched back to nvidia-current, that is currently 295.40.