CUDA Toolkit version for 285GTX?


I’m trying to execute CUDA software with Cuda Toolkit 4.1 installed:

SimplePrintf sample says:
cudaGetDeviceCount returned 35
-> CUDA driver version is insufficient for CUDA runtime version
value = 35 : Memory Allocation on GPU Device failed
Memory Copy from Host to Device failed.
Memory Copy from Device to Host failed.

Other more-complex program says:
WARNING: [?]Couldn’t initialize the GPU support component at runtime startup: CUDA driver version is insufficient for CUDA runtime version

Driver 285.05.33 installed. Aka lastest (the one provided with CUDA toolkit 4.1).

I’m running it on a shared machine where i’m not root, CUDA Toolkit is installed on my home directory, driver was installed and machine restarted. Machine has 285GTX SLI.

I need some permission or something else? Install older Cuda Toolkit?


PS: If i remove cudaStat = cudaMalloc ((void **)&cosArg, N * sizeof(cosArg[0])); all cudaMalloc calls, program will execute correctly and this cos_main<<<1,COS_THREAD_CNT>>>(funcParams) will work “fine” (altho printf wont show any change)
PS2: After installing toolkit 2.3 (the one for GTX 285 i think), i get -> no CUDA-capable device is available, i guess permission problem.

Looks like no luck :(

Just for being sure i’m not wasting my time (and machine’s administrator one), someone is running nvidia 200 series with CUDA Toolkit 3.1-4.0-4.1 compiled sources? (in any OS, win, linux, mac…).

Yes, it should work.

Double check that you are running the right driver (“cat /proc/driver/nvidia/version”) and that you have permission to read the /dev/nvidia* files.

Okay thanks,

After all… i was in the frontend :P, needed to switch into computing nodes @ cluster and fixed. It looks like the error was the “same”, but in CUDA 3.1 and 4.1 its just different message/error than 2.1.