Cuda 3.2 on Ubuntu 10.10 with GeForce GTX 260 deviceQuery works but bandwithTest fails

I recently installed the Cuda Toolkit 3.2.16 for 64bit Ubuntu 10.04 on my machine that is running 10.10 with a GeForce GTX 260 card. I was able to install and compile SDK examples fine. On running deviceQuery it passes:

./deviceQuery Starting…

CUDA Device Query (Runtime API) version (CUDART static linking)

There is 1 device supporting CUDA

Device 0: “GeForce GTX 260”
CUDA Driver Version: 3.20
CUDA Runtime Version: 3.20
CUDA Capability Major/Minor version number: 1.3
Total amount of global memory: 1878327296 bytes
Multiprocessors x Cores/MP = Cores: 24 (MP) x 8 (Cores/MP) = 192 (Cores)
Total amount of constant memory: 65536 bytes
Total amount of shared memory per block: 16384 bytes
Total number of registers available per block: 16384
Warp size: 32
Maximum number of threads per block: 512
Maximum sizes of each dimension of a block: 512 x 512 x 64
Maximum sizes of each dimension of a grid: 65535 x 65535 x 1
Maximum memory pitch: 2147483647 bytes
Texture alignment: 256 bytes
Clock rate: 1.08 GHz
Concurrent copy and execution: Yes
Run time limit on kernels: Yes
Integrated: No
Support host page-locked memory mapping: Yes
Compute mode: Default (multiple host threads can use this device simultaneously)
Concurrent kernel execution: No
Device has ECC support enabled: No
Device is using TCC driver mode: No

deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 3.20, CUDA Runtime Version = 3.20, NumDevs = 1, Device = GeForce GTX 260

PASSED

Press to Quit…

but running bandwithTest fails with:
./bandwidthTest Starting…

Running on…

Device 0: GeForce GTX 260
Quick Mode

bandwidthTest.cu(598) : cudaSafeCall() Runtime API error : all CUDA-capable devices are busy or unavailable.

Running matrixMul gives the same busy/unavailable error. I then ran matrixMulDrv to check the error code and got an unknown:

./matrixMulDrv
[ matrixMulDrv (Driver API) ]

Using CUDA Device [0]: GeForce GTX 260
GPU Device has SM 1.3 compute capability
Total amount of global memory: 1878327296 bytes
64-bit Memory Address: NO
cuSafeCallNoSync() Driver API error = 0999 from file <matrixMulDrv.cpp>, line 89.

I tried drivers 260.19.36 and 260.19.26.

Does anybody know what problem I might be having? Is it just not possible to run the toolkit designed for 10.04 on 10.10?

Thanks for any help!

UPDATE: Installed Ubuntu 10.04 64-bit on separate partition (same system) and everything seemed to work. deviceQuery and bandwithTest both passed. Then after some time (multipe reboots) I started receiving the same error again:
deviceQuery passes, bandwithTest fails, matrixMulDrv fails with cuSafeCallNoSync() Driver API error = 0999 from file <matrixMulDrv.cpp>, line 89.

UPDATE: While running on 10.04 I ended up getting a similar error as before (bandwidthTest failure due to busy or unavailable devices). As it turns out, there seems to be something weird going on. Occasionally, when Google Chrome is open and I try to run an example program, I get the error. Quiting Chrome frees up the device and bandwidthTest and others work. I can then reopen Chrome and the device continues to work. I’m not sure if this happens with other programs too, but it occurs on both 10.10 and 10.04 with toolkit 3.2 and driver 260.19.36. I tried resetting the compute mode using:

# nvidia-smi -g 0 -c 1

# nvidia-smi -g 0 -c 0

but this didn’t seem to do anything.

I have exactly the same problem. Same drivers and Ubuntu version (64 bit 10.04) although I am on a C2070. All was working fine running my own code, I installed Hoomd and then this problem kicked off. Restarts do nothing and I am now unable to run the same code which ran ten minutes before. Changing compute mode doesn’t work either.

If you have a solution to this that would be amazing as my PhD is kind of relying on getting this running.

Thanks

Dean