Context Creation Still Taking Time Context Creation Still Taking Time on CUDA 2.2 and Tesla

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello Everyone

We have a Tesla 1U server with Tesla C870 GPUs. Presently we are using CUDA 2.2.

When I use the cudaSetDevice command as shown, it is still slow ~ 2.1sec. Another older forum post also mentions it. Is this still an issue.

I know that CUDA 2.2 has a newer method to manage contexts. Would it make a difference?

Older Post

http://forums.nvidia.com/index.php?showtopic=93494

By adding the below snippet in the start of the deviceQuery example.

[codebox]

int device_no;

CUT_SAFE_CALL(cutCreateTimer(&timer));

CUT_SAFE_CALL(cutStartTimer(timer));

device_no=0;

CUDA_SAFE_CALL(cudaSetDevice(device_no));

CUDA_SAFE_CALL(cudaGetDeviceCount(&deviceCount));

CUT_SAFE_CALL(cutStopTimer(timer));

[/codebox]

I get two devices as expected for my system

Device 0: “Tesla C870”

&so on…

Device 1: “Tesla C870”

–

Thank You.

Perhaad