about cudaErrorMemoryAllocation

Could anybody tell me why
“cudaErrorMemoryAllocation”
happened when I startup a kernel

You’ll need to provide significantly more information on how this problem can be reproduced, along with the OS & graphics driver versions you’re using.

OS: Windows XP

Driver Version: 6.14.10.9773

My soft are composited by several processes, and the communication is done by COM.

One CUDA server is running, the second process also initialize its own cuda environments. I am sure the host memory and the GPU memory are all enough to run the two processes.

However, when I startup a kernel in the second process, it report the above error.

Any idea?

If you bind the device to the first process, the second cannot access it. You need to create a COM device handler routine that does all CUDA stuff and forward all requests from the two processes to it.

Peter

Thank you for your reply.

Did you mean it is impossible to access the device simultaneously by two process?

Even the kernel functions in two process are not simultaneously running?

It looks that is not the case. Or I make some mistake?

Any help, please.

Correct.

Peter