multi GPU questions

Accelerated Computing CUDA CUDA Programming and Performance
#1

I have a few questions about multi GPU programming.

  1. What is equivalent to “cudaSetDevice()” in driver mode?

  2. What’s the overhead of calling “cuCtxSetContext”? How expensive is it?

  3. When page-locked mapped memory is created, can all GPU
    devices access it directly or only the device which allocated it?

Thanks.

#2
  1. cuCtxSetContext
  2. basically nothing
  3. all GPUs can access it if it is allocated with the portable flag OR you are running with unified virtual addressing enabled