Hello All :) New feature of fermi is ability to execute several kernels concurrently. does it mean that on gt200 i could not do that if i have two different applications running concurrently? will kernels from these different applications run sequentially? Thank You.
Yes, they run sequentially. The driver will automatically handle this for your applications.
“A kernel from one CUDA context cannot execute concurrently with a kernel from another CUDA context.” it is said for 2.0 devices. so kernels from different applications also run sequentially?
Yes, it seems the only way to get concurrent kernel execution in compute capability 2.0 devices is to create several CUDA streams within one context, and issue kernels on both of them.