GPU sharing between programs that support OpenCL

Question

Is there a way to share the GPU between two separate programs that support OpenCL, or, more specifically, between two separate processes that simultaneously require the GPU to run OpenCL kernels? If so, how is this done?

Answer 1

It depends on what you call sharing.

In general, you can create 2 processes that simultaneously create an OpenCL device on the same GPU. Then the responsibility for the driver / OS / GPU should be a guarantee that everything will work.

, , ( ).

, , . ?

Answer 2

( NVidia Fermi) . , GPU ; ! . , GPU .

, GPU, , (< 100ms - ), GPU , GPU, , , GPU. , ​​ , -, .