cuda - Concurrently running two for loops with same number of loop cycles involving GPU and CPU tasks on two GPU -

i have 2 for loops in code running same number of loop cycles. these 2 loops independent (each loop works on different input data). within 1 loop, there cpu functions , several kernels not running concurrently.

can run these iterations on separate gpus?

you can run involved kernels separately on 2 different gpus.

you have take care synchronization of cpu processings on partial outcomes of 2 gpus. due presence of sequential part, perhaps not experience maximum possible speedup factor of 2 when working 2 gpus.

starting cuda 4.0, can use cudasetdevice() set current context corresponding given device without need of creating streams enable multi-gpu processing.