c++ - Is it possible to run a cuda kernel on multiple gpus -
this simple question googling doesn't seem have answer, so.
what want know if have 2 gpu cards (identical) capable of running cuda, can kernel span these cards? or bound 1 card or other? i.e. cuda presented entire set of available gpu cores, or ones on card run on.
if so, there special need know in order make happen , there examples on , above cuda sdk worth knowing about?
target language of course c/c++.
thanks in advance.
a single cuda kernel launch bound single gpu. in order use multiple gpus, multiple kernel launches required.
the cuda device runtime api focuses on whichever device selected. given kernel launch launched on whichever device selected using cudasetdevice()
examples of multi-gpu programming given in cuda samples simple multi-gpu p2p , simple multi-gpu
Comments
Post a Comment