ebfe
August 12, 2008, 8:39pm
1
Hi there,
the documentation states that
cudaSetDevice() is used to select the device associated to the host thread:
A device must be selected before any global function or any function from
the runtime API is called. If this is not done by an explicit call to
cudaSetDevice(), device 0 is automatically selected and any subsequent explicit
call to cudaSetDevice() will have no effect.
My question is simply what scope this has - per thread, per object, per process? From my understanding I’m required to work with multiple sub-processes when working on multiple cards.
My question is simply what scope this has - per thread, per object, per process? From my understanding I’m required to work with multiple sub-processes when working on multiple cards.
[snapback]424544[/snapback]
The scope of the context is per thread. You don’t need to fork multiple processes.