jaafar has quit [Read error: Connection reset by peer]
<mdiers_>
hi, does anyone have some time for me to help with a gpu application?
<mdiers_>
On a node with a pool of 16 gpu targets tasks (one per gpu) should be processed. for this I need a parallel executor on the gpu-targets. is there an example somewhere in the direction I can orient myself by?
david_pfander has joined #ste||ar
david_pfander has quit [Ping timeout: 250 seconds]
K-ballo has joined #ste||ar
<simbergm>
mdiers_: I think we don't have anything to do that directly, but block_executor is close to what you want, only for host targets
<simbergm>
with multiple gpus you have the additional fun of dealing with allocating memory on different gpus as well
<simbergm>
so we don't have that, but it wouldn't be an insane amount of work to implement it
<simbergm>
heller might know better
<mdiers_>
simbergm: yes, i have looked into the compute_cuda area and found only the parallelisation on a card
<mdiers_>
simbergm: many thanks for the tip with the block_executor. a good keyword with which i can search further