concurrency - Concurrent kernel execution and OpenCL device partition -


recently needed experiments need run multiple different kernel on amd hardware. have several questions before starting coding hence need help.

first, not quite sure whether amd hw can support concurrent kernel execution on 1 device. because when refer opencl specs, said command queue can created in-order , out-of-order. don't "out-of-order" mean "concurrent execution". there know info this? hardware amd apu a8 3870k. if processor not support, other amd products support?

second, know there extension "device fission" can used partition 1 device 2 devices. works on cpu now. in opencl specs, saw something, i.e. "clcreatesubdevice", used partition 1 device two? question is there difference between these 2 techniques? understanding is: device fission can used on cpu, clcreatesubdevice can used on both cpu , gpu. correct?

thanks kind reply!

real concurrent kernels not needed feature , causes troubles driver developers. far know, amd not support feature without subdevice split. mentioned, "out-of-order" not cuncurrent, out of order execution of queue.

but point in running both of them in parallel @ half speed instead of sequentially @ full speed? loose overall performance if in such way.

i recomend use more gpu devices (or gpu + cpu) if run out of resources in 1 of gpus. optimizing option too. splitting never option real scenario, academic purposes or testing.


Comments

Popular posts from this blog

Change php variable from jquery value using ajax (same page) -

Pull out data related to my apps from Android Play Store and iOS App Store -

How can I fetch data from a web server in an android application? -