CUDA Libraries and MPI+OpenMP+CUDA - Prace Training Portal
CUDA Libraries and MPI+OpenMP+CUDA - Prace Training Portal
CUDA Libraries and MPI+OpenMP+CUDA - Prace Training Portal
- No tags were found...
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
Know what you have…Let’s forget about <strong>CUDA</strong> 3.x <strong>and</strong> “old” GPU devices…Facts:• OpenMP <strong>and</strong> <strong>CUDA</strong> target local computations• MPI moves data (where is the data, HOST or DEVICE, is not conceptually soimportant) … <strong>and</strong> nothing else.• All modern compilers are OpenMP compatible (al least for OpenMP v2.5)• With <strong>CUDA</strong> 4.x it is now easier to manage multiple devices <strong>and</strong> toasynchronously off-load computation• Current hybrid HW “forces” people to think about complex computations likea set of tasks with dependency that can be dispatch to the bestcomputational device/unit available*February 10, 2012 PRACE Winter School 201236
Change language