Intel® oneAPI Math Kernel Library
Ask questions and share information with other developers who use Intel® Math Kernel Library.
7016 Discussions

Offloading to non-Intel GPUS with Intel Fortran (ifx).

Van_Veen__Lennaert
288 Views

A year ago, there was a question on an adjacent forum:

https://community.intel.com/t5/Intel-Fortran-Compiler/Fortran-2023-OneApi-OpenMP-offload-to-any-GPU/m-p/1443726/highlight/true#M164288

asking if offloading (in particular of MKL functions) to non-Intel GPUs can be done with ifx.
At that time, the answer was No, the right drivers have not been developed. In the mean time, offloading to non-Intel GPUs has been enabled for C(++) and SYCL
https://www.intel.com/content/www/us/en/developer/tools/oneapi/dpc-compiler.html#gs.9ux6z5

so I am dearly hoping that this feature is coming to ifx, too. Can anybody tell me if and when that will happen?

The background is that my code is highly optimized for ifx and the Intel MKL implementation and I want to offload the most costly computations (mainly BLAS XGEMM and FFTs) to GPU. All shared facilities I use have NVIDIA or AMD graphics cards, and I will need many.

0 Kudos
3 Replies
Gennady_F_Intel
Moderator
131 Views

The status is the same.

IFX doesn't support offloading to non-Intel(R) GPU.

here is the link to the Intel® Fortran Compiler Release Notes: https://www.intel.com/content/www/us/en/developer/articles/release-notes/fortran-compiler-release-notes.html



0 Kudos
Van_Veen__Lennaert
117 Views

Thanks for the confirmation. Do you happen to know if this is planned for a future release?

0 Kudos
Gennady_F_Intel
Moderator
111 Views

no comments here - we are not discussing our plans into open forums, sorry


0 Kudos
Reply