Does arrayfire use tensor cores?

28 views
Skip to first unread message

Brosnan Yuen

unread,
Feb 3, 2021, 2:29:06 PMFeb 3
to ArrayFire Users
Hello,

Does arrayfire use tensor cores in the CUDA backend to speed up matrix multiplication and tensor multiplication?

Thanks,
Brosnan Yuen

Pradeep Garigipati

unread,
Feb 8, 2021, 12:27:52 AMFeb 8
to Brosnan Yuen, ArrayFire Users
Hello Brosnan,

Yes, we do use tensor cores whenever we can. For example, BLAS routine GEMM uses tensor cores when ArrayFire is using CUDA toolkit versions greater than or equal to 10.0 (It had to be built with >= CUDA 10.0 in the first place - which our official website installers are).

Regards,
Pradeep.

--
You received this message because you are subscribed to the Google Groups "ArrayFire Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to arrayfire-use...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/arrayfire-users/c92f40f3-7cbd-4eb7-901a-299f37626acen%40googlegroups.com.


--
Reply all
Reply to author
Forward
0 new messages