[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: The state of Intel compute/AI packages.



What a surprise to me:

https://www.phoronix.com/news/Intel-Full-SYCL-Upstream-LLVM

Intel is finally going to upstream their SYCL implementation to LLVM. It will be easier to try that out in the future.

On 9/23/23 6:03 PM, M. Zhou wrote:
SYCL is not OpenCL. SYCL is an abstraction of OpenCL.
Debian has not yet started any work on SYCL/OneAPI AFAIK.
It requires significant amount of time for somebody to
start it from scratch.

SYCL/OneAPI do not seem mature enough, because SYCL is not
yet upstreamed to LLVM. And PyTorch's SYCL/OneAPI support
is still missing.

I feel it's still not the right time to start.
At least not myself.

On Sat, 2023-09-23 at 14:25 +0200, silex wrote:
Hello everyone,

I recently switched from AMD GPU to Intel ARC for my Blender and
Stable
Diffusion work.

Intel compute related packages in Debian are maintained under
pkg-opencl-devel@lists.alioth.debian.org. Unfortunately that mailing
list is riddled with spam, and from what I can see is not very
active.
Since the 4 of september intel-compute-runtime package has serious
bug
that makes it uninstallable with newest compiler:
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1051189 That
package
also is couple solid months behind upstream.

I'm writing about it here to maybe start a discussion whether it
would
be better to have AI/compute packages from all GPU vendors under the
same roof?
My apologies if this question is inapropriate.

Best regards, and thanks for your work,
Jakub Jaszewski



Reply to: