OpenCL 2.2-10 Released With Fixes

[phoronix]

Phoronix: OpenCL 2.2-10 Released With Fixes

While "OpenCL-Next" will hopefully be on track for releasing later this year as the next big update to OpenCL, OpenCL 2.2-10 was released today by The Khronos Group as the latest maintenance update to the nearly two year old OpenCL 2.2 specification...

http://www.phoronix.com/scan.php?page=news_item&px=OpenCL-2.2-10

[coder]

OpenCL-Next can't come soon enough to hopefully bolster OpenCL GPU programming adoption

Raspberry Pi v4 can't come soon enough to hopefully bolster OpenCL GPU programming adoption.

[boboviz]

OpenCL-Next can't come soon enough to hopefully bolster OpenCL GPU programming adoption and OpenCL 2.2 showing its age with the provisional specification for it approaching three years old

The 90% of OpenCl projects use OpenCl 1.2 (source: Neil Trevett), so the slow update of OpenCl is not a problem.

[coder]

The 90% of OpenCl projects use OpenCl 1.2 (source: Neil Trevett), so the slow update of OpenCl is not a problem.

That might be rather self-serving of him. If Nvidia (of which he's a VP) would support anything higher, that number would certainly drop.

[xiando]

That might be rather self-serving of him. If Nvidia (of which he's a VP) would support anything higher, that number would certainly drop.

Nvidia actively supporting OpenCL or promoting it at all would be foolish so they are not going to do that. There's this proprietary alternative called CUDA and it's exclusive to NVidia products. Make your code in OpenCL and you'll be able to use whatever GPU happens to provide the best price/performance when you upgrade your hardware down the line, choose CUDA and you're forever locked into using NVidia products. That's obviously not something NVidia wants, why would they? There's a reason AMD is hell-bent on promoting their open source ROCm alternative.

As for OpenCL 2.0: It's interesting to note that MESA's clover OpenCL for AMD GPUs is stuck at OpenCL 1.2.

[boboviz]

That might be rather self-serving of him. If Nvidia (of which he's a VP) would support anything higher, that number would certainly drop.

That's a great point. I cannot understand why the president of Kronos (leader in open source code) is a VP of a company devoted to closed source.

Returning to argument: i think that AMD is the disappointment of OpenCl ecosystem. They are working slowly and, seems to me, without conviction.

[Michael]

That might be rather self-serving of him. If Nvidia (of which he's a VP) would support anything higher, that number would certainly drop.

NVIDIA will support OpenCL Next.

[boboviz]

P.S. It's not only Trevett that says that most used version is 1.2.

[boboviz]

NVIDIA will support OpenCL Next.

5 years after the release :-P

[coder]

P.S. It's not only Trevett that says that most used version is 1.2.

Oh, I didn't say he's lying (though it would be interesting to know his sources).

My point is that highlighting it leads to somewhat circular reasoning that could justify why Nvidia hasn't moved beyond 1.2.