Why would you want that exactly rather than CUDA or OpenCL?

Although I haven't written a compute shader yet, compute in Vulkan looks trivial to use once you have waged your initial battles against the rendering pipeline setup. Most of the complexity with Vulkan lies in understanding descriptor sets, pipeline layouts and pipeline objects. I find that the Vulkan interface is remarkably consistent once you get over the initial - and substantial - hurdles.

Because Vulkan is cross-device, cross-platform. I dont want to tied to the Desktop or to Nvidia.

Vulkan Compute shaders work in recent Android devices and on every Windows/Linux PC that has a <5 year old dedicated GPU. And for iOS/OSX there is an interop layer in the works that will enable Vulkan over Metal.

> Vulkan Compute shaders work in recent Android devices

Actually no.

Vulkan is an optional API, so unless we are talking about Samsung and Google flagships phones, there is no guarantee of it being available.

So far Google hasn't even bothered to add it to the dashboards.

For compute, only Renderscript is required.

Apple has direct support for OpenCL. Why would you want to use a Metal wrapper instead?

iOS doesn't support OpenCL

Oh, I did not know that. Still, using Vulkan strictly for its compute shaders sounds weird to me.

OpenCL is merging into Vulkan: https://www.pcper.com/reviews/General-Tech/Breaking-OpenCL-M...

It is unclear to me whether the target of the statement is runtime interop between OpenCL and Vulkan or something else, like implementing (a subset of) OpenCL on top of Vulkan. I have not seen statements recently from Khronos that would hint at the later.

>The OpenCL working group has taken the decision to converge its roadmap with Vulkan, and use Vulkan as the basis for the next generation of explicit compute APIs – this also provides the opportunity for the OpenCL roadmap to merge graphics and compute.

From the article suggests at some point OpenCL is going to run on top of Vulkan.

If you ask me, the article is trying to interpret intentionally vague marketing speak in a very specific way. But that is only my opinion.

You're looking in the wrong spot, Vulkan compute is there to support the graphics. If you want pure compute, opencl/cuda are your best options.

Vulkan compute is super simple compared to vulkan graphics. In the graphics you have to consider renderpasses, tons of GPU state, and whatnot.

In compute you just dispatch kernels and put barriers to prevent race conditions. Pretty much like OpenCL Out Of Order queues except the arguments are passed via descriptor sets rather than given individually.

Well the author did say “if you have a different vision, or opinion, about what Learn Vulkan should be, don't hesitate to contact me as nothing is set in stone yet”, and while I think that may have been primarily about the book/website combo I think now is a perfect time for you to contact the author (contact details on the page) and ask them about including Vulcan compute on the site.