"The Unified Acceleration Foundation (UXL Foundation) is intended to be a cross-industry group to delivering an open standard programming model that simplifies the development of performant and cross-platform applications. This open standards for compute accelerators is bolstered by Intel’s oneAPI and intended to garner industry adoption of the oneAPI specification.
Now there are 3 competing standards Edit: 6ish accually
Obligatory xkcd
it is really 5 main ones now, long as i don’t need Cuda down the road i’m fine with oneAPI.
Vulkan, Apple, Nvidia, Intel, AMD.
I may be missing something, but the only machine learning focused api I know of are AMD’s ROCM, Nvidia’s CUDA, and now Intel’s oneAPI. I haven’t looked into Apple’s machine learning frameworks and I consider vulkan more of a general purpose api than a machine learning one.
Don’t forget about OpenCL.
Vulkan and Apple’s Metal both have machine learning functionality.
This is the best summary I could come up with:
Founding members of the UXL Foundation include Arm, Fujitsu, Google Cloud, Imagination Tech, Intel, Qualcomm Technologies, and Samsung.
But via Intel’s Codeplay Software acquisition work they’ve brought SYCL and various oneAPI elements already atop AMD and NVIDIA GPUs/drivers.
It’s also perhaps a bit surprising The Khronos Group wasn’t involved in this effort given their industry standards expertise and ecosystem.
Of the foundation, today’s press release adds, “enabling solutions that are productive, performant, and provide customers with the freedom to choose the hardware that best fits their unique needs.”
There are some good names involved already but without the backing of AMD and NVIDIA, the GPU compute space at least will likely remain a fragmented mess.
In any event I applaud today’s announcement and I will certainly be monitoring the UXL Foundation progress on Phoronix.
The original article contains 484 words, the summary contains 134 words. Saved 72%. I’m a bot and I’m open source!