Yes AMD’s support for older hardware does not go back sufficiently far and AMD takes the blame there but Blender 3D(The Blender Foundation) are the ones that stopped using OpenCL(Originally created by Apple but since given over to the Khronos industry standards body to manage) the older industry standard iGPU/dGPU compute API and the entire industry is fragmented now. And so AMD has ROCm/HIP, Apple Has Metal, Intel has OneAPI/level-0, and Nvidia has had CUDA(The One that was always supported by Nvidia for Blender 3D and with the best support for that Application).

And so ZLUDA was poised to take that support out of AMD’s hands and possibly move that over to be included the MESA GPU Graphics/Compute driver stack that ships with most Linux Distros but Nvidia’s licensing and legal arm put the kibosh on that and so here we are with AMD intentionally not supporting ROCm/HIP on its consumer iGPUs/dGPUs to any great degree.

It really comes down to the Linux Community not supporting, over the years, OpenCL to any great degree and making sure that the MESA drivers had a reasonably modern OpenCL implementation and some support for smaller GPU Kernels as well! So the Blender Foundation dropped support for OpenCL and moved to CUDA and later also Apple’s Metal where Apple provided the technical support to make sure that Apple’s iGPUs support Blender 3D’s iGPU accelerated Cycles rendering! And CUDA is just a C like programming language much like OpenCL but the CUDA Libraries are the most supported and advanced and ZLUDA was going to make that easy to access as all the software is written in high level languages like CUDA, and C/other and compiled down into portable Intermediate Language Representation cross platform “Binaries” that are actually re-compiled by the GPU drivers into that native machine language of the GPU hardware. And so ZLUDA was ROCm/HIP based but could use the CUDA libraries directly and thus that CUDA/PTX from Blender 3D’s back end could get cross-compiled to run on AMD or Intel/Other GPU hardware. The Major Attraction with Blender is that it’s free and open source software and thus does not cost thousands of dollars yearly to license like Maya/Solidworks/other software.

The big question is why did the entire Industry Abandon OpenCL for iGPU/dGPU compute for some fragmented non standard solution and that’s rather obviously for vendor Lock-in! And now maybe the competition authorities will have to come in via the Courts and Government Policy and force that Industry to support a standard iGPU/dGPU compute APU to put an end to the hardware fiefdoms attempts at cornering that iGPU/dGPU compute acceleration market!

It’s rather Ironic that the MESA drivers now have a modern OpenCL implementation in the Form of Rusticl(OpenCL Implementation written in the Rust Programming language) but that’s a little too late for the Blender 3D support as that application has dropped supporting OpenCL!

“No way to access that IGPU compute”? Sounds like you should be mad at AMD who should have supported their product properly in the first place.

That’s not for you to decide and Nvidia’s market actions fall firmly under Sherman Antitrust Act(As Amended) least there would be no pending actions concerning that! And the US regulators are not the only ones look at Nvidia’s actions in the market place!

Now AMD’s APUs have those iGPUs but no way to access that iGPU compute via the ROCm/HIP stack for generations of those devices based on Vega Integrated Graphics! And I know because I own 2 of those Ryzen APU based systems with Vega Integrated Graphics and I can never utilize that for Blender 3D’s iGPU accelerated Cycles rendering because Vega graphics, like Polaris before that, was dropped from AMD’s ROCm/HIP support matrix! But ZLUDA promised to change that but Nvidia’s Legal Eagles put a stop to that project with their licensing restrictions related to CUDA!

Now Blender 3D’s iGPU/dGPU Accelerated Cycles rendering is a Compute Workload(The Part of that’s related to Ray Tracing and Bounding Volume Hierarchy calculations and triangle intersection calculations) and all that math accelerated on old fashion GPU Shader cores that lack any Hardware Accelerated Ray Tracing on the GPUs, like Vega and Polaris graphics from AMD and the GTX series of GPUs from Nvidia!

And so if the regulators allow that ZLUDA code to be utilized I may have day where My Vega and Polaris Graphics based Laptop and Mini Desktop PC could actually make use if the Vega iGPUs and the Polaris dGPUs(my Laptop has a RX 560X dGPU) for Blender 3D iGPU/dGPU accelerated Cycles rendering that has the full Ray Tracing available for all render passes on Blender 3D. And the Modern Versions of Blender 3D(3.0/Later editions) dropped supporting OpenCL for iGPU and dGPU compute acceleration and modern Blender 3D only supports Nvidia CUDA(PTX Intermediate Language) and Apple Metal(Whatever Intermediate Language is used there) back ends!

I don’t believe so, either. But there is a non-zero chance something will. There are senators in Oregon and Texas….

It’s called supply and demand. Demand for those enterprise GPU’s is massive and Nvidia is not a charity.

More here in comment string on Nvidia q3 results, cost : price / margin, product category volumes, touches on prior horizontal and vertical tying examples that were cleaned up, and my natural monopoly take there are two major sections; near the top and in the middle, here in comment string;

https://seekingalpha.com/article/4741503-nvidia-stock-has-huge-potential-upside-for-2025

Mike Bruzzone, Camp Marketing