OneAPI can talk to Intel GPUs through Level Zero or OpenCL. I don’t know if AMD hardware is also working with that OpenCL backend or not, and unfortunately I don’t have the hardware to try. There is also a HIP backend for oneAPI DPC++, but I suspect that’s not what you’re asking for.
If someone with development experience and an AMD GPU wants to try, I’ll be happy to assist. I’m curious myself to see if it works.
That’s super interesting. Since the older devices are Opencl compatible I guess someone who has knowledge and experience with the OpenCl code could get the 99% of AMD GPUs that are currently unsupported by blender up and running ( probably slower but better than nothing ).
The HIP backend is kindda useless for older devices because they are unsupported by HIP anyway.
Thank you! ACES may not be perfect (as many are eager to point out) but having options is better than not. I appreciate the openness to implementing tools that simplify my pipeline.
With your very latest changes I can now successfully build under Linux and run Cycles as a Hydra delegate in Houdini:
This is so awesome! First thing I noticed is that the package JSON isn’t working. I have to declare PXR_PLUGINPATH_NAME in the command line while starting Houdini.
P.S.: Is there a separate thread for the Cycles Hydra delegate here in these forums already?
With that I totally agree, some times we may need to work with ACES, not because we want to or not, but because the client uses it, so the ACES usage is out of question in those situations 
Another quick update: With Brecht’s latest changes the package JSON workflow is now working out of the box.
P.S.: Still no dedicated Cycles Hydra thread? It would well deserve it.
I remember vega’s hip resubmitted
I’ve created a topic here now:
When will the eevee test branch be available?
The latest status update was provided here:
There are no concrete dates announced at this point, it’s under active development.