[rolleiflex]

Yes, they have the Radeon Open Compute project (ROCM) but they seem to be intent on following the tensorflow from a few stable versions behind. Additionally, and likely this is not something they can do anything about, but if you are using a Linux VM instead of running it natively, ROCM does not work. I had attempted to do this to get some use of the AMD GPU that was in my Mac — no dice.

The holy grail would be a direct replacement backend that could be fed into TF, like CUDA.

[sudosysgen]

If you are using a VM, you can never use your GPU unless you have a very expensive one or a spare GPU to passthrough.

ROCm has a direct replacement backend that can even take CUDA code (it's designed to be incredibly similar). It's called HIP. It's just that no one wants to support it. That is actually how TensorFlow on AMD works (mostly), and you can compile the latest stable release that way.

[tim--]

6600XT pass through works fine on QEMU/virtd. There is no need to have multiple GPUs in your machine to have a passable setup.

I can boot into Linux, and swap into Windows in 2 seconds with this setup. I have a dirty 20 line Bash script that deals with detaching the console, and passing the right things to the right place, but it all works.

ROCm on consumer cards does not work well. The tooling sucks. Massively. I don't understand why AMD doesn't have an extra team of 20 devs working just on the tooling.

Using DirectML with Windows Subsystem for Linux gives you better ML GPGPU support then AMDs native tooling.

[sudosysgen]

>I can boot into Linux, and swap into Windows in 2 seconds with this setup. I have a dirty 20 line Bash script that deals with detaching the console, and passing the right things to the right place, but it all works.

That's a matter of opinion I suppose, but I don't personally find that passable.

>Using DirectML with Windows Subsystem for Linux gives you better ML GPGPU support then AMDs native tooling.

DirectML sucks even more than ROCm, IMO. Also, WSL sucks more than a normal VM.

[tim--]

> I don't personally find that passable

What is a setup that would be passable then? I think a setup like the one that I have described [I believe] would be impossible with Hyper-V or ESXi (though, not that I have even attempted it with either).

[sudosysgen]

I agree witht the latter part about Hyper-V or ESXi, for it to work one would need a virtualizable GPU or two GPUs, so that both OSes can have a graphical environment concurrently.