Furry Technologists

I have 6800xt and it works well for stable diffusion under Archlinux, I can also run local LLMs but last I've worked with LLMs on GPU I had to do a bit of tweaking and installing rocm forks of things like bitsandbytes

As far as Linux goes, I have a RX 6700 XT and am running Arch. Stable diffusion works great (most of the time, see below), especially with the 12 GB or VRAM. Most games work flawlessly, as long as they are run through Proton. PC VR to the Quest 2 works okay, but I believe the issues I have there are mostly due to ALVR. I have had no issues with Blender (after ROCm is set up and HIP enabled in Blender).

I have experienced GPU crashes in certain niche situations (e.g. in the decode step of a too-large image in SD or in some versions of ALVR), but I have not figured out the source of the issue yet, and it is very infrequent and preventable. There is a possibility that the OC edition of the card I have is slightly unstable in those conditions, causing the crashes.

The only thing I would warn you about is that there is a fair bit of set-up necessary to get ROCm working. Just make sure you have the right packages, and that the ROCm version is set correctly in the environment. My card requires me to override the variable to get it working, like so:

export HSA_OVERRIDE_GFX_VERSION=10.3.0

I think it really depends on what you rely on ML and blender for. If you are making money using ML, then I would go for it. If you're just using ML as a hobby then I would try going for a 6800 and saving a little bit of cash. 6700 may also work well for you to, but you may want to look at an online comparison website and see which one you think would be best for you.

https://gpu.userbenchmark.com/Compare/AMD-RX-6950-XT-vs-AMD-RX-6800-XT/m1843533vs4089