I like their APUs a lot. Using a 4800U in the cable tray under the desk to drive screens on which I'm writing this. One in a laptop for whenever I'm away from the desk.
If you're sufficiently determined the compute units on these things are totally usable for running arbitrary code. As in a program that spawns a bunch of threads to work stuff out could have some of those "threads" running on the GPU cores.
I wouldn't say the software stack is totally there for out of the box convenience. As in you'll be writing in freestanding ~C and maybe a bit of assembly. I got partway through implementing that and got sidetracked. The GPU libc in LLVM is roughly the production version of some of that hacking. Between these machines coming out and the MI300A landing I really should put something up on github which looks like a pthread_create that executes on the GPU instead.