Thoughts on using the K3
So the varying VLEN across the x100 and a100 core cluster is going to make the a100 cluster harder to use -- we can't just present it as a 16c system where processes run wherever they want and migrate freely.
I'm not likely ever going to bother trying to explicitly launch things on the a100 cluster or setting up distcc to use that cluster (distcc would only help marginally for cases I care about). *But* I could easily see standing up a docker container on the a100 cluster and opening ports to make it appear as-if the a100 cluster is a distinct system. I can also limit the docker containe for the a100 to 16G of memory, which I already know is sufficient for my needs.
The net is I think I can utilize both cpu clusters meaningfully. A single unified 16c system would be best, but this is a viable second-best option AFAICT.