Greetings,
I'm asking, that you could last final time recycle datacenter GPU(MI) so you'd dable last level cache, and get memory to 256GB with twice memory bandwidth, if that's possible.
Like final for the Zen4.
Also I recommend you, to release computing toolkit that runs on this cards, with functions like furrier transform, FASTA DNA sequencing. PyTorch, but also GPU support for PyTorch Geometric would be nice.
But it'll be better if you released this PyTorch and PyTorch Geometric, not like python support, but you release quivalent of that code, in C++ that'll be open sourced with examples, so one can control his own streams.
Also I suggest you to pull out ATi's CROSSFIRE back from the grave, and interconnect those GPU's, so one can transfer GPU data straight from one card to another.
I think computation in genetics is what that GPU is really missing and I know it can be done on the GPU very efficiently and fast.
