This is just RV32I-ZMMUL not a custom architecture, bt if you do want to see something from scratch, without any premade ISA or anything, here is an NPU I built: https://www.armaangomes.com/blogs/kernn/

When you say "can upload any model to it" are you talking LLM models? Surely there is a limit eg. parameters

Yeah, anything that fits into memory, which on that FPGA is very limited. It was built for fun rather than production.

We hope to tape it out, along with an NPU and GPU once we raise $10k. Though first we have to find $10k :(

I think you need one more zero.

I'm kinda lazy lol. We are also implementing out-of-order rn, so we want to get a really solid rv32i setup going before adding multiplication, floating point, vector, etc.