As for Ethash and Vega, I don't think we would ever be able to maximize the use of HBM2's bandwidth (4096 bits)
as memory access to anything greater than sizeof(compute_hash_share), which is 512 bits, would be subject
to tCAS and tRAS. Just my two cents.
In theory you are absolutely correct, but there is still some improvements you can do to loosen HBM2 and achieve higher memclk.
But for now focus on ggsharp going on linux, there is no good cryptonote miner for linux...
P.S: I don't know if you haven't noticed yet ->
https://github.com/OhGodACompany/OhGodADecode/pull/6