π“Ÿ host1.bitchan.one

𓀐 @4iznkxcM

⏱ 6d seenβ›“ 0πŸ‘ 0β§– 1

filter [-]

languages:onother

I'm not going to be getting my mac studio until next year so I am devising other retard options. Rate the following idea: 5 ddr4 octa channel boards populated with 8x8gb 2666 and then using rdma capable network cards in the x16 slots to cluster them together. I expect that I will have to massively vibeslop the networking implementation but the transfer rates are hypothetically good enough for tensor parallelism.⏱ 5d↩