Nvidia Details Grace Hopper CPU Superchip Design: 144 Cores on 4N TSMC Process

NVlink is flying saucer grade tech. Im so envious. I wish there was more interconnect competition. CXL is interesting capabilities but over PCIe PHY & that's just so much slower (7x for 5.0) than this NVlink.

Part of me is still sad Hypertransport/HTX didnt stick around. Supposedly AMD will have some Infinity Fabric connected dGPUs at some point, be somewhat competitve here, but it was so cool having there be a semi-open alliance around things, the possibility for others to hack & play. It didnt get heavily used by 3rd parties so I understand AMD going closed, but I also think it was maybe a bit too far ahead of it's time, that chipdesign (& OSes) werent mature enough to make use of the opportunity then.