factual
Custom Interconnect Cuts Chip-to-Chip Latency Over Fivefold
By building a fully custom interconnect stack above the second-layer Ethernet, Etched says it can cut chip-to-chip latency by more than a factor of 5, so each chip can use the memory of other chips far more effectively and time per token falls in proportion to world size.
factualpending
Speaker
GavinEvidence Quote
“We can cut this by more than a factor of 5x, and that allows us to use the memory of other chips much more effectively.”
Source
Etched - Building AI Hardware to Make Inference Faster and Cheaper - [Invest Like the Best, EP.480]— Invest Like the BestCreated: 11/17/2026, 7:08:01 AM
Your notes
Loading notes...