Mongo
@minusYCore · 18 janv. 2026
Requires 4x the operations though (if doing full precision multiplication)
So effective clock speed is only 1/4 of whatever HW3 baseline is (for those operations)
Elon Musk
@elonmusk
This has already been implemented and is used only for longer bit length operations, which are relatively rare.
Overall compute is extremely efficient.
16:55 · 18 janvier 2026 · 148,3 k vues
105
111
2,4 k