3 ms·
They’re already optimizing GPU die area for LLM inference over other pursuits: the FP64 units in the latest Blackwell GPUs were greatly reduced and FP4 was adde
by conradev 1y ago
They’re already optimizing GPU die area for LLM inference over other pursuits: the FP64 units in the latest Blackwell GPUs were greatly reduced and FP4 was added