factual
Data movement dominates naive core area
In a pre-Tensor-Core CUDA/CPU core, the cost of moving data from the register file to the ALU and back vastly exceeds the cost of the logic unit itself: with three muxes (one per multiply-accumulate input) over an 8-entry register file, roughly seven-eighths of the circuit area is spent on reading and writing registers and only a tiny fraction on the actual computation.
factualpending
Speaker
Reiner PopeEvidence Quote
“almost all of the cost, seven-eighths of the cost, is in reading and writing the register file, and only a tiny fraction of the cost is in the logic unit itself.”
Created: 6/13/2026, 3:33:43 AM
Your notes
Loading notes...