Loading census data…
| GPU | Stream | tok/s (agg, b32) | stall % | H2D GB/s | $/1M tok | pred. error | status | $/hr |
|---|
MEASURED benchmark artifact on file · MODELED simulator prediction, unverified · SAMPLE illustrative value pending first sweep. Click column headers to sort. tok/s is aggregate decode throughput at batch 32, INT4 stream, 80-layer 70B-class model.