GPUUnlock

FP16 with an FP16 accumulator

The full-rate tensor-core path. On Ada this is already at the hardware limit before any unlock, so a measured 1.02× is best-of-three noise rather than a gain.

f16/f16 · TF/s · never rate-limited

1rows
1changed
1.02×peak gain

How this mode was measured

Register-resident mma.m16n16k16 issue rate, 8 independent accumulator chains per warp, best of three timed runs

Distance to the specification peak

  • GeForce RTX 4090: 331.6 TFLOP/s measured against a 330.3 TFLOP/s specification peak — 100.4% of it.

330.3 TFLOP/s (Ada whitepaper, RTX 4090, 128 SMs, 2.52 GHz, FP16 dense) = 1024 FLOP/cycle/SM

Every card measured on this mode

ModeStockUnlockedGainCard
f16/f16325.70 TFLOP/s331.60 TFLOP/s1.02×GeForce RTX 4090 · 128 SMs

Same card, same driver, same binary. The unlocked figure is the only variable.

Which cards this covers

Only cards with a same-card measurement on this mode appear in the table. A card missing from it has not been measured on this path, which is not the same as failing.

Other measured modes