TF32 → FP32 tensor rate
TF32 multiply with an FP32 accumulator, the default for unqualified fp32 matmul.
How this path is measured
Register-resident MMA issue rate at the card's stock power cap, 20,000 iterations and three timed repetitions, with every output cell checked per state.
What the unlock changes on this path
5 rows over 5 cards. 5 improved by more than 5%, 0 did not; the largest gain is GeForce RTX 4050 Laptop GPU at 2.00×.
| Mode | Stock | Unlocked | Gain | Card |
|---|---|---|---|---|
tf32 | 13.2 | 26.4 TFLOP/s | 2.00× | GeForce RTX 4050 Laptop GPU |
tf32 | 23.2 | 46.2 TFLOP/s | 1.99× | GeForce RTX 4060 Ti 16GB |
tf32 | 47.5 | 94.5 TFLOP/s | 1.99× | GeForce RTX 4070 Ti SUPER |
tf32 | 33.1 | 65.8 TFLOP/s | 1.99× | GeForce RTX 4070 (AD103) |
tf32 | 42.4 | 84.0 TFLOP/s | 1.98× | GeForce RTX 4070 Ti |
Reading the numbers
A 1.00× row is not a failed unlock. It says that particular path was never rate-limited, so the override had nothing to restore. If you benchmark your own card and see 1.00× on the mode you care about, the unlock may well have applied correctly and simply does not help that workload.