FP16 → FP16 tensor rate

FP16 multiply with an FP16 accumulator, the path that already runs at full tensor rate.

数据怎么来的

Register-resident mma.sync issue rate: all-one inputs, eight independent accumulator chains per warp, 20,000 iterations and three timed repetitions, best of three reported.

是否受 SM speed-select 熔丝约束：否

实测行

共 8 行，覆盖 8 张卡。0 行提升超过 5%，8 行没有变化；提升最大的是 GeForce RTX 4090，为 1.02×。

模式	解锁前	解锁后	提升	显卡
`f16/f16` · unlocked values are the 40,000-iteration run; f16/f16 was mostly unaffected, unlike f16/f32	325.7	331.6 TF/s	1.02×	GeForce RTX 4090 · AD102 · 128 SM
`f16/f16` · control mode, already at full rate	169.4	170.0 TF/s	1.00×	GeForce RTX 4070 Ti · AD104 · 60 SM
`f16/f16` · control mode, already at full rate	210.1	210.4 TF/s	1.00×	GeForce RTX 4080 SUPER · AD103 · 80 SM
`f16/f16` · control mode, already at full rate	153.4	153.6 TF/s	1.00×	GeForce RTX 4070 SUPER · AD104 · 56 SM
`f16/f16` · control mode, already at full rate	133.9	133.9 TF/s	1.00×	GeForce RTX 4070 (Gigabyte) · AD104 · 46 SM
`f16/f16` · control mode, already at full rate	36.6	36.6 TF/s	1.00×	GeForce RTX 3050 6GB · GA107 · 18 SM
`f16/f16` · control mode, already at full rate	214.2	213.9 TF/s	1.00×	GeForce RTX 4080 · AD103 · 76 SM
`f16/f16` · control mode, already at full rate	132.6	132.2 TF/s	1.00×	GeForce RTX 4070 (AD103) · AD103 · 46 SM

这组里最小的比值是 GeForce RTX 4070 (AD103)，为 1.00×。

1.00× 该怎么读

1.00× 不代表解锁失败。它说明那条路径本来就没有被限速，覆盖没有可恢复的东西。如果你在自己的卡上测出你关心的模式是 1.00×，解锁很可能确实生效了，只是对这个负载没有帮助。

FP16 → FP16 tensor rate

实测行

1.00× 该怎么读

其他测量路径