NVIDIA RTX GPUs Unlocked for Higher Performance
Modders bypass FP16 and BF16 compute limits on GeForce RTX 20, 30 and 40 series GPUs.
According to TweakTown, modders have bypassed FP16 and BF16 compute limits on NVIDIA's GeForce RTX 20, 30, and 40 series GPUs. The unlock doubles the issue rate of FP16 and BF16 operations with FP32 accumulation on GA102, GA103, GA104, and GA106 cards.
TweakTown reports that an RTX 3090 Ti saw a performance boost from 87.46 to 118.39 TFLOPS in FP16 after the unlock.
Key facts
- RTX 20, 30 and 40 series GPUs affected
- FP16 and BF16 compute limits bypassed
- Issue rate of operations doubled
- RTX 3090 Ti performance boost from 87.46 to 118.39 TFLOPS
Sources
- TweakTownModders unlock NVIDIA GeForce RTX 20, 30 and 40 series compute limits, nearly doubling some Tensor Core throughput
- Tom's HardwareNvidia RTX 20, 30, and 40-series GPUs hacked to remove a half-speed AI throttle
- The FPS ReviewCompute Performance for GeForce RTX 20, RTX 30, and RTX 40 GPUs Nearly Doubled in Linux Thanks to New Mods
- Club386Nvidia GeForce RTX 40, 30, and 20 GPUs get 2x compute performance boost, thanks to this mod
