RTX 3090 Compute Unlock via Firmware

• Someone unlocked an RTX 3090 via a firmware exploit, raising PyTorch performance from 75 TFLOPS to 135 TFLOPS at the same 360W limit, an about 80% gain.
• The unlock targets registers controlling compute speed by exploiting a GPU boot firmware vulnerability to gain write access before the NVIDIA driver loads, then running with the original driver.
• The unlocked limits mainly cover FP16/BF16 input with FP32 accumulation AI operations.
• A separate branch project did the same on the 4090, raising instruction throughput for similar operations from about 165 TFLOPS to 331 TFLOPS.

Key points and the reaction summary are written by AI from the posts on this page. Check the original post. How we use AI