NVIDIA GeForce RTX 30 GPUs Get Up To 2X Faster AI Compute Performance With New GPU Unlock Mod, Real-World Gains Of Up To 60%

NVIDIA’s aging GeForce RTX 30 Series (codename Ampere) GPUs apparently have a considerable amount of untapped computing potential. A new open-source project called GPU Unlock, developed by GitHub user Zagitta, allows Ampere GPUs to overcome certain artificial performance restrictions, nearly doubling their Tensor Core compute throughput in specific workloads without requiring a modified NVIDIA driver or a permanent VBIOS flash.

The project specifically targets the instruction issue-rate limitation affecting FP16/BF16 Tensor Core operations with FP32 accumulation, which operates at a reduced rate on consumer Ampere graphics cards compared to their professional counterparts. By removing this restriction, the tool effectively enables additional computing performance that the GPUs were already capable of delivering.

Interestingly, GPU Unlock builds upon earlier research involving NVIDIA’s CMP cryptocurrency mining GPUs. It combines the firmware unlocking techniques of the CMP project with an early initialization approach, using a Linux kernel module that executes before NVIDIA’s GPU driver loads. The module modifies the relevant hardware speed selectors, performs a Function Level Reset (FLR), and then allows the unmodified driver to initialize the GPU normally.

This also means that users won’t have to patch their NVIDIA drivers every time a new version is…