Free firmware unlock nearly doubles AI speed on your RTX 30 and 40 GPU - MakeUseOf
Independent modders have released firmware-level utilities that nearly double the AI compute performance of NVIDIA's RTX 20, 30 and 40 series graphics cards on Linux systems.
The engineering feat allows consumer cards like the GeForce RTX 4090 to achieve performance jumps from 165.6 to 331.6 trillion operations per second for certain AI math calculations. This level of performance is typically only found in expensive professional cards like the NVIDIA RTX 6000 'Ada Lovelace'.
The utilities, including open-pu-unlock and related modules, intercept the boot process before official drivers bind to the kernel, overriding PCIe Function Level Reset restrictions. Rather than modifying the permanent video BIOS, the utility removes software bottlenecks on instruction-issuing rates for FP16 and BF16 matrix calculations.
Performance Gains
For the RTX 30-series 'Ampere,' the tool enables cards like the GeForce RTX 3090 Ti to perform matrix calculations up to 80% faster. The solution works on GeForce RTX cards with GA102, GA103, GA104, and GA106 chips. For newer RTX 40-series 'Ada Lovelace' cards, the unlock doubles the speed for certain types of math used in AI applications.
Real-World Impact
The main beneficiaries of this unlock are hobbyists, developers, and researchers running their own local LLMs or using AI upscalers for video. The faster math throughput allows processing more tokens in these applications. However, PC gamers should note these changes provide no benefit to traditional games, real-time rendering, or frame rates.
if you're going to unlock enterprise level AI performance on an older consumer architecture, you're going to create a lot of secondary market demand
Market Concerns
The community has expressed concerns about the secondary market impact. If hobbyists and small developers purchase older generation consumer cards to build affordable ML rigs, demand and prices could rise, potentially pricing out average gamers seeking entry-level cards.
Currently, the firmware unlocks are Linux-only, require manual configuration at boot, and void the hardware warranty. This limits their adoption but doesn't diminish their significance for those willing to take the risk.
Related
