Nvidia RTX 20, 30, and 40-series GPUs hacked to remove a half-speed AI throttle
Linux kernel modules lift Nvidia’s half-rate AI tensor limit on GeForce RTX 20, 30, and 40-series GPUs.
Modders have discovered a way to bypass the half-speed AI performance limits imposed on Nvidia's RTX 20, 30, and 40-series GPUs using a firmware exploit, as reported by Tom's Hardware. This unlocking process can significantly enhance the performance of these graphics cards, with the RTX 4090 achieving double its peak tensor rate and the RTX 3090 Ti experiencing up to 80% improvement in matrix math.
While the exact gains vary across different cards, real-world workloads benefit less than the peak rates would suggest, with chatbot token speeds remaining relatively unchanged due to their dependency on memory bandwidth. Nvidia had intentionally imposed a limit on FP32-accumulate math, running at half the FP16-accumulate rate for all impacted GeForce generations, except for Turing and Ampere pro cards.
Brief written by urgent.news from Tom's Hardware's own syndicated text. Machine-written — may contain errors; check the original before relying on it.
