Nvidia RTX GPUs Unlocked From AI Throttle
A firmware exploit reportedly removes a half-speed AI throttle on Nvidia GeForce RTX 20, 30, and 40-series GPUs, according to Tom's Hardware. The method uses Linux kernel modules to lift the half-rate tensor limit on these cards.
The report says the gains could be significant. An RTX 4090's peak tensor rate is said to double, and an RTX 3090 Ti's matrix math can rise by up to 80 percent. For anyone building local AI workloads on consumer hardware, that is a notable performance question.
The bigger story is about control. When a vendor limits performance in firmware, users and developers start looking for ways around it, and that raises questions about warranties, stability, support and what 'supported' really means for AI tools on consumer GPUs.
For engineers, IT teams and hardware buyers, this is a reminder to watch how GPU capabilities are segmented and documented. Performance you pay for should be predictable, and workarounds carry real risk in production.
Key points
- What happened A firmware exploit reportedly lifts a half-rate AI tensor limit on GeForce RTX 20, 30, and 40-series GPUs.
- The key detail Linux kernel modules are the method described. The report says an RTX 4090's peak tensor rate can double, and an RTX 3090 Ti's matrix math can rise up to 80 percent.
- Why it matters Performance limits set in firmware raise questions about warranties, stability and support for local AI work on consumer GPUs.