A software tool has been released that unlocks higher FP/BF16 instruction issue rates on Nvidia RTX 3000 series GPUs, effectively doubling their compute performance for AI workloads.

What it means

A new open-source Linux kernel module named 'gpu-unlock' by developer Zagitta lifts the firmware-level cap on FP16 and BF16 instruction issue rates for Nvidia RTX 3000 series GPUs. This effectively doubles their compute throughput for AI workloads, matching the rates of professional-grade Nvidia cards such as the A-series. The tool has been released on GitHub and confirmed to work on RTX 3080 20GB and similar cards, with reports extending the same principle to RTX 20 and 40 series cards.

Accuracy & sourcesGrounded

The claim is solid and directly backed by the primary source: the Zagitta/gpu-unlock GitHub repository, which provides the actual kernel module and documentation. Multiple secondary sources (Hacker News, AIWeekly, TechPowerUp) independently describe the same performance doubling and firmware cap removal, lending strong corroboration. The provenance tier is correctly 'primary source' because the link points to the actual released tool.

How new

This is genuinely new: the GitHub release and initial coverage all appeared on October 23–24, 2026.

Go deeper