Breaking

nv-uefi-unlocker Brings GeForce Tensor Unlock to Windows

October 29, 2026 at 06:16 EDT

Infra & Chips
Open Source

A community-developed Windows utility called nv-uefi-unlocker documents a UEFI-based way to remove a GeForce tensor-throughput cap before NVIDIA's Windows driver starts. In its October 27, 2026, v0.2.0 RTX 2070 test, FP32-accumulation GEMM throughput rose from 36 to 60–63 TFLOPS, a reported gain of 67–75%.

OCT. 29, 2026 · NVIDIA GEFORCE / WINDOWS

A Windows path to fuller tensor throughput

RTX 2070 tensor GEMM throughput: 36 to 60–63 TFLOPS. This specific benchmark does not promise faster performance across every AI workload.

RTX 2070 GEMM throughput

FP32 accumulation · GEMM benchmark

36 TFLOPS

Before

60–63 TFLOPS

After

Each block represents 3 TFLOPS; the right column shows the 63 TFLOPS upper end.

What improves

The unlock targets tensor matrix math used in compute-heavy AI work.

What remains limited

Real application gains vary; memory-bound token generation may see little change.

A low-level, per-boot change

RTX 20-series users must enable GSP firmware mode and restart. The documented workflow also requires Secure Boot to be disabled, and power loss clears the unlock.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.