Last week, modders discovered a way to double tensor throughput on GeForce RTX 20, RTX 30, and RTX 40 cards, though Zagitta’s original tool only ran on Linux. Now Chinese modder physicsdolphin has released nv-uefi-unlocker, a UEFI utility that lifts the same limit on Windows. The tool removes NVIDIA’s half‑rate cap on FP16, BF16, and TF32 math before the driver loads, without any VBIOS flashing. Gains are substantial: one user’s RTX 4070 Ti Super jumped from 57.00 to 150.93 TFLOPS in FP16 (2.65×) while board power rose from 227 W to nearly 292 W. An RTX 2070 reportedly climbed from 36 to about 60‑63 TFLOPS. RTX 20 owners must enable GSP firmware mode first, and RTX 50 isn’t supported. The unlock resets on every power‑off, requiring a USB stick and either disabled Secure Boot or a self‑signed certificate each cold boot—potentially problematic for games with anti‑cheat that enforce Secure Boot. The modification only affects AI workloads such as image generation and fine‑tuning, leaving games and regular FP32 CUDA unchanged. Pairing it with a power‑limit unlock can push temperatures beyond stock limits, so users should monitor thermal conditions.
expert_faq:\
question: How does nv-uefi-unlocker work?\
answer: It disables NVIDIA’s tensor‑math cap before the driver loads, allowing consumer GPUs to run FP16, BF16, and TF32 operations at full silicon speed, matching professional cards, without flashing VBIOS.
internal_link_anchor: Windows GPU Unlock Tool
primary_keyword: NVIDIA tensor cap unlock
secondary_keywords: ["RTX 20-40 performance", "Windows UEFI mod", "AI compute boost", "GPU tensor math", "modder tool"]
content_type: article
"key_entities": {"companies": ["NVIDIA", "TechPowerUp"], "people": ["physicsdolphin", "impacifist", "Zagitta"], "locations": []}
Source: TechPowerUp News · Summarized by HeadlinesBriefing