Windows GPU unlock arrives: RTX 3090 Ti matrix performance jumps 86%
lxfater · x · 2026-10-26
Following the earlier Ubuntu unlock, the community has ported the RTX 30-series hidden-performance unlock to Windows: unlock at boot via UEFI, then hand off directly into the OS with stock NVIDIA drivers.
- A contributor's 3090 Ti went from 80.4 to 149.3 TFLOPS in FP16-input/FP32-accumulate matrix tests (86% gain)
- Still experimental: self-compile required, and you must verify GPU, motherboard and BIOS compatibility
- Verified configs so far: 3090 FE, 3070 Ti and 3090 Ti
- The 86% figure is matrix-only; gaming frames and LLM inference speedups are untested
The author notes his own 4070 Ti SUPER was only verified on Ubuntu and isn't yet on the Windows project's list.
Related event: Community Unlocks Hidden Tensor Performance on RTX 20/30/40 GPUs(3 posts)→
More from Infra
- You Can Train a 1B Model in 3 Weeks: Dev Shares 15.9-Day Training Run Log — arthurcolle · 2026-10-26
- TWIG: weights as shared discrete decision paths unify pruning, quantization, and adder-based circuits — ITMO · 2026-10-26
- BofA sees 21M incremental CPUs by 2030 from agentic AI — analyst calls the math nonsense — AccBalanced · 2026-10-26
- Resin author: KV slabs inspired by TokenSpeed, self-contained forward pass per model+quant — ekzhang1 · 2026-10-26
- Why NVL72 scale-up bandwidth sits mostly idle — cutting it won't hurt inference — bookwormengr · 2026-10-26
- Custom OpenAI-branded gigabit network switch PCB shown off in photos — mckbrando · 2026-10-26