Kernel module unlocks RTX 4090 software cap, doubling FP16 tensor throughput
CeFurkan · reddit · 2026-10-24
A Redditor found the RTX 4090 is also software performance-capped and shared the open-source gpu-unlock-ada project: a Linux kernel module that enables the SM speed-select override before the stock NVIDIA driver loads. On an RTX 4090 (AD102), FP32-accumulate tensor throughput roughly doubles — FP16→FP32 goes from 165.6 to 331.6 TF/s, BF16→FP32 from 165.5 to 330.6 TF/s. Includes a copy-paste Quick Start; RTX 4070 Ti SUPER is also supported.
More from Infra
- capy-flash runs Qwen3.8-27B at 379 tok/s on two RTX 5070 Ti cards — FeydRowan · 2026-10-24
- Why KV sharing works: 60-layer model keeps 21% of KV cache with lower loss — yoavartzi · 2026-10-24
- The 3090 unlock method reportedly works on 4090s too — TheZachMueller · 2026-10-24
- YC Paper Club: inference matters 12x more than training, worth $220B in market cap — Y Combinator · 2026-10-24
- Scoring 20 criteria: AWS edges Azure 84.7 to 82.7 as best public cloud, GCP trails at 78.9 — DavidLinthicum · 2026-10-24
- Customers prepay billions for AI chip capacity: what the terms reveal about who wins — tengyanAI · 2026-10-24