Benchmarks

Every figure below is labelled by where it came from. This site has not re-run these benchmarks — they are the project maintainer's published measurements and the community reports the project links to, gathered so you don't have to dig through issue threads.

Sources checked 2026-10-25 · Upstream README pinned at commit af28cbb (2026-10-24) · Percentages computed from the published before/after figures

The headline result — RTX 3090 Ti, dense GEMM

Project-reported Dense 8192³ GEMM on an RTX 3090 Ti at a 450 W power limit, five-sample medians, FP32 accumulation. Published in the gpu-unlock README and its linked benchmark report.

Input typeBefore (locked)After (unlocked)Change
FP1687.46 TFLOPS118.39 TFLOPS+35%
BF1687.50 TFLOPS157.41 TFLOPS+80%
TF3242.58 TFLOPS82.61 TFLOPS+94%

Why BF16 gains the most: the lock halves the FP/BF16 FMA issue rate into FP32 operations, so removing it can at most double that rate — BF16's +80% captures most of the theoretical headroom in this benchmark. FP16 improved less in the same run (+35%); the README doesn't break down the difference, and real workloads mix many operation types anyway, so treat the trio as "measured on this benchmark" rather than a formula. And these are one card, one benchmark, one power limit: "what the unlock can do on a 450 W flagship", not a promise for every card.

Inside issue #1 — Pento95's 3090 Ti report, read closely

Community-reported Issue #1 is useful because its reporter, GitHub user Pento95, recorded the machine around the number, not just the number. The card was an MSI RTX 3090 Ti (10de:2203, subsystem 1462:5090) that also drove the console, running Fedora 44 with dracut, Secure Boot off, the NVIDIA 615.78.08 closed module with GSP enabled, and one exact target address (gpu_unlock.targets=0000:2d:00.0). Its cold-boot log records the selector writes (SS0=0x88888888, SS1=0x8), the post-FLR state and pci 0000:2d:00.0: gpu_unlock: unlocked before the stock driver bound normally.

RunPower limitFP16BF16TFLOPS per MHz
Before300 W59.6 TFLOPS63.0 TFLOPS0.043
After, cold boot380 W100.3 TFLOPS105.9 TFLOPS0.083
After, warm reboot350 W89.5 TFLOPS95.1 TFLOPS0.083

All three runs use the project's 8192³ torch.matmul with FP32 accumulation. Read that table the way its author did: the raw TFLOPS rows are not a like-for-like pair, because the power limit — and therefore the SM clock — moved between runs. The comparable figure is TFLOPS per MHz, which rose from 0.043 to 0.083, roughly a doubling. The warm reboot also succeeded on this one MSI AM4 board; the reporter suggested its reboot path resets the slot far enough, which is a board-specific observation, not a general promise. For an application-shaped counterweight, the same issue reports an INT8-heavy vLLM fork whose end-to-end gain was about 5% — within run-to-run noise. Source: gpu-unlock issue #1.

Per-card results — community reports

Community-reported BF16 dense-GEMM gains from the user reports in the project's tested-cards list. Each report ran its own benchmark, so rows aren't directly comparable — and power limits explain much of the spread.

CardChipReported BF16 GEMM gainNote
RTX 3090 TiGA102+80% (450 W)The 450 W headline run above. A separate issue #1 report showed +68% raw BF16 at unequal power limits (300 W vs 380 W) — not a controlled unlock-only gain; see "Inside issue #1" below.
RTX 3090GA102+25–67%Lowest figure from a card capped at 225 W.
RTX 3080 10 GBGA102+70–73%—
RTX 3080 20 GB (mod)GA102+56–59%Memory-upgraded card.
RTX 3070GA104+73–81%—
RTX 3060 12 GBGA106~+80%—
RTX 3080 Ti 20 GB (OEM)GA102-250no gainUnlock completes; BF16 stays at half rate. The known exception.

Real workloads

Community-reported Collected in the project's README from its issue reports:

How to read these numbers honestly

"Up to ~2×" applies to one microbenchmark class (dense GEMM issue rate), on tested cards, at generous power limits. End-to-end application gains are smaller — LLM prefill's +14–36% is the realistic shape of it. If a workload of yours is memory-bound, expect nothing. Anyone quoting you a single "speedup percentage" for this unlock without naming the workload is selling something.

Gaming performance

There is no verified gaming benchmark data for gpu-unlock as of this writing — no FPS tables in the README or its linked reports. The lock it removes targets FP/BF16 compute issue rates used by AI and HPC workloads, and every published measurement is a compute benchmark. If gaming numbers matter to you, the honest answer today is "unmeasured", and you should treat any FPS claims circulating without a methodology as unverified.

Primary source: github.com/Zagitta/gpu-unlock (README benchmark table and tested-cards list, with per-report links), pinned at commit af28cbb. ComfyUI figures: Reddit user under-_-null’s RTX 3090 Ti test, as reported by Notebookcheck and WccfTech. RTX 40 series figures live on the RTX 40 page — different project, different evidence base.