About this site
RTX AI Unlock is an independent, single-person resource about one narrow thing: removing NVIDIA's artificial Tensor Core throughput cap on consumer GeForce cards, and being honest about what that does and doesn't get you.
It exists because in the week after the story broke, every search result was either a Reddit thread, a GitHub issue, or a news article that said "this exists" without saying "here is exactly what you do." Nobody was writing down the parts that actually trip people up — the exact-hardware requirement on Windows, the cold-boot reset, the fact that half the cards people ask about have no support at all.
Who runs it
One person, an RTX 4090 owner, who spent a weekend testing both the Linux kernel module and the Windows UEFI build on their own hardware. No hardware sponsorships, no vendor relationships, no PR agency feeding talking points.
This is not a news outlet. There is no editorial team, no review pipeline, and no second shift. If something is wrong here, it is wrong because one person got it wrong — which is exactly why the corrections section below exists.
Where the numbers come from
Benchmark claims are the easiest thing to fake on a page like this, so every number on the benchmarks page is tagged with how strong it is:
- Measured, controlled before/after — same card, same driver, same power limit, run twice. Strongest tier. Currently only the RTX 4090 Linux figure (165.6 → 331.6 TF/s) qualifies.
- Single-rig, before/after — a real user reported both numbers from one machine, but without controlled conditions. Includes the 3090 Ti and 3060 12GB Windows builds.
- Extrapolated or unverified — derived from silicon specs or community chatter. We label these as such, and we do not put them in comparison tables as if they were measured.
When a vendor, a fork, or a news outlet makes a claim we can't reproduce, we say so on the page rather than repeating it. That includes claims that would be convenient for this site's traffic.
Corrections
The first version of the Windows guide on this site was wrong. It described a download-and-copy-unlock.efi flow with an "Ampere or Ada" choice, which is the legacy manual implementation kept in the repository for historical reference — not the current supported path. A reader opened an issue on the upstream project pointing this out, and the page was rewritten from the README and the official compatibility matrix instead of from memory.
What changed: the page now leads with "there is no ready-made binary," lists the four verified exact hardware profiles, documents the actual build toolchain (clang/lld-link 20.1.8, firmware assets generated locally), and states plainly that no RTX 40 series profile exists for Windows at all.
The benchmarks page was corrected the same day: invented per-card figures for RTX 40 series were removed, and the table now only contains numbers with a traceable source.
What this site is not
- Not affiliated with NVIDIA, Zagitta, impacifist, or any hardware brand.
- Not a download site. No binaries are hosted here. Every build is compiled from the upstream repositories by the user.
- Not a repair service. If a card is already bricked, this site cannot help. See the recovery notes in the Windows guide and the upstream issue tracker.
- Not a reason to buy a GPU. Cards outside the verified compatibility list are unsupported, and buying hardware for this is a bad trade.
How it's funded
Display advertising (Google AdSense) and Amazon Associates links, both disclosed on every page. Affiliate links are marked rel="sponsored" in the source and never influence which products appear in a guide — the recommendation on the Windows page is a $25 SSD, and the page explicitly tells readers not to buy a GPU for this.
There is no paid placement, no sponsored post, and no affiliate link in any benchmark table.
Questions, corrections, or a test result worth adding? Use the contact form.