# AI use cases for GPU Unlock Ada (JustAResearcher/gpu-unlock-ada) Repo: [JustAResearcher/gpu-unlock-ada](https://github.com/JustAResearcher/gpu-unlock-ada) | Use cases: 3 | Page: https://dedupedai.com/use-cases/justaresearcher-gpu-unlock-ada/ GPU Unlock for Ada is a Linux kernel module that runs at boot, before the stock NVIDIA driver loads, and switches supported RTX 40 cards to full-rate FP16 and BF16 tensor math with FP32 accumulation. The authors measured about 2x in instruction benchmarks on an RTX 4090 (165 to 331 TFLOPS) and smaller gains in real GEMM workloads. This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL. - Unlock full-rate FP16 and BF16 tensor math on an RTX 4090: https://dedupedai.com/use-cases/justaresearcher-gpu-unlock-ada/unlock-full-rate-fp16-bf16-on-rtx-4090/ - Benchmark stock vs unlocked tensor throughput: https://dedupedai.com/use-cases/justaresearcher-gpu-unlock-ada/benchmark-stock-vs-unlocked-tensor-throughput/ - Test the experimental RTX 4070 Ti SUPER or 4060 Ti path: https://dedupedai.com/use-cases/justaresearcher-gpu-unlock-ada/test-experimental-rtx-4070-ti-super-4060-ti-unlock/ ## How to unlock full-rate FP16 and BF16 tensor math on an RTX 4090 with GPU Unlock Ada Repo: [JustAResearcher/gpu-unlock-ada](https://github.com/JustAResearcher/gpu-unlock-ada) | Difficulty: Advanced | Time: 1 hour | Basis: From the README Build the kernel module, add one boot option, cold boot, and check the kernel log for the unlock. **Who it is for:** An ML engineer with an RTX 4090 Linux box who wants more FP16 or BF16 throughput for training or inference. **The problem:** GeForce cards run FP16 and BF16 tensor math with FP32 accumulation at half rate. **The outcome:** A card that reports full-rate tensor math after boot, with the stock NVIDIA driver loaded on top. ### Steps 1. Run `git clone https://github.com/JustAResearcher/gpu-unlock-ada.git` and `cd gpu-unlock-ada` 2. Run `make install` to build the package locally, which downloads and verifies the pinned NVIDIA inputs and generates the firmware on your machine 3. Find the card's full PCI address with `lspci -D` and add `gpu_unlock.targets=` with that exact address to the kernel command line 4. Power the machine off and start it again for the first attempt, since a warm reboot may not give the card a fresh state 5. Run `journalctl -k -b | grep gpu_unlock` and look for that card's `gpu_unlock: unlocked` line; a loaded module alone does not prove it worked ### Prompt to paste into your agent ```text On my Linux machine, build and install JustAResearcher/gpu-unlock-ada with `make install`, add gpu_unlock.targets= to the kernel command line, keep a stock boot entry, and tell me when to cold boot. After boot, report the gpu_unlock lines from the kernel log and whether `gpu_unlock: unlocked` appears for that card. ``` ### Caveat The 2x figure is a register-level instruction rate; one matched run found real cuBLAS GEMM gains of 12 to 34 percent for FP16, so measure your own workload before trusting any headline number. Risk, reverse engineering: The unlock runs a payload derived from published analysis of NVIDIA's debug booter and changes GPU register state from a kernel module before the driver loads. It is unofficial, unsupported by NVIDIA, may void your warranty, and has been validated on very few cards. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/justaresearcher-gpu-unlock-ada/unlock-full-rate-fp16-bf16-on-rtx-4090/ ## How to benchmark stock vs unlocked tensor throughput with GPU Unlock Ada Repo: [JustAResearcher/gpu-unlock-ada](https://github.com/JustAResearcher/gpu-unlock-ada) | Difficulty: Advanced | Time: 1 hour 30 minutes | Basis: From the README Run matched stock and unlocked measurements so you know what the unlock does for your card and workload. **Who it is for:** A developer who saw the viral 2x claim and wants numbers from their own GPU before relying on it. **The problem:** Headline TFLOPS come from one card and instruction benchmarks, not whole applications. **The outcome:** A matched before and after comparison on the same kernel, driver and power cap. ### Steps 1. Measure stock throughput first, using the CUDA instruction benchmark in diagnostics/bench and one real workload of yours 2. Note your kernel, driver version and power cap, and keep all three the same for both runs 3. Enable the unlock with `gpu_unlock.targets=` and power off and on 4. Confirm the `gpu_unlock: unlocked` line, then rerun the same benchmarks 5. Compare FP16, BF16, INT8 and FP32 results against the testing guide in docs/testing.md ### Prompt to paste into your agent ```text Using JustAResearcher/gpu-unlock-ada on , run the diagnostics/bench instruction benchmark and in stock mode, then again after a cold boot with the unlock enabled, with the same kernel, driver and power cap. Report a before and after table and confirm the kernel log shows `gpu_unlock: unlocked`. ``` ### Caveat Results so far are single runs on single cards with no extended stability test; INT8, dp4a and scalar FP32 stayed near baseline. Risk, reverse engineering: The unlock runs a payload derived from published analysis of NVIDIA's debug booter and changes GPU register state from a kernel module before the driver loads. It is unofficial, unsupported by NVIDIA, may void your warranty, and has been validated on very few cards. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/justaresearcher-gpu-unlock-ada/benchmark-stock-vs-unlocked-tensor-throughput/ ## How to test the experimental RTX 4070 Ti SUPER or 4060 Ti path with GPU Unlock Ada Repo: [JustAResearcher/gpu-unlock-ada](https://github.com/JustAResearcher/gpu-unlock-ada) | Difficulty: Advanced | Time: 1 hour | Basis: From the README Opt in to the experimental AD103 or AD106 target with its flag and one exact PCI address. **Who it is for:** An RTX 4070 Ti SUPER or RTX 4060 Ti 16GB owner willing to be an early tester on Linux. **The problem:** Only one card of each model has been validated, so the module never selects them by default. **The outcome:** A tested unlock on your exact card, or a clean refusal in the kernel log. ### Steps 1. Install with `make install` as for the RTX 4090 2. For an RTX 4070 Ti SUPER, add `gpu_unlock.targets=0000:01:00.0 gpu_unlock.ad103_experimental=1` with your own address 3. For an RTX 4060 Ti 16GB, add `gpu_unlock.targets=0000:01:00.0 gpu_unlock.ad106_experimental=1` with your own address 4. Power the machine off and start it again 5. Run `journalctl -k -b | grep gpu_unlock` and remove `gpu_unlock.targets=` from the kernel command line to disable it ### Prompt to paste into your agent ```text Install JustAResearcher/gpu-unlock-ada and enable the experimental path for my at with its experimental flag, keeping a stock boot entry. After a cold boot, report the gpu_unlock kernel log lines and whether the card unlocked. ``` ### Caveat targets=all never selects these cards, other Ada models are not accepted, and shared firmware does not mean an untested card will work. Risk, reverse engineering: The unlock runs a payload derived from published analysis of NVIDIA's debug booter and changes GPU register state from a kernel module before the driver loads. It is unofficial, unsupported by NVIDIA, may void your warranty, and has been validated on very few cards. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/justaresearcher-gpu-unlock-ada/test-experimental-rtx-4070-ti-super-4060-ti-unlock/