DEDUPED Hot AI repos, turned into use cases Get the Monday 5

How to benchmark stock vs unlocked tensor throughput with GPU Unlock Ada

Run matched stock and unlocked measurements so you know what the unlock does for your card and workload.

Short answer: Measure stock throughput first, using the CUDA instruction benchmark in diagnostics/bench and one real workload of yours; then Note your kernel, driver version and power cap, and keep all three the same for both runs; enable the unlock with gpu_unlock.targets= and power off and on. All steps.

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.

Who it is forA developer who saw the viral 2x claim and wants numbers from their own GPU before relying on it.

The problemHeadline TFLOPS come from one card and instruction benchmarks, not whole applications.

The outcomeA matched before and after comparison on the same kernel, driver and power cap.

What you need first

  • A Linux machine you fully control, with a spare stock boot entry to fall back to
  • RTX 4090 (10de:2684); RTX 4070 Ti SUPER (10de:2705) and RTX 4060 Ti 16GB (10de:2805) are experimental and validated on one card each
  • Kernel headers and a working DKMS toolchain, with CONFIG_CRYPTO_LIB_SHA256 in the kernel
  • The stock NVIDIA driver
  • Secure Boot turned off, or the out-of-tree module signed with your own key
  • Your own stock benchmark numbers taken before you enable it

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 for your agent

Using JustAResearcher/gpu-unlock-ada on <my GPU>, run the diagnostics/bench instruction benchmark and <my workload> 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.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for JustAResearcher/gpu-unlock-ada and "Benchmark stock vs unlocked tensor throughput"

Add Deduped to Claude Code once

claude mcp add --transport http deduped https://dedupedai.com/mcp

Cursor, ChatGPT, others

Share: X LinkedIn Hacker News Email Copy link

Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer