# Welcome! Welcome to 170th Street — the most comprehensive community resource for the NVIDIA CMP 170HX, a misunderstood piece of hardware that deserves a second life. The CMP 170HX is built on the same GA100 silicon as NVIDIA's $10,000 A100 datacenter GPU. Sold in 2021 as a dedicated Ethereum mining card, it was deliberately crippled by NVIDIA to prevent general compute use — but the underlying hardware is extraordinary. With 1,493 GB/s of HBM2e memory bandwidth that beats even the RTX 4090, 42 TFLOPS of FP16 compute, and a PCB nearly identical to the A100, this card has enormous untapped potential. This documentation covers everything we know about unlocking that potential: * **Hardware** — full teardown, PCB analysis, comparison to the A100, and what NVIDIA removed and why * **Software Setup** — driver installation, power configuration, and clock tuning on Linux * **Performance** — honest benchmarks, what the FMA throttle actually costs you, and what workloads still shine * **Modifications** — VBIOS flashing, PCIe capacitor mod, watercooling installation, and NVLink research * **AI & ML** — LLM inference, FP16 workloads, the FMA workaround, and real-world usability * **Security Research** — firmware signing, the path toward unlocking FMA, and tools for investigation This is a living document. The goal is to turn a piece of discarded mining hardware into a viable budget compute platform — and eventually, to fully unlock the GA100 silicon inside. This project is community-driven and everyone is welcome to contribute. Whether you've done your own benchmarks, discovered a new workaround, successfully performed a hardware mod, or are researching the firmware — your findings belong here. Open a pull request, submit a change request, or reach out directly. No contribution is too small. The more people working on this, the faster we unlock everything this card can do. --- # Agent Instructions: Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter: ``` GET https://170th-street.gitbook.io/hx/readme.md?ask= ``` The question should be specific, self-contained, and written in natural language. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.