$GPUFARM IS LIVE
GPUFarm
Become a host

Plug in a GPU.
Put it to work.

Run the GPUFarm host daemon on your rig. Every GPU is verified by a real benchmark, jobs run sandboxed, and you're paid for verified GPU-seconds through escrow on Robinhood Chain.

Install the daemon Estimate earnings
No machines on Robinhood Chain yet. Yours can be the first.
From box to earning
4 steps
  1. 01Pair
    $ gpufarm-host pair --server https://robingpu.farm

    Generates the daemon's own host key (kept in your OS keystore) and prints a code. Approve it in the browser while signed in.

  2. 02Benchmark
    $ gpufarm-host benchmark

    Server-issued challenge per GPU: FP32/FP16 matmul with checked results, VRAM bandwidth and a 30 s stability run. Raw numbers vs. the class reference decide the tier.

  3. 03Set pricing

    Price per GPU class in the dashboard, with the network's real median prices next to yours. Add an autoscheduling policy: max temp, hours, minimum rate.

  4. 04Go online
    $ gpufarm-host run

    Heartbeats every 10 s, receives job offers, runs each workload in a sandbox on its reserved GPUs, uploads the result hash and a signed receipt.

What you need

The daemon checks each of these at startup and reports what it finds on your dashboard.

An NVIDIA GPU + driver

Any CUDA GPU the driver reports: RTX 30/40/50 series, A100, H100, L40S and more. nvidia-smi must work. AMD/ROCm comes later.

Docker + NVIDIA Container Toolkit

Customer jobs run as containers with GPU access only to their assigned devices. The toolkit is what passes the GPU through.

gVisor (optional)

A stronger sandbox runtime. Used where GPU compatibility allows; recommended for hosts who want an extra isolation layer.

Outbound internet only

The daemon dials out to GPUFarm (HTTPS / WebSocket). No inbound ports, no port forwarding, and your IP is never published.

A wallet

Sign in with a message signature, approve your machines and receive payouts. The daemon never touches this wallet's keys.

Linux, Rust to build

Linux x86_64 or aarch64 with a recent kernel is the supported host. Building from source needs Rust 1.88 or newer.

Install

Two ways to get the binary. Either way, you can read and verify exactly what runs on your machine. Linux x86_64 or aarch64; the full guide is in the host docs.

Build from source

Available now

From a checkout of the GPUFarm repository (the daemon is the Rust crate in host/):

build
# Rust 1.88+ (https://rustup.rs)
$ cd host
$ cargo build --release --locked
$ sudo install -m 0755 target/release/gpufarm-host /usr/local/bin/gpufarm-host
$ gpufarm-host version
$ gpufarm-host detect

Signed release

Not published yet

Each release publishes the binary, its SHA-256 and an Ed25519 signature over gpufarm-host <version> <sha256>. Check both yourself before anything runs.

Signed releases aren't hosted yet. When they are, the release-signing public key appears here and these exact steps verify a download. Until then, build from source.
verify a release by hand
# <release host>/gpufarm-host/<version>/: gpufarm-host-<target> plus its .sha256 and .sig
$ VERSION=0.1.0 TARGET=x86_64-unknown-linux-gnu
$ sha256sum -c gpufarm-host-$TARGET.sha256
# the signature covers the exact bytes "gpufarm-host <version> <sha256>" (OpenSSL 3+)
$ printf 'gpufarm-host %s %s' "$VERSION" "$(sha256sum gpufarm-host-$TARGET | cut -d' ' -f1)" > msg
$ xxd -r -p gpufarm-host-$TARGET.sig > sig.bin
$ printf '302a300506032b6570032100%s' "<release key hex>" | xxd -r -p | openssl pkey -pubin -inform DER -out gpufarm-release.pem
$ openssl pkeyutl -verify -pubin -inkey gpufarm-release.pem -rawin -in msg -sigfile sig.bin
# only after "OK" and "Signature Verified Successfully":
$ sudo install -m 0755 gpufarm-host-$TARGET /usr/local/bin/gpufarm-host
or: the installer (same checks, plus service user + systemd unit)
# download it, read it, then run it; never pipe it into a shell
$ curl --proto '=https' --tlsv1.2 -fsSLO https://<release host>/gpufarm-host/install.sh
$ less install.sh
$ sudo sh install.sh --base-url https://<release host>/gpufarm-host --public-key <release key hex>
No curl | sh one-liner
There is no curl … | sh installer, on purpose: piping a script into a shell skips every check above. The installer is downloaded, read, then run, and it refuses to install unless both the SHA-256 and the signature verify. Updates are held to the same rule: gpufarm-host update checks the signature before downloading and the hash after — no signature, no install.

Pair, verify, go online

  1. 01Pair

    Generates the daemon's own host key (kept in your OS keystore) and prints a code. Approve it in the browser while signed in.

    terminal
    $ gpufarm-host pair --server https://robingpu.farm
  2. 02Benchmark

    Server-issued challenge per GPU: FP32/FP16 matmul with checked results, VRAM bandwidth and a 30 s stability run. Raw numbers vs. the class reference decide the tier.

    terminal
    $ gpufarm-host benchmark
  3. 03Set pricing

    Price per GPU class in the dashboard, with the network's real median prices next to yours. Add an autoscheduling policy: max temp, hours, minimum rate.

    Pricing in the dashboard
  4. 04Go online

    Heartbeats every 10 s, receives job offers, runs each workload in a sandbox on its reserved GPUs, uploads the result hash and a signed receipt.

    terminal
    $ gpufarm-host run

Developing locally? gpufarm-host run --simulate runs a fixture host on the local chain only (31337): it executes the selftest workload and is labeled SIMULATED everywhere.

Earnings estimator

Built from the prices hosts on Robinhood Chain list right now, not a marketing number. The assumptions are spelled out next to the result.

$

No market data yet: no GPU on Robinhood Chain has a listed price, so there is nothing honest to estimate from. Enter your own rate to see the math.

4
50%

Share of the time your GPUs are running paid jobs. Not guaranteed: it depends on demand, your price and your measured reliability.

Estimated gross revenue
—/ 30 days
Pick a GPU or enter a rate
  • · Rate: your own rate
  • · Utilization 50% is the number you chose above.
  • · Gross: rate × billable GPU-hours — you receive your full rate. The platform fee (2.50%) is added on top of the customer's bill and goes to the protocol fee recipient, not out of your earnings.
  • · If your machine runs in someone else's farm, that farm owner's commission (registered onchain, max 20%) comes out of this. Electricity, hardware and internet costs are yours and not included.
Trust & security

Your machine stays yours.

Customer code is untrusted by default. The daemon isolates every workload and never lets it near your keys, your network or your files.

Trust modelSecurity docsHost docs
  • Sandboxed containers

    No privileged mode, no host PID or network namespace, no Docker socket, read-only base filesystem, CPU/RAM/runtime limits.

  • Only its own GPUs

    A job sees exactly the GPU UUIDs it reserved. Outbound network is off unless the customer requested an allowlist, and LAN ranges are never allowed.

  • Keys out of reach

    The host key lives in your OS keystore and only signs logins, benchmarks and receipts. It holds no funds; your wallet keys never touch the daemon.

  • Honest verification

    Results are labeled Result verified, Execution verified or Host reported, never all “verified”. Benchmarks are checked against reference ranges.