OPEN COMPUTE EXPERIMENT · AMD GPU NETWORK

AI COMPUTE POWERED
BY THE PEOPLE.

Putting overlooked GPUs to work. Boats N Hoes is an open-source experiment exploring useful AI compute, starting with two Radeon GPUs and a mission to make distributed computing practical, transparent, and verifiable.

EARLY-STAGE PROTOTYPE · HARDWARE VALIDATION IN PROGRESS

◈

Start Your AI Journey

Explore practical AI workloads, model inference, and the tools behind an open compute experiment.

✣

Contribute Your GPUs

Help test how accessible hardware can participate in coordinated compute jobs. Worker onboarding is on the roadmap. Check your GPU →

▣

Build in the Open

Inspect the source, follow development, reproduce tests, and help us publish real performance results.

01 / The mission

Compute first.
Hype never.

Boats N Hoes explores a small, auditable path from local GPU inference to coordinated useful-compute jobs. We publish what works, what breaks, and what remains unproven.

01 — HARDWARE

Start with what we have

Reported setup: Ryzen 5 5600X, 32 GB RAM, Ubuntu, and two RX 470 cards with different VRAM capacities.

02 — USEFUL WORK

Run real inference

Test a small Vulkan-capable workload on each GPU independently. Measure memory, speed, temperature, and output quality.

03 — VERIFICATION

Check before counting

Validate task outputs, reject duplicate jobs, and independently audit selected results before recording accepted work.

02 / Proposed architecture

A simple stack.
One layer at a time.

This is the intended design, not a claim that a distributed network is already running.

01
Job coordinatorCreates and assigns bounded tasks
↓
02
GPU worker agentsIndependent RX 470 workers
↓
03
Verification engineTask checks, duplicate detection, audits
↓
04
Off-chain reward ledgerPrototype accounting only
03 / Project status

Transparent by default.

No fake live counters. These labels reflect the current documented state of the repository.

DesignDocumented
2 GPUsReported hardware
PendingVulkan validation
Not liveNetwork / rewards
Important: No working miner, validated GPU inference stack, live compute network, deployed token, audited contract, live rewards, or guaranteed payouts are claimed. The RX 470 cards and their software compatibility still need to be tested on the actual machine.
04 / Build sequence

From bench test
to proof of concept.

Milestones are deliberately ordered to prove the hardware before adding infrastructure or token complexity.

M0

Inventory the machine

Capture Ubuntu, kernel, PCI IDs, drivers, Vulkan devices, and permissions.

M1

Run inference per GPU

Complete a small Vulkan smoke test on each card and record benchmark results.

M2

Build the worker agent

Registration, signed job retrieval, execution, receipts, retries, and logs.

M3

Build coordinator + verifier

Job lifecycle, deterministic checks, duplicate rejection, audit sampling, SQLite ledger.

M4

Prove two-worker operation

Run both GPUs independently; test restarts, malformed jobs, and throughput.

M5

Simulate rewards

Calculate hypothetical credits from accepted work. No real payouts.

M6

Consider testnet settlement

Only after the compute prototype works: capped test token, tests, and independent review.

NEXT

Contribute and reproduce

Read the hardware notes, run the checks, and publish reproducible results.

OPEN SOURCE / EARLY STAGE

Don't take the word for it. Check the work.

Read the prototype notes, inspect the code, and help turn the first hardware test into a reproducible result.

Go to GitHub ↗