ProjectsAugust 9, 2026 · updated August 29, 2026

XDNA NPU Miner

Evidence & reproducibility

Integrity 67% · 4/6 evidence signals

RESEARCHRESEARCH

The project does not claim mining performance yet. Its current evidence is the explicit CPU-reference-first validation plan and the requirement to prove actual NPU execution before publishing acceleration results.

Machine-readable checks

✓ Pinned repository commit exists
✓ CPU reference is required before NPU benchmark claims
◇ End-to-end mining performance demonstrated

Evidence

Trusted CPU reference required for candidate kernels
NPU execution proof required before acceleration claims
Correctness is checked before benchmark interpretation

Environment

accelerator: AMD XDNA1 / AIE2
runtime: MLIR-AIE / IRON / XRT
host: Linux

Last verified: 2026-08-29Verified commit 85e623b1 ↗

Reproduction

Follow the repository preflight and validation milestones at the pinned commit before interpreting any acceleration claim.

Experiment ledger

1 recorded experiment

EXP-2026-08-10-PREFLIGHT · 2026-08-10
Testnet preflight handoff
PARTIAL
Hypothesis: The project can define a measurable validation path before making any mining-performance claim.Pinned the repository milestone that finalizes the preflight handoff while keeping the implementation explicitly experimental.
- CPU-reference-first validation remains a requirement.
- Actual NPU execution proof remains a gate for future acceleration claims.
Experiment commit 85e623b1 ↗

Status

This project is deliberately early. Mining functionality is not implemented yet. The current milestone is repository bootstrap and technical validation: checking the upstream Qubic algorithm and protocol, licensing constraints, and which compute kernels are actually suitable for AMD XDNA1 acceleration.

Target platform

• AMD Ryzen AI Hawk Point• XDNA1 / RyzenAI-npu1• AIE2 four-column array• Fedora Linux• MLIR-AIE / IRON / XRT

Research loop

Evidence-first workflow

Upstream algorithm
CPU reference
Candidate kernel
NPU execution proof
Correctness
Benchmark

Every NPU kernel should have a trusted CPU reference, and NPU execution or benchmark claims need hardware evidence. That makes the repository useful before a miner exists: architecture, milestones, test strategy, upstream research, and the decision log remain explicit.

Limitations

No mining performance claim is made while the implementation is incomplete. The experiment is specifically about whether suitable kernels can be mapped honestly and measurably onto XDNA1. Follow the experiment on GitHub.

Related projects

Hawk Point NPU LLM

Active

PythonC++ / AIE2MLIR-AIEXRTOpenAI API

★ 9 stars⑂ 0 forks◌ 63 commits◉ 0 issues

PythonUpdated Sep 17, 2026

Running small transformer models directly on first-generation AMD XDNA NPUs with MLIR-AIE/IRON, plus an OpenAI-compatible local API.

Read case study View project

Samsung Odyssey G5 Firmware Research

Research

PythonSPARC V88051FirmwareReverse Engineering

★ 17 stars⑂ 0 forks◌ 20 commits◉ 0 issues

PythonUpdated Aug 27, 2026Release v1.0.0

Reverse engineering and reproducible tooling for the Samsung Odyssey G5 G55C firmware, including a minimal patch that exposes an existing hidden factory page.