[ANN] Open Pearl Miner - open source miner for Pearlhash (Nvidia/AMD)

[Neilquick]

Open Pearl Miner (Nvidia/AMD)
A high-performance Pearl (PRL) proof-of-work miner for NVIDIA and AMD (beta) GPUs.

Mixed-GPU rigs are supported. Ampere/Ada cards run a fused int8 tensor-core kernel (GEMM + in-mainloop transcript fold + tensor-core noise generation), bit-exact with the Pascal DP4A path; Pascal cards run DP4A.

Features
AMD cards beta support
~7.0 TH/s sustained on a single Tesla P40 (Pascal DP4A).
~30 TH/s on a single RTX 4050 mobile (Ada) — fused int8 tensor-core GEMM + in-mainloop transcript fold + tensor-core noise generation, all bit-exact with DP4A.
~220 TH/s on RX 9900
Multi-GPU — auto-detects every GPU and runs one worker per card, near-linear scaling.
Continuous mining — no idle time waiting between pool jobs.
Background proof submission — finding a share never stalls the search.
Solo mining — mine to your local pearl-gateway node.
Pool mining — built for LuckyPool's Pearl stratum (default) with more pool protocols planned.
HiveOS — full HiveOS custom-miner package.
Dev fee - 1%
Available for Windows/Linux/Hive

Quick Start

Windows

Code:

p40-miner.exe --wallet prl1YOURWALLET --worker rig1

Linux

Code:

./p40-miner --wallet prl1YOURWALLET --worker rig1

That's it — it auto-detects all GPUs and starts mining.

Official links
Source: https://github.com/neilquicks/open-pearl-miner
Latest version: https://github.com/neilquicks/open-pearl-miner/releases/tag/v1.7.1

HiveOS
Download p40-miner-1.7.1.tar.gz and install it as a Custom miner:

Code:

Flight Sheet → Miner = Custom.
Set the Installation URL to the tarball, or scp it to the rig and run tar -C /hive/miners/custom -xzf p40-miner-1.6.1.tar.gz.
Wallet and worker: your Pearl wallet prl1... (worker name auto-appended).
Pool URL: pearl-eu2.luckypool.io:3360 (default LuckyPool GPU pool).
Extra config arguments (optional): --devices 0,1, --region 4096, etc.

The miner reports per-GPU TH/s and accepted shares to the HiveOS dashboard. Built against glibc 2.31, so it runs on both the focal (20.04) and jammy (22.04) HiveOS images and newer.

[Neilquick]

Open Pearl Miner v1.8.0 public release

What new
Added support dual mining ModelOS and Pearl (MDL+PRL)
Fix error on 6700XT
Support 9700XT cards

What changes
Intel Arc XMX — Implements an 8×8×32 INT8 cooperative matrix fold shader using SIMD8. Produces a bit-exact transcript through per-lane XOR combined with subgroupXor across four 8×8 fragments that form a single 16×16 hash tile. Validated against CPU results.

RDNA1 / Vega 56 (SWINT8) — Uses a software-emulated INT8 dot product by manually sign-extending operands and performing INT32 multiply-add operations when shaderIntegerDotProduct is unavailable. Shares a unified execution path with the DP4A implementation.

Architecture gating — Updated to use the exact RDNA2 device ID range (0x73BF–0x7447), allowing RDNA4 devices (0x7550) to bypass the RDNA2-specific path correctly.

Vendor detection — vendor_id is captured during initialization; Intel's vendor ID (0x8086) enables acceptance of the XMX 8×8×K matrix shape during the cooperative matrix capability probe.

Also
Available for Windows/Linux/Hive
Support Nvidia and AMD cards
Open source code
Dev fee 1%

Official links
Source: https://github.com/neilquicks/open-pearl-miner/
Latest version: https://github.com/neilquicks/open-pearl-miner/releases/tag/v1.8.0

[Neilquick]

Open Pearl Miner v1.9.0 Public release. High performance for middle cards

What new
Fused CuTe Ampere/Ada kernel (sm_80+). 128×256 CTA tile, in-mainloop transcript fold, multistage cp.async.cg + ldmatrix, <8,1,1> TiledMMA (each warp owns a 16-row band so the R-block fold is a direct in-register shfl_xor). Engaged automatically for the real mining shape (region multiple of 128×256, R multiple of 32). Measured 42.7 TH/s on an RTX 4050 (Ada), bit-exact with the DP4A transcript.
Dependency-free ldm fallback. Raw-PTX mma.sync.m16n8k32 with 16 N-fragments per warp (A fragment loaded once, reused ×16) for shapes the CuTe fold doesn't cover, plus a cascade of 64×256 / 64×128 tilings.
Pascal DP4A dual-tile (NT=2). Each warp now computes two adjacent 16×16 tiles, loading the A fragment once for both (shared-memory traffic per FLOP cut ~1.5×), with int2 global→shared staging. Bit-exact: identical dp4a inputs and order.
Leaner bundle. Excluded stray transitive deps (numba/llvmlite, pandas, lxml, pywin32) from the frozen build — back to ~30 MB.

Experimental
CUDA path builds clean for sm_61 / sm_86 / sm_89; the Windows DLL is smoke-tested on a Tesla P40 (Pascal DP4A, pow_split OK).
The Pascal NT=2 change is functional (P40 smoke test OK) but transcript equality on your card is worth a first-shares check as well.

Performance
The CuTe kernel benchmarked 42.7 TH/s bit-exact on the RTX 4050

Source: https://github.com/neilquicks/open-pearl-miner
Latest version: https://github.com/neilquicks/open-pearl-miner/releases/tag/v1.9.0

[Neilquick]

Open Pearl Miner v1.9.1 Public release

Main: fix continuous mining and AMD cards performance

What new
-

Code:

mine_job

now processes each job continuously. A pool job only defines the key (derived from the block header) and the target, while the A and B matrices are generated by the miner from a fresh Philox seed. As a result, a single job becomes an effectively endless stream of unique grids. The miner repeatedly generates a new grid and continues scanning the same job_id without any idle time, submitting every qualifying tile it finds. Mining only switches to a different job when a newer pool job arrives or when a scheduled dev-fee wallet rotation occurs.
-Corrected the TH/s display

Custom miner for HiveOS now available

Official links
Source: https://github.com/neilquicks/open-pearl-miner/
Latest version: https://github.com/neilquicks/open-pearl-miner/releases/tag/v1.9.1