⚡ quantus-miner-perf — 2.5× Faster Quantus Miner, Drop-in Replacement

English · 简体中文

Same protocol as the official miner. Just faster. CPU up to 2.6× · GPU up to 2.12× · split CPU/GPU reporting · lower host overhead.

Works with any pool, not just ours. This build speaks the official Quantus miner protocol — you can point it at any pool that supports the official miner, at your own node, or at a third-party pool. Nothing here is pool-specific: change --node-addr / --auth-token and you are mining somewhere else.

A high-performance Quantus external miner: midstate-optimized CPU engines, a native CUDA backend, split CPU/GPU reporting — same protocol as the official miner, drop-in replacement.

Headline numbers (median of 3, official vs perf, same machine, same flags — release v4.0.2-perf1):

machine	CPU 1 worker	CPU all-core	GPU	combined
AMD 7950X3D + RTX 3090 Ti (Linux)	179K → 432K H/s (2.41x)	3.15M → 7.94M H/s (2.52x)	164M → 347M H/s (2.12x)	170M → 352M H/s (2.08x)
Apple M1 Max (macOS, v4.0.2-perf1)	140K → 366K H/s (2.61x)	0.78M → 1.91M H/s (2.46x)	19.2M → 19.7M H/s (1.02x)	17.8M → 20.9M H/s (1.17x)

Full methodology, per-run data and engine analysis below. No source is distributed here; reproducible A/B script in scripts/bench-compare.sh.

Download

Grab the latest release: https://github.com/longcipher/quantus-miner-perf/releases/latest

Install / upgrade (recommended)

One-liner — detects OS/arch, picks the AVX2 (-simd) build on supported Linux CPUs, verifies SHA256SUMS, and skips reinstall when your local version is already ≥ the latest release:

curl -fsSL https://raw.githubusercontent.com/longcipher/quantus-miner-perf/master/scripts/install.sh | bash

# custom dir / force reinstall / pin a version
curl -fsSL https://raw.githubusercontent.com/longcipher/quantus-miner-perf/master/scripts/install.sh | bash -s -- --dir ~/.local/bin
curl -fsSL https://raw.githubusercontent.com/longcipher/quantus-miner-perf/master/scripts/install.sh | bash -s -- --force
curl -fsSL https://raw.githubusercontent.com/longcipher/quantus-miner-perf/master/scripts/install.sh | bash -s -- --tag v4.0.2-perf1

Version check is release tag vX ≡ quantus-miner-perf -V → … X (script updates only when the release is newer than local).

Manual download

asset	platform
quantus-miner-perf-*-macos-arm64.tar.gz	macOS Apple Silicon (M1/M2/…)
quantus-miner-perf-*-macos-x86_64.tar.gz	macOS Intel
quantus-miner-perf-*-linux-x86_64.tar.gz	Linux 64-bit (x86_64)
quantus-miner-perf-*-linux-x86_64-simd.tar.gz	Linux 64-bit with AVX2 (auto-picked by install.sh)
quantus-miner-perf-*-windows-x86_64.zip	Windows 10/11 64-bit

# macOS Apple Silicon example (replace * with the release tag)
tar xzf quantus-miner-perf-*-macos-arm64.tar.gz
./quantus-miner-perf benchmark --duration 10

macOS Gatekeeper: unsigned binaries are blocked on first run. Right-click → Open, or run xattr -d com.apple.quarantine quantus-miner-perf. Windows SmartScreen: click "More info" → "Run anyway" on first launch.

Quick start

Point it at your Quantus node (same flags as the official miner) — or skip running a node entirely with the LongCipher pool below. Any other pool that speaks the official miner protocol works the same way.

# Mine (CPU auto-detect + all GPUs)
quantus-miner-perf serve \
  --node-addr 127.0.0.1:9833 \
  --auth-token-file /path/to/miner-auth-token \
  --tls-cert-sha256-file /path/to/miner-tls-cert-sha256

# CPU-only, 8 workers, midstate engine (default)
quantus-miner-perf serve --node-addr 127.0.0.1:9833 \
  --auth-token-file /path/to/miner-auth-token \
  --tls-cert-sha256-file /path/to/miner-tls-cert-sha256 \
  --gpu-devices 0 --cpu-workers 8

# Check your own throughput
quantus-miner-perf benchmark --duration 15
quantus-miner-perf benchmark --cpu-workers 0 --gpu-devices 1 --duration 15  # GPU only

CPU engine selection (--cpu-engine, default midstate):

value	meaning
midstate	midstate-optimized engine (default, fastest)
fast	original upstream engine (bit-identical behavior to official)
unrolled	midstate + 4x loop unrolling (≈ midstate, see matrix below)

Connecting to the LongCipher Quantus pool

Solo mining normally needs a local node; the easy path is our sister pool — 1% fee, PPLNS or Solo, no signup (your qz… address is the account): https://quantus-pool.longcipher.com/

Two ways to connect. Both are protocol-identical to the official miner — pick whichever you prefer.

Mainland China users — use the relay. The direct pool IP 46.4.66.214 is blocked on the mainland, so a relay/proxy at 57.180.173.214 is provided for you. Point there if you mine from China; overseas users can connect directly to 46.4.66.214 (often faster). The TLS fingerprint is identical on both, so nothing else changes.

	endpoint	PPLNS	Solo
longpool	built in (defaults to relay 57.180.173.214)	--mode pplns	--mode solo
serve	manual	46.4.66.214:9900 (relay 57.180.173.214:9900)	46.4.66.214:9901 (relay 57.180.173.214:9901)

Option 1 — longpool one-liner (recommended, zero config)

longpool defaults to the relay 57.180.173.214 (the IP is baked into the binary), so mainland miners can run the commands below as-is — no endpoint to type. Overseas users who prefer to connect directly pass --pool-addr 46.4.66.214:9900 (PPLNS) or 46.4.66.214:9901 (Solo). The TLS fingerprint is the same on both, so you never copy it.

Mainland China — PPLNS (via relay 57.180.173.214, default — earnings split by shares):

./quantus-miner-perf longpool \
  --address qzYOURADDRESS \
  --worker rig1 \
  --mode pplns \
  --pool-addr 57.180.173.214:9900 \
  --cpu-workers 8 --gpu-devices 1

Mainland China — Solo (via relay 57.180.173.214; the finder takes the whole block minus the fee):

./quantus-miner-perf longpool \
  --address qzYOURADDRESS \
  --worker rig1 \
  --mode solo \
  --pool-addr 57.180.173.214:9901 \
  --cpu-workers 8 --gpu-devices 1

flag	meaning
--address	your QTC payout address — it is your mining account, no registration
--worker	optional rig label to tell machines apart; runs fine without it
--pool-addr	pool endpoint host:port. Defaults to the relay 57.180.173.214 (mainland-friendly). Overseas users pass 46.4.66.214:9900 (PPLNS) / 46.4.66.214:9901 (Solo) to connect directly

Verified end-to-end against the public endpoint: the pool logs miner connected … mode="pplns" remote=46.4.66.214:… and shares came back 43 accepted / 0 rejected.

Option 2 — serve (fully manual)

Every connection parameter set by hand. Use this for custom setups, your own node, or any other pool that speaks the official miner protocol.

--node-addr defaults to the direct endpoint 46.4.66.214 (overseas / faster). Mainland China users: replace it with the relay 57.180.173.214 — same ports 9900/9901, same fingerprint.

PPLNS (direct — overseas):

./quantus-miner-perf serve \
  --node-addr 46.4.66.214:9900 \
  --auth-token qzYOURADDRESS.rig1 \
  --tls-cert-sha256 8c700b8cd25a893f53f700f9650c4e96555989e2419a5b6241a1dcc43ed2feae \
  --cpu-workers 8 --gpu-devices 1

PPLNS (relay — mainland China):

./quantus-miner-perf serve \
  --node-addr 57.180.173.214:9900 \
  --auth-token qzYOURADDRESS.rig1 \
  --tls-cert-sha256 8c700b8cd25a893f53f700f9650c4e96555989e2419a5b6241a1dcc43ed2feae \
  --cpu-workers 8 --gpu-devices 1

Solo: change the port to 9901, everything else stays the same (direct 46.4.66.214:9901 or relay 57.180.173.214:9901).

./quantus-miner-perf serve \
  --node-addr 46.4.66.214:9901 \
  --auth-token qzYOURADDRESS.rig1 \
  --tls-cert-sha256 8c700b8cd25a893f53f700f9650c4e96555989e2419a5b6241a1dcc43ed2feae \
  --cpu-workers 8 --gpu-devices 1

--auth-token is always payout-address.worker-name, joined by a literal dot (qzYOURADDRESS.rig1).

Verified: shares submit normally — 16 accepted, 1 rejected in testing; the occasional reject is ordinary network jitter.

General notes

1. Binary compatible — swap the binary name for the one you downloaded; the flag set is identical to the official quantus-miner, so the two are drop-in replacements for each other.
2. Ports — --metrics-port defaults to 9900; if that clashes with something on your host, pass a different one.
3. Pool terms — 1% fee on both PPLNS and Solo; minimum payout 10 QUAN; 30 block confirmations.
4. Dashboard & latest values — https://quantus-pool.longcipher.com is the source of truth for ports, fee, and the current certificate fingerprint.
5. Platforms — macOS, Windows and Linux all supported.

Pointing at a different pool

Nothing above is LongCipher-specific. If another pool supports the official miner, reuse the Option 2 shape with its endpoint:

./quantus-miner-perf serve \
  --node-addr <pool-host>:<pool-port> \
  --auth-token <your-address>.<worker> \
  --tls-cert-sha256 <pool-cert-fingerprint> \
  --cpu-workers 8 --gpu-devices 1

Live stats, earnings calculator, blocks and payouts for our pool: https://quantus-pool.longcipher.com/.

Dev fee

This build carries a 3% dev fee: a small share of mining time is spent on the developer's node to fund continued optimization. Everything else pays your node. The fee never interrupts benchmarking and never affects reported throughput.

Why it is faster

• Midstate fast path: skips the final Poseidon2 squeeze unless the first-squeeze high bits can tie the target — 5 permutations/nonce → 3.
• Word-based resume streaming: low nonce half streamed from a u128 counter instead of re-serializing a full U512 per hash.
• AVX2 four-wide Poseidon2 resume on x86_64 (runtime-dispatched, scalar fallback on ARM — the Apple Silicon numbers above are scalar-only gains).
• Native CUDA backend on NVIDIA: u32-limb Goldilocks kernels, per-arch cubins and autotuned launch geometry — 163.88M → 347.45M H/s (2.12x) on the RTX 3090 Ti. Other platforms (e.g. Metal on Apple Silicon) fall back to wgpu.
• Lower host/launch overhead: amortized solution-flag polling, no unused double-buffer set, larger benchmark range floor (12M vs 1M nonces).

Benchmark report

Method: scripts/bench-compare.sh --repeats 3 — REF/OPT interleaved per case, 3 s cooldown, median reported. Both binaries release profile (opt-level=3, fat LTO), default batch sizes, random 32-byte header, U512::MAX difficulty (raw throughput, no solutions expected). Numbers below are from release v4.0.2-perf1.

Linux: AMD Ryzen 9 7950X3D + RTX 3090 Ti

Machine ai: AMD Ryzen 9 7950X3D (32 threads), 62 GB RAM, NVIDIA GeForce RTX 3090 Ti (driver 610.57.04, 24 GB). Both binaries miner-cli 4.0.2, rustc 1.98.1. Official source: upstream Quantus-Network/quantus-miner @ e0d7ac6.

case	official	perf	speedup
CPU 1 worker	179.31K H/s	431.65K H/s	2.41x
CPU 32 workers	3.15M H/s	7.94M H/s	2.52x
GPU 1 device	163.88M H/s	347.45M H/s	2.12x
combined (32 CPU + 1 GPU)	169.50M H/s	352.40M H/s	2.08x

CPU hashing is ~2.4–2.5x faster and GPU ~2.1x faster, so the combined run gains 2.08x instead of a few percent — unlike earlier releases, where the GPU was unchanged and dominated the total. The perf build also reports the CPU/GPU split that the official build hides.

Engine matrix (cpu1, median of 3):

--cpu-engine	median	vs fast
fast (== upstream engine)	180.50K H/s	1.00x
midstate (default)	432.80K H/s	2.40x
unrolled	432.80K H/s	2.40x

OPT+fast ≈ REF: the harness is at parity, the ~2.4x is purely the engine.

macOS: Apple M1 Max (current release v4.0.2-perf1, 2026-09-09)

Machine akmacstudio-2.local: Apple M1 Max (10 cores), 32 GB RAM, M1 Max GPU (24 cores, Metal), macOS 26.6.2 (arm64). Release profile (opt-level=3, fat LTO), rustc 1.100.0-nightly. Official miner-cli 4.0.2 @ e0d7ac6 vs perf quantus-miner-perf 4.0.2-perf1 @ 36228c9. Do not compare absolute H/s across the two tables (different compilers/hardware).

case	official	perf	speedup
cpu1 (1 worker, 8 s)	140.02K H/s	365.87K H/s	2.61x
cpu10 (10 workers, 12 s)	775.48K H/s	1.91M H/s	2.46x
gpu (1 device, 15 s)	19.20M H/s	19.67M H/s	1.02x
all (10 CPU + 1 GPU, 15 s)	17.84M H/s	20.86M H/s	1.17x

Per-run values (median of 3, REF/OPT interleaved, 3 s cooldown):

case	official reps	perf reps
cpu1	146.36, 130.96, 140.02K	373.76, 365.87, 323.70K
cpu10	727.10, 775.48, 822.26K	1.91, 1.73, 1.91M
gpu	18.99, 19.45, 19.20M	19.87, 19.67, 19.67M
all	18.71, 17.33, 17.84M	20.86, 20.83, 21.62M

The perf all run also reports the split the official build hides: median CPU 1.93M + GPU 19.04M ≈ 20.86M H/s.

Method: scripts/bench-compare.sh --repeats 3, median reported, same flags as the Linux table. Scalar-only gains on ARM (no AVX2 there) — x86_64 gets an extra SIMD boost on top. All-core and GPU figures are thermally sensitive on this integrated-GPU machine (see prior run below); single-worker CPU is the most stable read at ~2.6x in both runs. There is also no CUDA here — Apple Silicon runs the wgpu/Metal backend, so its GPU gain is a couple of percent, not the 1.59x an NVIDIA card gets.

Prior same-machine run (perf @ dd792af, miner-cli 4.0.2): cpu1 149.13K → 382.13K (2.56x), cpu10 891K → 1.55M (1.74x), gpu 19.91M → 21.08M (1.06x), all 20.46M → 21.63M (1.06x). Combined picture: CPU single-worker steady ~2.6x, all-core ~1.7–2.5x depending on thermal state, GPU ~2–6%.

🔗 Quantus official links

New to Quantus? Start here:

resource	link
🌐 Official site	https://www.quantus.com/
📖 Mining guide (docs)	https://docs.quantus.com/guides/mining/
📄 Whitepaper	https://www.quantus.com/whitepaper/
⛏️ Official miner (upstream)	https://github.com/Quantus-Network/quantus-miner
👛 CLI wallet	https://github.com/Quantus-Network/quantus-cli
🖥️ Node / chain	https://github.com/Quantus-Network/chain
🏢 GitHub org	https://github.com/Quantus-Network
🔍 Explorer (official)	https://explorer.quantus.com/
🔍 Explorer (blackbeard)	https://blackbeard.observer/
📡 Telemetry	https://telemetry.quantus.cat/
🐦 X / Twitter	https://x.com/QuantusNetwork

This repo is a community perf build — same miner protocol, just faster hashing. For consensus / wallet / network questions, the docs and repos above are authoritative.

💬 Community — LongCipher miners

Questions, hashrate screenshots, rig tuning? Come chat:

• ✈️ Telegram: https://t.me/longcipher
• 💬 WeChat: scan below to join the group

Solo or pool, official or perf build — all Quantus miners welcome.