Quantus is a proof-of-work Layer 1 whose selling point is that its cryptography is post-quantum from block one: ML-DSA signatures, the NIST standard finalised in 2024, and a hash-based proof of work (QPoW, on Poseidon2) rather than SHA-256. Its mainnet went live on 9 September 2026, open to miners from the first block, with a supply capped at 21 million QTC.
It mines on ordinary NVIDIA GPUs. We rented an RTX 4090 on Paralon, set it up against the largest pool, and measured. This guide is what we did, every parameter included, and the numbers we saw. As with the Pearl guide before it, this is a technical explainer, not financial advice: everything below moves block to block.
What you need
- A QTC address, which starts with
qz…. The official mobile wallet (iOS and Android) creates one; so does thequantus-clifrom the project's GitHub releases withquantus wallet create --name pool. - An NVIDIA GPU with a CUDA driver. A rented JupyterLab on Paralon is exactly that: you get a terminal in a container with the card attached and outbound internet. No Docker is needed and none is available inside the rental; the miner is a single binary.
- Outbound UDP. The pool speaks QUIC. We had no trouble from a rental; a corporate network that blocks UDP would.
Pool or solo
Solo mining means running your own Quantus node (about 500 GB of disk, port 30333 open to the internet) and an external miner talking to it on a private port. You are paid only when you find a block. On a single card, against a network that already had ~25 TH/s of hashrate two days after launch, that is a lottery ticket.
Pool mining means the miner connects outbound to the pool's node, you are
credited for shares, and the pool pays your address every hour once you
pass its threshold. That is the only option that makes sense from a rented
GPU, which has no inbound connectivity anyway, so it is what this guide
covers. The solo path is documented in the project's MINING.md if you
want it on a machine you own.
The pool parameters
We used quanpool (quanpool.com), which at the time of writing carries most of the network's hashrate. Its connection details are published on its site and API, and this is what the miner needs:
| parameter | value |
|---|---|
| endpoint (pool) | 37.187.143.115:9834, hostname mainnet.quanpool.com, QUIC over UDP |
| endpoint (solo through the pool's node) | 37.187.143.115:9844 |
| TLS certificate SHA-256 | 87dc37af6096a3ddc860b94368ca087775f3ad3e0c4e9bcff3b07ea08d8abef6 |
| authentication | your qz… address, optionally address.rigname to name the machine |
| pool fee | 1%, PPLNS |
| payout | hourly, once pending balance exceeds the pool's threshold |
| miner | the pool's own CUDA build, quanpool-miner, version 6.0.0 at the time of writing |
Two things to know about the miner. It is the pool's own build, not the
stock quantus-miner from GitHub, and it is much faster on NVIDIA (the
pool's published benchmark for an RTX 4090 is 1,123 MH/s versus 178 MH/s
for the stock miner). It also carries a 5% developer fee: three minutes
of every hour mine for the miner's author. That is on top of the pool's 1%.
The stock miner has no such fee and works against the same endpoint; on a
4090 you would still come out far ahead with the fast one.
Check the pool's page for the current values before you start; the TLS fingerprint in particular changes if the pool rotates its certificate.
Setup on a rented GPU
Rent a GPU on the Rent marketplace with JupyterLab access, open a terminal in it, and:
# 1. the pool's miner (Linux x86_64)
curl -fsSL -o quanpool-miner https://download.quanpool.com/quanpool-miner-6.0.0-linux-x86_64
chmod +x quanpool-miner
# 2. hashrate of this card, nothing sent to the pool
./quanpool-miner benchmark --gpu-devices 1 --cpu-workers 0
# 3. mine
./quanpool-miner serve \
--node-addr 37.187.143.115:9834 \
--auth-token qzYOURADDRESS.rental-1 \
--tls-cert-sha256 87dc37af6096a3ddc860b94368ca087775f3ad3e0c4e9bcff3b07ea08d8abef6 \
--gpu-devices 1 --cpu-workers 0
--cpu-workers 0 keeps the CPU out of it; on a rental the CPU is shared and
contributes a rounding error next to the card. To keep the miner running
after you close the notebook tab:
nohup ./quanpool-miner serve ... > miner.log 2>&1 &
tail -f miner.log
For a second machine, change the rig name after the dot. The pool groups every rig under the one address and pays once.
If you run it on a machine with Docker instead, the same binary works in a
container built on an nvidia/cuda base image with the NVIDIA container
toolkit; the flags are identical.
What we measured
RTX 4090, rented on Paralon, quanpool-miner 6.0.0, GPU only:
| hashrate | 1.17 GH/s, steady, engine cuda/gpu |
| power | 501 W, at the card's 500 W cap |
| temperature | 71 °C at 59% fan |
| clock | 2,745 MHz |
| first accepted share | within the first minute |
| shares in the first 3 minutes | 47, 0 invalid, 0 stale |
That is slightly above the pool's own published figure for the card.
We also ran it on a rented 8x RTX 3060 mining rig, same miner, all eight
cards (--gpu-devices 8): 1.47 GH/s in total, 181 to 186 MH/s per card,
each at 107 to 148 W and 58 to 70 °C. Eight entry-level cards out-hash one
4090 here, which matters once you put a rental price next to the number
(below).
The pool's per-address hashrate lags: for the first ten minutes it showed us 16 MH/s, because it estimates from accepted shares over a window, and the window was mostly empty. The miner's own display is the number to trust until the pool catches up.
The math, so you can do your own
Everything you need is public on the pool's stats API and changes hourly. On 10 September 2026, the day after launch:
| network hashrate | ~25.4 TH/s |
| block time | ~8.6 s measured, 6 s target |
| block reward | 3.0 to 3.2 QTC |
| pool share of network | ~72% |
An RTX 4090 at 1.17 GH/s is 0.0046% of 25.4 TH/s. At roughly 10,000 blocks a day paying about 3.1 QTC each, that is about 1.4 QTC per day gross, about 1.3 after the 1% pool fee and the 5% miner fee. Multiply by whatever QTC is worth on the day, subtract your rental cost per hour times 24, and you have your answer for that day. Two things will move it:
- Network hashrate only goes one way after a launch. Every card that joins takes a proportional slice from yours. The 25.4 TH/s figure was measured 30 hours after mainnet went live; do not plan on it.
- The price. QTC has been tradable for a matter of days. We have no view on it and this guide does not offer one.
The two machines we measured, at the marketplace's list prices the day we ran them, needed QTC to be worth roughly $6 to $7 for a day of mining to cover a day of rent, before any change in network hashrate. The 8x3060 rig gave the better hashrate per dollar of the two.
The rental cost side is the one number you control. Paralon bills by the minute at a price each GPU's owner sets, so the honest way to decide is: rent one card, run it for an hour, read your pending balance on the pool, and compare against what the hour cost. Then scale, or stop.
Things that bit us
- The mobile wallet still pointed at the testnet the day after launch, and the block explorer's account page crashed on an address with no history yet ("Cannot mix BigInt and other types"). The pool's own page for your address, reached by searching it on quanpool.com, was the reliable place to watch shares and pending balance. The pool says as much in its announcement banner.
- The solo node needs a different kind of address. The node pays
rewards to a "wormhole" address derived from a secret preimage
(
quantus-node key quantus --scheme wormhole), not directly to aqz…wallet. If you go solo on your own machine, that preimage is the key to your rewards; keep it like one. lshides.env. Not Quantus-specific, but if you keep your address in a dotfile on the rental and cannot see it,ls -a.
Ready to try it?
Rent a GPU with CUDA on the Rent marketplace,
paste the three commands above with your qz… address, and read your first
accepted share within a minute. Hourly rates per card are on the
pricing page; there is no minimum and
nothing to cancel, so an hour of measurement costs exactly an hour.
