MeshMiner 0.8.2

MeshPool Miner 0.8.2

Changes since 0.7.5 — a new coin with two algorithms, one miner that runs
both at once, a faster Plaine, and two fixes that were quietly costing people
shares.

---

New coin: Avian (AVN)

Avian mines two algorithms against one chain at the same time. A block goes
to whichever side finds it first, and both pay the same wallet.

Flag	Hardware	Port	Speed
--algo x16rt	Graphics card (NVIDIA)	3344	~41 MH/s on an RTX 3080 Ti
--algo minotaurx	Processor	3345	~32 kH/s on 32 threads of a Ryzen 9 9950X3D

No ASICs exist for either, and stock miners do not implement X16RT.

Fee: 0.5% on MeshPool, 1% elsewhere. Printed in the startup banner.

---

Dual mining: one miner, two algorithms

--cpu-algo turns it on. The card mines one algorithm and the processor mines
another, from a single process, at the same time.

• --cpu-pool and --cpu-user set the second stream's pool and address. Both
  default to the first stream's.
• The two streams are independent: different coins and different wallets
  work, and if one pool goes down the other keeps mining.
• This is mainly for rented servers, which load only one container image.
  Without it the processor half of the machine is wasted.
• Share lines and the stats table tag the streams s1 and s2. Read those
  rows, not the total — adding a card's hashrate to a processor's gives a
  number that means nothing.

---

Plaine (Isochron) is faster

• It is checked before it is used: the chain's own test vectors plus 64 random
  programs run through both at startup. One mismatch and the miner falls back to
  the chain's code, mines at the old speed, and says so.
• Batch size now defaults to 32 instead of 8. On a 128-thread machine that is
  roughly four times the hashrate; on a desktop it changes little.
  --batch N sets it by hand.

---

Fixes

• Shares after a reconnect were thrown away. After a pool restart the miner
  kept the old extranonce1, and every share it sent from then on was quietly
  discarded. A rig that had reconnected once looked fine and earned nothing.
• Blocks under the share target were never sent. Hashes that met the network
  target but not the pool's share target went unsubmitted. On Dragonchain that
  threw away nineteen of every twenty blocks found.

---

Display

• Hashrate prints with units: 4.40 GH/s instead of 4396536593.82 H/s.
• The hardware report names the processor model, so the pool's Hardware page can
  rank processors the way it ranks cards. No rig names or addresses are sent.

---

Removed

--algo oms. ONE MILLION SATOSHIS was delisted from MeshPool on 2026-09-17. The
flag is refused rather than mining a chain the pool no longer serves.

---

Commands

Avian, both algorithms at once:

MeshMiner 0.7.5

MeshPool Miner 0.7.5 - changes since 0.7.4

NEW COIN: ONE MILLION SATOSHIS (OMS)

• --algo oms (aliases: blake2bv2, blake2b, onemillionsatoshis).
• Pool port on MeshPool: 3343.

CPU

• BLAKE2b on every thread, about 400 MH/s on 32 threads.

NVIDIA GPU

• New BLAKE2b kernel. --backend both runs the card and the CPU together.
• 4.56 GH/s on an RTX 3080 Ti, 40 registers, held at the card's 250 W cap.

DEV FEE

• OMS: 0.5% on MeshPool, 1.2% elsewhere, same as Lumenite. Address and
  rate print in the startup banner; covered by the tamper check.

PACKAGING

• start-oms.bat / start-oms.sh in the package; set WALLET and run.
• HiveOS Flight Sheet algo field accepts "oms" or "blake2b".

UNCHANGED

• Lumenite, Nonsense, Plaine and Dragonchain mine exactly as in 0.7.4.

MeshMiner 0.7.4

MeshMiner 0.7.4

Two new coins, both on the CPU backend.

New

• Plaine (Isochron): --algo isochron
  Add --backend both to let your GPU help out.
• Dragonchain (yespower): --algo yespower. CPU only.

Fees
Lumenite 0.5 % mesh / other 1.2 %
Plaine 0.5 % mesh / other 1.2 %
Dragonchain 0.5 % mesh / other 1.2 %
Nonsense 2 % mesh / other 7 %

Packaging

• Windows, Linux and HiveOS archives are now released side by side.
• Docker image: bawls/meshminer:0.7.3

Usage
meshpool-miner --pool host:port --user

. --algo isochron --backend both
meshpool-miner --pool host:port --user . --algo yespower --backend cpu

MeshMiner 0.7.2

MeshPool Miner 0.7.2

Two coins, fallback pools, a power cap, and a hardware report that helps us tune the miner for your card.

NEW IN THIS RELEASE

Mine Nonsense (NNN) as well as Lumenite

The miner now runs two algorithms. Pick one with --algo:

Nonsense (NNN) --algo fishhashplus NVIDIA GPU
Lumenite (LMT) --algo v12 NVIDIA GPU or CPU

FishHashPlus needs an RTX 20-series or newer with at least 6 GB of memory.
The first start builds a 4.5 GB dataset on the card, which takes about
fifteen seconds.

meshpool-miner --algo fishhashplus --backend cuda --pool pool.meshpool.net:3338 --user nonsense:YOUR_ADDRESS.rig1
meshpool-miner --algo v12 --backend cuda --pool pool.meshpool.net:3335 --user lmt1YOUR_ADDRESS.rig1

Fallback pools

--pool takes a list. The first is the one you want to be on; the next takes
over when it cannot be reached. Every minute the miner checks whether a pool
higher on the list is back, and moves up when it is.

--pool pool.meshpool.net:3338,backup.example.com:3338

Power cap

--power W sets a fixed board power limit on every mined card and puts the
original back when the miner exits. Needs administrator on Windows or root
on Linux, the same as fan control.

--power 250

On this algorithm a 3080 Ti holds full speed at 250 W and runs a good deal
cooler.

Hardware report

The miner tells the pool what your card is doing: model, hashrate, power,
clocks and temperature, once a minute after connecting and then every five
minutes. Nothing changes on screen. The pool uses it for the public Hardware
page, which shows how each card model performs per algorithm, and it is how
we will tune the miner for cards we do not own ourselves. --no-telemetry
turns it off.

PACKAGES

meshminer-win-0.7.2.zip Windows
meshminer-linux-0.7.2.tar.gz Linux
meshpool-miner-0.7.2.tar.gz HiveOS custom miner
bawls/meshminer:0.7.2 Docker, run with --gpus all

Each package has a start script per coin. Open it, put in your wallet
address, save, run. check-gpu verifies that your card computes correct
hashes for both algorithms.

HiveOS: Flight Sheet, custom miner, installation URL pointing at the
tarball, miner name meshpool-miner-0.7.2, algo fishhashplus or v12, wallet
template %WAL%.%WORKER_NAME%.

FEES

Nonsense 2% on MeshPool 7% elsewhere
Lumenite 0.5% on MeshPool 1.2% elsewhere

The fee address and rate are printed in the banner at every start.

REQUIREMENTS

NVIDIA RTX 20-series or newer with a current driver. No CUDA toolkit needed.
GTX 10-series and older are not supported. Lumenite also runs on any 64-bit
CPU.

MeshMiner 0.6.9

MeshMiner 0.6.9 — Windows, Linux, HiveOS and Docker.

A HomeScrypt miner for Lumenite (v1.1 and v1.2). NVIDIA GPU, CPU, or both.

Downloads

• Windows: meshminer-win-0.6.9.zip
• Linux: meshminer-linux-0.6.9.tar.gz
• HiveOS: meshpool-miner-0.6.9.tar.gz
• Docker: bawls/meshminer:0.6.9

Dev fee 1.2%, or 0.5% on meshpool.net, unchanged and shown in the startup banner.

---

One more thing for the autotune to measure

The miner does its work in phases, and two of them have always run together
because they were written together — not because they belong together. They can
now run apart, each sized for what it actually does.

Whether that helps depends on the card, so the autotune measures it and
decides. On both cards we can test here it decided not to, by a margin too
small to be worth taking. That is a real answer and the tuner is allowed to
give it: where splitting buys nothing, the extra work is pure cost.

We are shipping it because the cards where it might pay are the ones we cannot
measure on. A 24 GB card runs roughly seven times denser than the 12 GB one
this was developed against, and the answer there is not ours to guess.

Upgrading

Nothing to configure. Your saved autotune still works, but this release adds a
setting it has never measured, so retune once:

meshpool-miner --retune

MeshMiner 0.6.8

MeshMiner 0.6.8 — Windows, Linux, HiveOS and Docker.

A HomeScrypt miner for Lumenite (v1.1 and v1.2). NVIDIA GPU, CPU, or both.

Downloads

• Windows: meshminer-win-0.6.8.zip
• Linux: meshminer-linux-0.6.8.tar.gz
• HiveOS: meshpool-miner-0.6.8.tar.gz
• Docker: bawls/meshminer:0.6.8

Set your wallet in the start script and run. Dev fee 1.2%, or 0.5% on
meshpool.net, unchanged and shown in the startup banner.

Requirements: NVIDIA RTX 20-series or newer for GPU, or any 64-bit CPU.
Windows 10/11 or 64-bit Linux.

---

The largest release yet, and it lands unevenly

Measured against 0.6.6 on the same machine, alternating between builds:

card	0.6.6	0.6.8
rented RTX 4090	5207	6149	+18%
RTX 3080 Ti desktop	2306	2481	+7.6%

Same card model, same clock, wildly different gain — and the difference is not
the graphics card. It is the processor driving it.

The miner prepares each batch of work on the processor before handing it to the
card. That preparation used to happen with the card sitting idle. On a 32-core
desktop it is 8 ms against a 232 ms kernel and reads as noise. On a rented host
with few cores it was 42 ms — the card was doing nothing for 15% of every
batch. It now happens while the card works on the previous batch, so it costs
nothing on any host.

If your processor is modest next to your graphics card, this release is worth
a lot to you. Rented instances, small hosts, an old CPU driving a big GPU.
If you run a fast desktop, expect the smaller number. Nothing is worse either
way and there is nothing to configure.

MeshMiner 0.6.6

MeshMiner 0.6.6 — Windows, Linux and HiveOS.

A HomeScrypt miner for Lumenite (v1.1 and v1.2). NVIDIA GPU, CPU, or both.

Downloads

• Windows: meshminer-win-0.6.6.zip
• Linux: meshminer-linux-0.6.6.tar.gz
• HiveOS: meshpool-miner-0.6.6.tar.gz

Set your wallet in the start script and run. Dev fee 1.2%, or 0.5% on
meshpool.net, unchanged and shown in the startup banner.

Requirements: NVIDIA RTX 20-series or newer for GPU, or any 64-bit CPU.
Windows 10/11 or 64-bit Linux.

---

+39% over 0.6.4

Measured on an RTX 3080 Ti, back to back on the same machine and pool:

	H/s
0.6.4	1652
0.6.6	2290

The proof-of-work is unchanged — same shares, nothing to update on the pool side.

Unlike 0.6.4, this does not depend on your card having spare memory, so it
should hold across the range rather than favouring big cards. Post your model
and your before-and-after if you run it — we have measured two cards and would
like more.

Autotuning measures the card instead of assuming

It used to search one setting and take two others on faith. It measures all
three now, and the two it assumed matter more than the one it searched.

Runs once per card, about half a minute, and is remembered. Every card in a
rig is measured on its own — cards that report the same model name are not
always the same card, so a reflashed CMP 170HX and a stock one are told apart
and tuned separately.

meshpool-miner --tune-only     measure every card and exit; no pool or wallet needed
meshpool-miner --retune        measure again

Retune after a driver update, a memory overclock, or a new miner version. A
stale result costs hashrate quietly.

--bench is gone. It printed a table for you to act on; the autotune now
measures the same thing and acts on it.

MeshMiner 0.6.4

WHAT IS NEW IN 0.6.4

Uses more of a big card's memory. The batch used to be sized by holding back
8% of whatever the card had free, and 8% is the wrong shape: the driver needs
a few hundred megabytes, and that need does not grow because the card is
bigger. On a 32 GB card the old rule discarded 2.4 GB -- about 150 concurrent
hashes -- and on a 64 GB card, 5 GB. A fixed reserve replaces it.

That is close to a straight hashrate gain, because on this algorithm hashrate
tracks the number of concurrent hashes almost exactly: on an RTX 5090,
doubling the batch returns 99% of double the hashrate. Expect roughly 4% on a
5090 and 5% on a 64 GB card. Cards of 8 to 12 GB are unchanged -- the old
reserve was about right for them, and it is kept.

Smaller launch block sizes are tried. The tuner's range now reaches 4, and on
a card with many idle SMs the smallest size keeps winning: measured 1% faster
than the previous best on an RTX 3080 Ti. The chosen size is remembered, so
delete

%LOCALAPPDATA%\meshpool-miner\tune.json

to measure again -- otherwise the size picked by an older version stands and
the new one is never tried.

The scratchpad is aligned to the memory system. Batches are rounded so that
no row of the scratchpad straddles an extra 32-byte memory sector, which the
fill and fold phases were paying for on most batch sizes.

It no longer fails to start when memory is briefly short. If something else
took memory between the miner asking how much was free and allocating it, the
miner exited and was restarted on a loop. It now backs off and starts a
little smaller.

--bench answers "would more VRAM help?" properly. It reports what the last
step of the sweep actually bought, instead of an average over the whole range
that reads as saturation when there is none, and it splits the time between
the fill and fold phases and the mix phase.

The proof of work is unchanged. Same digests, same shares. Upgrading is
replacing the binary and nothing else.

MeshMiner 0.6.3

MeshMiner 0.6.3 — Windows, Linux and HiveOS.

A HomeScrypt miner for Lumenite (v1.1 and v1.2). Runs on your NVIDIA GPU, your CPU, or both.

Downloads

• Windows: meshminer-win.0.6.3.zip
• Linux: meshminer-linux-0.6.3.tar.gz
• HiveOS: meshpool-miner-0.6.3.tar.gz

Set your wallet in the start script and run. Dev fee 0.5%, or 0.1% on meshpool.net — shown in the startup banner.

Requirements: NVIDIA RTX 20-series or newer for GPU, or any 64-bit CPU. Windows 10/11 or 64-bit Linux.

HiveOS packages now work

The 0.6.1 package could not start, and left no log to explain why. Three separate causes, all fixed:

• The folder inside the tarball did not match its filename, so anyone who typed the filename as the miner name got a directory Hive could not find. Folder, filename and CUSTOM_NAME are now all meshpool-miner-0.6.3 — copy the filename and it works.
• The scripts resolved paths against the working directory, which Hive does not promise. They now resolve against their own location.
• An empty Flight Sheet field produced --pool --user, which started the miner with the next flag as its pool address. It now names the field that is empty and stops.

Older NVIDIA drivers work again

The miner exited with CUDA_ERROR_UNSUPPORTED_PTX_VERSION on anything below CUDA 13.3. It never needed a driver that new; it runs on CUDA 11.0 and up, and if it ever happens again the message says what to do.

RTX 50-series

Already supported, and always was — the kernel is compiled by the driver for whatever card it finds. The wider block-size search from 0.6.0 matters most there: a 5090 has far more idle SMs than a smaller card, and the tuner can now reach the sizes that suit it.

HiveOS setup

Miner: Custom · Miner name: meshpool-miner-0.6.3 · Installation URL: the meshpool-miner-0.6.3.tar.gz link from this release
Pool URL: pool.meshpool.net:3335 · Wallet and worker template: %WAL%.%WORKER_NAME%

Leave "Extra config arguments" empty — it is only for extras like --device 0,1.

Upgrading: replace the binary, nothing else. The proof-of-work is unchanged — same digests, same shares.

MeshMiner 0.6.1 with hive

**WHAT IS NEW IN 0.6.1

Fixes a startup failure on older NVIDIA drivers. The miner would exit with
CUDA_ERROR_UNSUPPORTED_PTX_VERSION -- "the provided PTX was compiled with an
unsupported toolchain" and retry until it gave up. It needed a driver as
new as CUDA 13.3 for no reason; it now runs on anything from CUDA 11.0.
If it still happens, the message now tells you what to do about it.

Mines on every card in the machine. Until now it opened the first card and
only the first card, so a rig ran at one card's speed with all of them listed
in the banner. Each card gets its own row in the table and sizes its own
batch, so mixed rigs work without being told about it.

  --device 0,1     use only these cards (default: all of them)**