Bitaxe Naja Duo: The First Single-Digit Bitaxe

Every Bitaxe before this one has been a BM1370 machine or older. The Naja Duo is the first built on Bitmain’s 3nm BM1373 — and it arrives with a number no Bitaxe has posted before: 9.5 joules per terahash, at stock, out of the box. Single-digit efficiency used to be the exclusive territory of industrial hardware in a warehouse. It now sits on a desk, draws 42 watts, and runs firmware you can read on GitHub.

In brief: The Bitaxe Naja Duo (Model 1201) is an open-source Bitcoin solo miner built by Solo Satoshi in Conroe, Texas, using two Bitmain BM1373 3nm ASICs. It produces about 4.4 TH/s at 42 watts — 9.5 J/TH — and runs esp-miner with AxeOS, which supports Stratum V2 natively from version 2.14.0.

This guide covers the machine end to end, with every figure verified as of late August 2026: specifications, what the overclock actually costs, how to run it on Stratum V2, and a clear-eyed look at what 4.4 TH/s means when you point it at the Bitcoin network alone.

Key takeaways

• 9.5 J/TH at stock — the first Bitaxe under 10, against 15 J/TH on the Gamma 602 and 18 on the GT 801.
• Two chips doing four chips’ work. 4.4 TH/s from two BM1373s lands in the same class as a four-chip BM1370 board, at a fraction of the power.
• The overclock is real but expensive. 8 TH/s is tested and reachable — at 92 watts and 11.5 J/TH, meaning you trade away the efficiency that makes the board special.
• Native Stratum V2 through AxeOS 2.14+, including encrypted transport, pool authentication, and coinbase verification from the miner itself.
• Around $3.61 a month to run at default on $0.12/kWh power — roughly the cost of a coffee, for a machine hashing continuously.
• Genuinely open source. esp-miner firmware, published schematics, USB-C flashing, no vendor lock — which is what separates a Bitaxe from a marketplace clone.

What is the Bitaxe Naja Duo?

The name deserves a moment. “Duo” is the chip count — two ASICs, where a Gamma has one and a Hex has six. “Naja” follows the Bitaxe habit of naming boards after snakes. Model number 1201.

What makes it a landmark rather than another revision is the silicon. The BM1373 is the chip Bitmain built for the Antminer S23 generation, fabricated on a 3nm process, with eight hashing domains per chip. Every previous Bitaxe used the BM1370 or something older, on a larger process. Our BM1373 era article covers what that generational jump means across the whole home mining category; the short version is that a single BM1373 delivers around 2.5 TH/s at roughly 25 watts, which is the 10 J/TH figure the chip exists to hit.

Two of them on one board, tuned slightly below the chip’s nominal point, gives you 4.4 TH/s at 42 watts. That is the whole product thesis in one sentence.

Full specifications

All figures below are from the Solo Satoshi product listing for Model 1201, read 26 August 2026, published with a stated ±15% tolerance.

Performance at both operating points

Read those two columns against each other and the trade is unmistakable. Hashrate rises 82%. Power rises 119%. Efficiency degrades by a fifth. The overclock is legitimate engineering — the two-phase TPS546024A regulator was specified for exactly that 600 MHz, 1,120 mV profile, so this is a supported operating point rather than a gamble. But it converts the board’s headline advantage into an ordinary one. At 11.5 J/TH the Naja Duo is merely good; at 9.5 it is the most efficient desktop miner you can buy.

A transparency note, because we hold vendors to this standard and the rule applies here too: all figures above are Solo Satoshi’s own, and the company’s launch announcement on 23 August 2026 quoted 9.8 J/TH where its product page says 9.5. Both are bench numbers from the same company two days apart — a reminder that until independent wall-socket measurements exist for a board this new, every efficiency claim in the category deserves the same one-word suffix: claimed. The 24-hour meter test in the buying section below is how you produce your own number.

Which you should choose depends entirely on what you are optimising, and we come back to that below.

How does the Naja Duo compare to other BM1373 miners?

Efficiency claims mean little without neighbours to measure against. Here is the current desktop-class landscape, all figures at default settings.

Two comparisons carry the argument.

Against the Bitaxe GT 801: identical 42-watt draw, roughly double the hashrate. Same power budget, twice the lottery tickets. There is no configuration in which the older board wins that matchup.

Against the NerdQAxe++: more interesting, because the four-chip BM1370 board still produces more raw hashrate — 4.8 against 4.4 TH/s. It just needs 69.5 watts to do it, where the Naja Duo needs 42. Over a year of continuous running at $0.12/kWh that difference is roughly $29 in electricity, which will not decide many purchases on its own. What it does tell you is that chip generation now matters more than chip count, and that a two-chip board has caught a four-chip board from the previous generation. Our NerdQAxe++ guide covers that board in its own right.

Worth stating plainly, because marketing in this category is loose: the efficiency advantage evaporates if you overclock it away. Push a BM1373 hard enough and it lands back around 15 J/TH — precisely where BM1370 hardware already lives. Solo Satoshi, which builds the board, puts it more bluntly than we would dare — maxed out, in its own words, “you paid a next-gen premium for last-gen efficiency.” When the company selling the board warns you off its own overclock headroom, believe it. The honest generational gain of the BM1373 over the BM1370 is about a third — roughly 15 down to 10 J/TH — not the “up to 50%” that circulates, and the arithmetic behind spec-sheet inflation is the same one we work through in six mining specs that fail the math.

One forward-looking note: the firmware ecosystem already contains a quad-chip BM1373 board target of NerdQAxe++ class, and the chip’s supplier quotes 10–12 TH/s for that architecture. The dual-chip Naja Duo will not be the top of this family for long — which is not a reason to wait, only a reason to buy the board for what it does today rather than for its position in a ranking that resets every quarter.

How loud is the Naja Duo, and how does it stay cool?

The thermal solution is a Thermalright AXP90-X47 — a low-profile tower cooler from the PC world, spanning both ASICs on a single fin stack with heat pipes — paired with a 92 mm slim PWM fan rated 42.58 CFM at up to 22.4 dBA.

Those numbers are more meaningful than they look. 22.4 dBA is the fan’s maximum, and AxeOS only ramps toward maximum when chip temperature demands it; at the 42-watt default the fan spends most of its life well below that. For reference, a quiet library sits around 30 dBA. At stock settings this is a machine you can run in a room where people sleep.

Overclocked, that changes — 92 watts is more than double the heat, the fan works considerably harder, and the acoustic profile moves from unnoticeable to present. Anyone planning to run the 8 TH/s profile in a bedroom should audition it first.

The detail worth appreciating is per-chip temperature monitoring. Each BM1373 reports its own temperature, so throttling and fan curves respond to the hotter of the two rather than to an average that could hide a problem. On a dual-chip board this matters: uneven paste application or a partially blocked fin stack shows up as a divergence between two numbers, which is a diagnosis. A board average would simply read slightly high and tell you nothing.

Community builders have already gone further, replacing the air cooler with an all-in-one liquid loop and reporting figures above 11 TH/s. That is enthusiast territory rather than a supported configuration — but it indicates the silicon has headroom the stock cooler does not intend to use. If liquid cooling interests you, our cooling comparison covers the trade-offs honestly, including the ones vendors skip.

How do you set up a Bitaxe Naja Duo?

First power-on takes about five minutes and needs no software beyond a browser.

1. Connect power. XT30 plug into the board, then AC. The board boots and broadcasts its own Wi-Fi hotspot.
2. Join the hotspot from a phone or laptop. A configuration page opens automatically; if it does not, browse to the address printed on the LCD.
3. Enter your home Wi-Fi credentials. The board restarts and joins your network, then displays its new IP address on the LCD.
4. Open that IP in a browser. This is AxeOS, and everything from here happens in the dashboard.
5. Enter your Bitcoin address as the stratum user. Any on-chain address works; a bech32 address starting bc1q is the sensible default.
6. Point it at a pool and save.

That final step is where the interesting decision lives, because this board can do something most miners cannot.

Stratum V2 configuration

The Naja Duo runs esp-miner with AxeOS — the mainline Bitaxe firmware from the bitaxeorg/ESP-Miner repository — and AxeOS gained native Stratum V2 in version 2.14.0, with significant hardening in 2.15.0, released 21 August 2026: larger protocol frames, fractional pool difficulty, and a per-pool authentication requirement. That makes this board a first-class V2 device out of the box.

Check your firmware version in AxeOS before configuring. If you are below 2.14.0, update first, either over the air from the dashboard or over USB-C with the official web flasher.

In Settings → Pool Configurations, AxeOS 2.14+ gives you separate fields rather than a single URL:

Stratum Host          eu-btc.solofury.com
Stratum Port          3333
Stratum Protocol      Stratum V2
User                  bc1qYourAddress.najaduo
Password              x
SV2 Channel Type      Extended
SV2 Authority Pubkey  9cLif4sCxvAz7FBP7GPvYG8Mv586ZhdgNbn3f4PsrM56gboSZEp
Require Auth          enabled  (2.15.0+)
Decode Coinbase Tx    enabled

Swap the host for whichever of the nine regional endpoints is nearest you — the Start wizard measures latency and picks for you, and it also calculates the difficulty a 4.4 TH/s device should ask for, which saves waiting through vardiff settling on its own.

Three of those fields deserve explanation, because they are the difference between “encrypted” and “encrypted and verified.”

The authority public key is how your miner proves the endpoint answering on port 3333 really is the pool it expects, rather than something that intercepted your DNS. It works like an SSH host key. Leave it blank and AxeOS still connects — in trust-on-first-use mode — so the session encrypts, the dashboard looks perfectly normal, and you have quietly skipped the one check that authenticates the other end. Paste the key. On 2.15.0 and later, also switch on Require Auth so the shortcut is unavailable.

Extended channel means the pool sends you the coinbase transaction template rather than just a header. That is what makes the next section possible.

Decode Coinbase Tx turns that template into something readable on the board’s own display.

Set a second pool as fallback while you are in there, pointing at a V1 endpoint. AxeOS handles any pairing of protocols across primary and fallback, keeps probing the primary with a heartbeat while running on the backup, and switches back on its own when the primary answers. Full detail on all of this is in our Stratum V2 solo mining guide.

Reading the coinbase from the board itself

With an extended channel and coinbase decoding active, the Naja Duo displays the outputs of the block it is currently working on:

Block Header
  Height       963788
  Scriptsig    SoloFury
  Value        3.14278556 BTC

Outputs
  bc1q68eg...vyd2ey      0.03142785 BTC     (1%)
  bc1qavkk...n0tlw9      3.11135771 BTC     (99%)

Understand what you are looking at. That is not the pool’s dashboard and not a block explorer — it is the miner on your desk, showing you that if this block is found, 99% of the reward is addressed to your wallet. Before a single hash is spent on it. Every solo pool says the reward goes to your address; with Stratum V2 you no longer have to take anyone’s word for it.

How far can you overclock the Naja Duo?

The tested profile is 600 MHz at 1,120 mV, yielding roughly 8 TH/s at 92 watts. The method matters more than the destination: in Settings → Mining Settings, raise frequency in 25 MHz steps, let each step soak for thirty minutes, and watch the hardware error rate — it should stay under 2%. The moment errors climb past that, step back down: you have found your chip’s ceiling, wherever the vendor’s number sits. When you land on a setting you like, give it a full 24 hours before calling it stable, and keep both chips under 60°C sustained — above 70°C the firmware begins stepping the board down to protect it, and everything you gained goes to thermal management instead of hashes.

Four cautions, in order of how often they cause trouble.

Silicon varies. Two boards off the same reel will not reach the same ceiling. 600 MHz is a tested target, not a promise, and a chip that refuses it is not defective.

Expect the hashrate to drop to zero when you change settings — that is the chip, not a fault. Unlike BM1370 hardware, which switches operating modes on the fly, the BM1373 restarts when its operating point changes. The dashboard reads zero for a moment, then climbs back as the new frequency takes hold. Knowing this in advance saves the standard panic of assuming a save just killed the board.

Size the power supply properly. A 124W unit running a 92W load sits at 74% — comfortable. The rule of thumb worth keeping is not to run a supply above about 80% continuously, because that is where heat, ripple and long-term degradation start compounding. A supply that is fine on a bench for an hour is not automatically fine for a year.

Watch both chip temperatures, not the average. The board reports each separately for a reason. A growing gap between them means a thermal contact problem, and it will always appear before the failure does.

Consider what you are giving up. This is the one people skip. The Naja Duo is remarkable because it is the first Bitaxe under 10 J/TH. Overclocked, it is a 11.5 J/TH board — still respectable, no longer special. If your electricity is expensive, the overclock is a bad trade. If your electricity is cheap and you want maximum lottery tickets per board, it is a good one. There is no universally correct answer, only your kWh rate. Our power measurement article covers why the wall reading and the dashboard reading rarely agree, and which to trust.

What are the odds of finding a block with 4.4 TH/s?

Here is the part most product pages leave out, and the part that determines whether you will still be happy with this purchase in a year.

Bitcoin’s network hashrate is measured in hundreds of exahashes per second. A single exahash is a million terahashes. Your 4.4 TH/s is therefore a very small fraction of a very large number, and the honest description of what you are doing is buying lottery tickets that happen to warm your desk. Expected time to a block at this hashrate is measured in tens of thousands of years.

That figure is real, and it is also less discouraging than it sounds, for a reason worth internalising: hashing is memoryless. Every hash you compute has exactly the same probability as the first one you ever computed. There is no progress bar, no accumulating credit, no “due” for a win. A machine that has run for three years is no closer than one switched on this morning — which means the block could land tonight. Blocks have been found by home miners far smaller than this one. Our odds calculator and the Poisson maths behind mining variance lay out the real distribution without softening it.

So the sensible frame is this. If you buy a Naja Duo expecting income, you have made a mistake that arithmetic will eventually explain to you. If you buy it because you want to participate in Bitcoin’s consensus with your own hardware, hold a real ticket in a real lottery, learn how mining works from the inside, and spend about $3.61 a month doing it — the purchase makes complete sense. Our lottery mining explainer makes that case in full.

One structural note that matters more than most people realise: on a non-custodial solo pool, the reward is paid by the network’s coinbase directly to your address. The pool never holds your funds. Combine that with Stratum V2 coinbase verification and the trust requirement shrinks about as far as it can — you can see where the money goes before you do the work.

Buying one without getting burned

The Naja Duo is open-source hardware, which means anyone may legally build one. That is the design working as intended, and it also means the name on a listing tells you less than you would like.

Three checks, in order of importance:

Verify the firmware first. Before entering pool credentials, reflash with the official open-source web flasher for esp-miner. A failed flash attempt does not damage the hardware, so the test costs nothing — and a board that refuses official firmware has told you exactly what it is.

Check the chip count with arithmetic. A genuine dual-BM1373 board hashes near 4.4 TH/s at stock. One that stalls near 2.2 contains one chip, whatever the silkscreen says. Per-chip output is among the most stable numbers in this industry and it does not negotiate.

Measure at the wall. A twenty-dollar energy plug over 24 hours settles any dispute about the 42-watt figure, and catches a marginal power supply before it degrades into a problem.

One supply-chain detail worth knowing when comparing channels, because it explains this generation’s prices: early BM1373 chips are salvage. They are harvested from Antminer S23 and S23 Hydro hashboards at roughly $60 per chip — a figure both TinyChipHub, which sells BM1373 chips in sealed reels, and the Chinese OEM DigLucky put in the same range — which puts about $120 of raw silicon on a Naja Duo before a single other component — and around $240 on a four-chip board, which is why nothing in this generation is cheap yet. Sealed factory reels exist from exactly one supplier, at a premium; everything else is desoldered silicon of uncertain history, and second-hand chips carry documented risks of degraded hashrate and premature failure. A chip that hashes 10% low forever is a defect no reflash can cure — one more reason a board’s provenance matters as much as its spec sheet, and why a retail Bitmain chip channel, if it ever opens, would reset every price in this category overnight.

Our guide to spotting Bitaxe clones covers the counterfeit question in depth, and AliExpress versus official vendors covers the channel trade-off — the discount is real, and so is everything the discount excuses you from.

Quick fault diagnosis

Four patterns cover most of what goes wrong with boards in this class, and each points somewhere specific:

Anything beyond these, the troubleshooting bible has the full diagnostic tree.

Who should buy the Bitaxe Naja Duo?

The honest summary

The Naja Duo is the first Bitaxe where the efficiency figure is genuinely notable rather than merely acceptable. Single digits at stock is a threshold the category has been approaching for two years, and crossing it in a 42-watt desktop board is a real engineering result — not a marketing rounding.

The overclock will tempt you, and it works, and it undoes the specific thing that makes the board worth buying. That is not a reason to avoid it, only a reason to decide deliberately rather than by reflex.

And whatever you do with the frequency slider, the machine’s actual purpose is unchanged: it computes hashes against a target almost nobody hits, on a network that has never once cared how much you spent. Point it somewhere non-custodial, verify the coinbase yourself, and let the arithmetic run.