From 0.5 to 2.5 TH/s: The Complete BM1366→BM1373 Chip Lineage for Home Bitcoin Miners

Every home Bitcoin miner you've seen in the last three years — the Bitaxe, the Nerdaxe, and everything Ix Tech builds — exists because of one pipeline: Bitmain designs an industrial ASIC chip, deploys it in a fleet-scale miner, and eventually that silicon trickles down to the open-source community.

The chip is everything. It determines your hashrate, your power draw, your thermal challenge, and ultimately whether home mining is viable or just a novelty.

I have all four of the chips that define this era sitting on my bench right now — the BM1366, BM1368, BM1370, and the new BM1373CC. This isn't a spec-sheet comparison. It's a hands-on lineage breakdown from someone who has built products around most of these.

Here's how we got from 500 GH/s to 2.5 TH/s per chip — and why the BM1373 changes the math for pleb miners in 2026.

CHIP COMPARISON TABLE

Why ASIC Chip Generations Matter for Home Miners

Every ASIC chip Bitmain ships is designed for one customer: a large-scale industrial operator buying miners by the container. The Antminer S19, S21, S23 — those machines exist to fill warehouse-scale farms running on subsidized power. Home miners were never the target market.

That changed when open-source hardware projects figured out how to harvest the ASIC chips from those same industrial machines and redesign the board from scratch. Instead of buying a 200-pound miner and paying $0.10/kWh to run it, a home miner can build a device the size of a paperback book, plug it into a standard outlet, and point it at a solo mining pool. The chip does the same SHA-256 work — it just does it at a scale that fits on a desk.

This is the open-source pipeline: Bitmain ships a chip generation, industrial miners buy the machines, the open-source community reverse-engineers the chip, and new designs follow. Ix Tech sits at the end of that pipeline — we design boards, source chips, and ship hardware that's assembled in the USA.

When you're evaluating a chip for a home miner build, two numbers matter: per-chip hashrate (how much work one chip does) and efficiency (joules per terahash, J/TH — lower is better). Everything else is noise. A chip that does 2.5 TH/s at 10 J/TH is categorically better than one doing 0.5 TH/s at 24 J/TH. The difference between those two numbers is four chip generations — and the gap between them is what this post is about.

BM1366 — Where Open-Source Mining Started

5nm | ~0.5 TH/s per chip | ~24 J/TH | Source: Antminer S19 XP / S19k Pro

The BM1366 is where the current wave of open-source Bitcoin mining hardware became possible. At 5nm, it was Bitmain's most advanced chip at the time — and it was capable enough that small, single-chip and multi-chip designs could produce a meaningful hashrate without drawing industrial power levels.

On the bench, the BM1366 is easy to identify: it's the largest die of the four generations we're covering here. Measured against calipers, it's visibly bigger than its successors — a reminder that process improvements aren't just about performance, they're about fitting more transistors into less physical space.

The BM1366's 24 J/TH efficiency isn't impressive by 2026 standards, but it was the floor that made everything else possible. Without a chip the community could reverse-engineer, clone PCB designs around, and run reliably on consumer power infrastructure, none of the subsequent generations would have open-source hardware to land in. 

BM1368 — Refinement, Not Revolution

5nm refined | ~0.65–0.75 TH/s per chip | ~17 J/TH | Source: Antminer S21

The BM1368 didn't change process nodes — it's still 5nm. What Bitmain changed was the architecture inside: hash core layout, power delivery, and die utilization. The result is a 30–50% per-chip hashrate improvement and a meaningful efficiency gain at the chip level.

Here's the counterintuitive part that trips people up: if you build a system equivalent to 1 TH/s using BM1368 chips, it draws slightly more power than a 1 TH/s BM1366 system. That seems backwards given the 17 J/TH vs 24 J/TH specs. The reason is that you need fewer BM1368 chips to reach the same hashrate — so your total chip count drops, but each chip running draws more than a BM1366 at comparable load. The efficiency gain is real at the chip level; it just doesn't automatically translate to a system-level power reduction when you're comparing equivalent total hashrate.

Where the BM1368 actually shines is in density. You get more hash per board, which means smaller PCBs, simpler thermal management, and lower BOM cost per terahash. That's what made it a natural fit for a generation of Ix Tech designs.

The BM1368 is still a solid chip for home mining builds. If you have BM1368 hardware, it's not obsolete — it's just not the most efficient option available today.

BM1370 — The Node Jump Changes Everything

3nm | ~1.2 TH/s per chip | ~15 J/TH | Source: Antminer S21 Pro

The jump from BM1366/BM1368 (5nm) to BM1370 (3nm) is the most significant process node transition in open-source mining history so far. Everything changed: the transistor density, the core voltage, the thermal profile, and the performance ceiling.

At stock clocks, the BM1370 delivers approximately 1.2 TH/s per chip — more than double the BM1368 and nearly 2.5x the BM1366. Efficiency lands around 15 J/TH at stock. But the real story is what happens when you push it.

The BM1370 has substantial overclocking headroom. With adequate voltage and proper cooling, it's capable of exceeding 1.8 TH/s per chip — we have been able to achieve higher with the NerdQX that we designed for example. The PLL-based frequency control that all these chips use means hashrate is a tunable parameter, not a fixed spec. You write a higher frequency to the PLL register; the chip runs faster; you need more voltage and more thermal dissipation to keep it stable.

This is where the El Mirage copper-plated heatsink matters. Stock thermal solutions max out at stock clocks. If you're pushing a BM1370 past 1.5 TH/s, you need a heatsink designed for it — something with the surface area and thermal interface to pull heat off a 3nm die running at elevated voltage. El Mirage was built for this.

The Bitaxe Gamma is the reference design for BM1370-based open-source miners. It's the benchmark — the product the community validated the chip on, and the foundation for Ix Tech's Gamma builds.

BM1373 — The Chip That Resets Expectations

3nm refined | ~2.4–2.5 TH/s per chip | ~10 J/TH | Source: Antminer S23 Hyd

The BM1373 is the same process node as the BM1370 — 3nm — but it's not an incremental improvement. It's an architectural revision of what 3nm can do when Bitmain has had time to mature the design. The numbers tell the story: roughly 2.5 TH/s per chip and 10 J/TH. That's approximately double the hashrate of the BM1370 and a 33% efficiency improvement.

The physical chip we have on the bench is designated BM1373CC, lot P2L925CC09 — Q3 2025 production. It measures approximately 13mm × 13mm against calipers. The die is dense. The pad count visible on the BGA package is notably higher than the BM1370, which is part of why you cannot simply drop a BM1373 onto a BM1370 board and expect it to work.

Pinout incompatibility is absolute. Different ball count, different pitch, different pad layout. Different power delivery requirements — the BM1373 runs at a lower core voltage than the BM1370, which is a primary driver of its efficiency gain alongside the hash core architecture changes. Different register map. The BM1373 requires a complete PCB redesign. There is no adapter, no rework, no shortcut. If someone tells you otherwise, they're wrong.

Here is a 4-chip BM1373 prototype built by the creator of the NerdQ* line of home miner, Pmaxsd. As you can see, this is whole new board design. 

One of the most amazing things about this new chip is its overall performance that we've talked about previously. In these images below, you can actually see that there are massive gains in both hashrate and power efficiencies, and a lot of head room to overclock.  

Here are the 4 chips in that prototype being pushed to the extreme. These numbers are insane!

Supply chain reality: BM1373 chips today come from harvested Antminer S23 Hyd boards. The S23 Hydro is Bitmain's liquid-cooled flagship — not a machine that was being run by typical small operators. Harvested board quality varies, and the yield from a given lot of boards isn't 100%. This is normal and expected. It's why sourcing matters, and it's why Ix Tech is doing the qualification work before products ship. The air-cooled versions of the S23 have been constantly delayed, and as of the writing of this blog, they are expected to start shipping in July 2026. 

The BM1373 is the basis for the next line of Ix Tech product's currently in development. Follow for updates. What we can say now: if the BM1370 made home mining practical, the BM1373 makes it compelling.

Conclusion

From 0.5 TH/s at 24 J/TH to 2.5 TH/s at 10 J/TH — four chip generations, each one expanding what's possible for a miner building outside a data center. The BM1366 opened the door. The BM1368 refined it. The BM1370 kicked it off its hinges. The BM1373 builds a new door entirely.

Special thanks to Pmaxsd for providing pictures of his prototype.