// blog · June 16, 2026 · by Nathan Baldwin

Pushing a Bitaxe Past 700 MHz: The Silicon Lottery and How to Chase 900

// Bitaxe Baller's manual-tune cap is now 900 MHz. Here's what a good BM1370 chip can actually hit, why voltage — not frequency — is the thing that kills boards, and how to climb safely.

We just raised the manual-tuning ceiling in Bitaxe Baller from 700 MHz to 900 MHz. The old cap was conservative to a fault: BM1370 silicon-lottery winners with real cooling were hitting the wall at 700 and leaving stable hashrate on the table. If you’ve got a good chip and a heatsink that can keep up, 800–900 MHz is now yours to chase. This post is the honest version of how to do that — including the part where most chips won’t get there, and the part where the number that actually destroys boards isn’t frequency at all.

What “stock” and “tuned” really mean

A Gamma ships around 525–575 MHz out of the box. That’s the safe-for-everyone default, sized so the worst chip in the bin runs cool and stable forever. It is not where a good chip wants to live.

Past that, you’re in silicon-lottery territory. Every BM1370 is slightly different — same part number, different tolerance for clock and voltage. A mediocre chip taps out in the 650–700 range. A genuinely good one, kept cool, will hold 800–900 MHz. There is no way to know which one you have except to climb and watch.

Frequency scales hashrate roughly linearly, so the jump from 575 to 850 MHz is real money in share terms — on the order of a 45–50% bump if the chip holds and the errors stay down. The “if” is doing a lot of work in that sentence.

The thing that actually kills boards: voltage

Here’s the part people get backwards. Raising frequency makes a chip run hotter and, past its limit, makes it spit hardware errors. Both are visible, both are recoverable — drop the clock and the chip is fine. Frequency is a knob you can turn both ways.

Voltage is different. The BM1370’s ceiling is 1300 mV — that’s Bitmain’s hard spec, and Bitaxe Baller will not let you exceed it. But the real-world caution line is lower: sustained operation above roughly 1225 mV accelerates electromigration and wears the silicon out faster, regardless of how good your cooling is. Cold doesn’t save you here. Heat damage and voltage damage are two different failure modes, and a frosty heatsink only fixes the first one.

So the discipline is: climb on frequency, spend voltage reluctantly. Every extra millivolt above ~1200 is buying you a little stability now in exchange for lifespan later. Sometimes that trade is worth it. It should always be a conscious choice, not an accident.

How to actually climb

Do this with cooling already sorted. If your VR or ASIC temps are marginal at stock, fix airflow first — overclocking a thermally-limited board just means you abort earlier and learn nothing.

1. Start from a known-good baseline. Note your stock frequency, voltage, hashrate, and HW error rate. You need something to fall back to.
2. Step frequency up in small increments — 25 MHz at a time — and hold each step for a couple of minutes. Watch three numbers: hashrate (should rise), HW error rate (should stay near zero), and temps (ASIC and VR).
3. Only raise voltage when frequency stalls. If a frequency step makes errors climb or hashrate stops scaling, the chip wants a little more voltage to hold that clock. Give it the minimum that stabilizes — and remember the 1225 mV caution line.
4. Back off at the first sign of instability. A rising HW error rate is the chip telling you it’s past its comfortable limit. Errored shares are wasted work; a slightly lower stable clock out-earns a higher unstable one every time.
5. Find the highest stable point, then sit there. The goal isn’t the highest number you can flash for thirty seconds. It’s the highest clock that runs clean for days.

If you’d rather not babysit a spreadsheet, Bitaxe Baller’s auto-tune sweep (Pro) does exactly this climb for you — frequency-only, +25 MHz per observation window, with a hard abort and baseline restore if temps or error rate cross the guardrails. It records the highest stable frequency it found and applies it. Voltage is never touched during a sweep, by design — see the section above for why.

The reality check

Most chips are not 900 MHz chips. If yours settles in happily at 700 and gets cranky above it, that’s a perfectly normal result and not a defect — it’s just where that particular piece of silicon lives. Chasing the last 50 MHz with more and more voltage is usually a bad trade: you’re spending lifespan to win a few percent of hashrate that variance will swamp anyway.

The raised cap is there for the chips that have earned it. Climb carefully, watch your errors, keep voltage on a short leash, and let the good ones show you what they’ve got.

Try it yourself: Bitaxe Baller is a free desktop app for Mac and Windows — live monitoring, one-click tuning recommendations, and manual control up to 900 MHz with hardware-safe bounds. Pro adds automated tuning sweeps with abort guardrails. Open source on GitHub.

← back to blog