// blog · · by Nathan Baldwin
// Bitaxe Gamma overclocking with real measurements — frequency, voltage, and why VR temperature kills your board before the ASIC ever gets hot.
A stock Bitaxe Gamma hashes around 1.2 TH/s at 14W. Push it to 575 MHz and 1200 mV and you’ll see 1.35 TH/s at 17W — same efficiency, more raw output. Push it to 625 MHz and 1250 mV and you’ll see 1.5 TH/s at 20W, with the VR temperature climbing past 70°C. That’s the line where the regulator, not the chip, becomes the part that fails first.
If you only watch ASIC temp while you tune, you’ll cook a VR before the ASIC ever throttles.
The BM1370 has two knobs that matter: core frequency and core voltage. Frequency is set in MHz, voltage in millivolts. Both are exposed in AxeOS’s web UI and in the Bitaxe API.
Higher frequency means more hash attempts per second per chip — linear scaling, up to the point the chip can’t complete an attempt before the next clock edge. That point is voltage-dependent: more voltage stabilizes higher frequency.
Higher voltage means each transistor switch costs more energy. The penalty isn’t linear — it goes roughly with voltage squared. Doubling the voltage doesn’t double power draw, it quadruples it (for the chip itself; the rest of the board is constant overhead).
So the math says: push frequency as high as the silicon will hold, push voltage only as high as you need to hold that frequency. That math has been right since the first ASIC, and it’s still right on a Gamma.
These numbers are from my three Gammas, all on a Bitcoin Merch v2 board, running 24/7 in a 22°C room with the stock fan profile. Your silicon will vary by ±25 MHz on the high end.
Tier 1 — Stock (525 MHz / 1166 mV) - Hashrate: ~1.20 TH/s - Wall power: ~14W - Efficiency: ~11.7 J/TH - ASIC temp: ~55°C - VR temp: ~52°C - HW error rate: <0.1%
The factory defaults. Reliable forever, low noise, low heat. If you’re running a Gamma in a bedroom or near a desk, stay here.
Tier 2 — Light OC (550 MHz / 1180 mV) - Hashrate: ~1.27 TH/s - Wall power: ~15W - Efficiency: ~11.8 J/TH - ASIC temp: ~58°C - VR temp: ~55°C - HW error rate: <0.2%
A free 6% bump with no thermal cost. This is the setting most fleet owners settle on for daily-driver miners.
Tier 3 — Sweet spot (575 MHz / 1200 mV) - Hashrate: ~1.35 TH/s - Wall power: ~17W - Efficiency: ~12.6 J/TH - ASIC temp: ~62°C - VR temp: ~60°C - HW error rate: ~0.3%
The efficiency curve starts bending here. You’re paying ~8% more power for ~13% more hashrate, but HW errors creep up. Good silicon does this cleanly; marginal silicon starts dropping shares.
Tier 4 — Aggressive (625 MHz / 1250 mV) - Hashrate: ~1.50 TH/s - Wall power: ~20W - Efficiency: ~13.3 J/TH - ASIC temp: ~68°C - VR temp: ~70°C - HW error rate: ~0.5–1.0%
The chip is happy. The VR is not. This is the tier where the bottleneck shifts from “is the silicon stable?” to “can the regulator dump heat fast enough?”
Push past 1.50 TH/s and you’re typically increasing voltage faster than frequency to keep stability, which means the efficiency curve gets ugly fast.
The voltage regulator on a Bitaxe board is a small MPS step-down converter rated for around 30W of throughput. At stock voltage (1166 mV) and ~12A of core current, it’s converting 5V → 1.17V with maybe 88-92% efficiency. The other 8-12% becomes heat, dissipated through a small thermal pad to the PCB ground plane.
At Tier 4, you’re asking it to push 17A at 1.25V. The math: 17A × 0.13V drop across internal resistance = ~2.2W of heat dissipated in a component the size of a fingernail. The pad and ground plane can sink that for a while, but the steady-state temperature climbs into the 70°C range — and the IC itself is rated for 125°C junction.
So you don’t fail at 75°C VR. You just start aging the regulator faster. MOSFETs are a thermal-cycling part: every degree above 65°C cuts expected life roughly in half over years of continuous operation. The chip will outlive the VR.
The OLED only shows ASIC temp by default. The VR temp comes from a separate sensor (NTC thermistor near the regulator) and is exposed in the /api/system/info JSON under vrTemp. AxeOS shows it in the dashboard but doesn’t alarm on it. Anything over 70°C sustained is the warning zone.
Walk up frequency 25 MHz at a time. After each step:
Stop when one of the four trips. The chip silicon lottery is real: I’ve seen Gammas hit 625 MHz at 1220 mV (great silicon) and others that need 1240 mV to hold 575 MHz (mediocre silicon). The reading on the dashboard is the truth, not the spec sheet.
For voltage steps, 10 mV is fine-grained enough. 20 mV is faster. The penalty for being 10 mV over what you need is small; the penalty for being 10 mV under is share drops.
Sustained VR temps above 80°C will pop the regulator. You’ll see the device reboot itself when the VR cuts out at thermal shutdown, then come back, then cut out again. By that point the IC has likely lost some headroom and you’re a few months from a board replacement.
Sustained ASIC temps above 80°C trigger the chip’s internal throttle — you’ll see frequency drop on the dashboard while voltage stays put. Not destructive, but you’re paying full power for half the output.
Dust in the heatsink fins is the silent killer. A 20°C delta from clean-to-dusty is normal after six months. Compressed air every quarter is enough.
Pull up your Gamma’s dashboard right now and write down the four numbers at stock: hashrate, ASIC temp, VR temp, HW error rate. Then bump to Tier 2 (550 MHz / 1180 mV) and let it run for 24 hours. Write the same four numbers down again.
If Tier 2 looks clean, repeat for Tier 3. Do not jump from stock to Tier 4 — you’ll miss the inflection point where your specific board starts losing efficiency. Walk the curve.
The goal isn’t the highest TH/s. The goal is the best J/TH at the lowest VR temp you can hold for years. That number is unique to your board.
Try it yourself: Bitaxe Baller is a free Mac app that surfaces these recommendations automatically across your fleet — live monitoring, tuning suggestions, pool config, all in a native window. Open source on GitHub.
