On this page6 sections
A miner arrives tuned to the manufacturer's default, which is a compromise chosen for a generic customer. Your site is not generic. If your power is cheap and abundant you want output. If it is expensive or capped you want efficiency. Tuning is how you move the machine from one to the other, and on a fleet of any size it is worth real money.
What tuning actually changes
Every ASIC chip runs at a frequency, and holding a frequency requires a voltage. Push the frequency up and you need more voltage, which costs power out of proportion to the extra hashrate you gain, because power scales roughly with the square of voltage. Pull the frequency down and the voltage can drop with it, and efficiency improves for the same reason in reverse.
So the machine has a curve, not a setting. Along that curve you trade terahash against joules per terahash. Tuning is choosing your point on it.
Underclocking, and why most operators should
Dropping frequency and voltage below default typically buys a meaningful improvement in joules per terahash for a smaller percentage loss in hashrate. On an expensive power contract that trade is straightforwardly profitable: you give up some revenue and give up more cost. It also lowers chip temperature, which extends hardware life and reduces failure rates, and it reduces heat and noise, which can matter as much as the arithmetic.
The test is simple. Work out your breakeven power price at default and at the underclocked profile. If underclocking moves your breakeven further above your actual power price, you have bought yourself margin and headroom for the next bad month.
Overclocking, and when it pays
Pushing above default earns more terahash at worse efficiency. That is a good trade in exactly one situation: your power is cheap enough, or fixed enough, that the extra revenue outruns the extra cost, and you have cooling headroom to absorb the heat. Operators with stranded gas, curtailment credits or a flat rate contract are the natural case.
The costs are not only electrical. Higher voltage and temperature shorten chip and PSU life, warranty cover generally does not survive it, and the extra heat has to go somewhere your cooling was not sized for. Overclocking a fleet into a cooling system already at its limit is a common route to a summer of failures.
Custom firmware: what you gain and what you give up
Third party firmware gives finer control of the frequency and voltage curve, tuning per chip that can lift efficiency beyond what stock allows, better telemetry and features like automatic tuning to a power target. On a large fleet, a few percent of efficiency is a serious number.
What you give up is stated plainly in every hardware agreement in this trade, ours included: installing third party or modified firmware generally voids the manufacturer warranty, and where it causes a fault it removes your cover. Most custom firmware also carries a developer fee taken as a share of hashrate, which belongs in your calculation. And it is another software dependency on a site that already has several.
The sensible pattern is to run stock through the warranty period on new machines and consider custom firmware afterwards, or to run it from day one on used machines where the warranty question is already settled, as it is on anything sold with a short dead on arrival window rather than a manufacturer warranty.
Tuning to your contract, not to a leaderboard
- Fixed cheap power with cooling headroom: tune towards output.
- Expensive or metered power: tune towards efficiency and take the hashrate loss.
- Demand response or curtailment obligations: run a profile you can drop quickly and predictably, and test the drop before you need it.
- Hot months, hot site: underclock seasonally rather than fighting the ambient temperature with fans that are already at maximum.
- Ageing fleet approaching its breakeven: underclocking often adds quarters of useful life to machines that would otherwise be switched off.
Doing it without breaking things
Tune on a sample, never on the fleet. Take ten machines, apply the profile, and leave them for a week watching hashrate, rejected shares, chip temperatures and hardware error rate. A profile that looks excellent for an hour and produces errors on day four is common. Roll out in stages, keep the stock profile documented so you can return to it, and record which machines run which profile, because in six months nobody will remember.
Run this against your own numbers
Tell us your power situation, your site and the scale you are aiming at. We come back with a hardware shortlist and a firm quote.