Updating Miners and Drivers: When to Do It and When to Leave It Alone
Every changelog promises "+5% and stability", every forum is full of "updated — everything broke". The truth sits in between: some updates are necessary, others are harmful, and you can tell them apart in advance. Here is an update discipline for a working farm.
TL;DR (for those skipping the read)
- Update for a reason, not on a schedule: a new algorithm, a fix for your specific problem, security. "A new version is out" is not a reason.
- A stable rig on an old version earns more than a rig on a new version that spends a day being reconfigured.
- The one-rig rule: update one machine first, watch it for a couple of days — then the rest.
- Before updating — a config backup and a known rollback path. Without that, an update is an irreversible experiment.
- Download miners only from official sources. A "forum build" is the most popular way to lose a farm.
When an update actually makes money
Three valid reasons. First — a new algorithm or coin: a miner that does not support your algorithm earns exactly zero. Second — a fix for your specific problem: the changelog explicitly mentions your GPU model or your symptom (drop-offs, memory errors, pool incompatibility). Third — security: closed vulnerabilities in software that faces the internet around the clock.
A special case is the developer-fee change: a new version with a lower fee is direct money; an "improved" version whose higher fee eats the entire hashrate gain is not. Count what reaches the wallet, as in the fees article.
Drivers follow the same logic but more conservatively: a driver touches all hardware at once. The reason to update a driver is a requirement of the new miner version or a bug fixed for your GPU series — not "a new game driver dropped".
When to leave it alone
The farm has been stable for weeks, the algorithm has not changed, the new changelog is about other people's cards and other people's systems — do not update. Every update is risk and time: reboots, re-tuning, stability checks. The cost of downtime easily eats the mythical "+3%" from a changelog: one unstable day on an eight-GPU rig costs more than a month of a three-percent gain.
It is especially dangerous to update "while at it" before a trip or on a Friday night: if something breaks, you will be fixing it remotely and in a hurry. Updates are scheduled for when you are nearby and unhurried.
How to update safely: the checklist
- Read the changelog in full — look for yours: your cards, your algorithm, your symptoms. Nothing of yours — postpone.
- Save the current configuration and software versions, so a rollback takes minutes instead of reconstruction from memory.
- Update one rig and watch it for at least a couple of days: hashrate, errors, stability. All clean — roll out to the rest.
- Re-verify the overclock after updating: a new miner version may load the card differently, and the old preset may end up past the edge (see the stability article).
- Download only from official sources and verify checksums where published. A miner with rights over the whole farm is a perfect theft vector.
Where automation fits
After a miner update the overclock usually needs re-checking — and that is the most tedious part. OCBot re-tunes the working point for the new version and immediately shows if hashrate or stability sagged compared to "before". And when something goes wrong at night, you learn it from a notification, not from an empty chart in the morning.
This article is general information summarizing common practice; actual version behavior depends on your software and hardware. OCBot is an independent project, not affiliated with NVIDIA or HiveOS/Hiveon. Use at your own risk: disclaimer.