A mining rig in the heat: how to survive summer without losses
A rig that ran like clockwork all winter starts acting up in summer: fans scream, temperatures climb, a stable overclock suddenly is not stable. The cards did not go bad — the room got ten degrees hotter. Summer is survived with three things: exhaust, fewer watts, and a fall inspection.
The short version (for the no-time-to-read)
- A card lives in the room's air. Every +10 °C indoors is roughly +10 °C on the cards: fans will cover part of it, not all.
- Exhaust beats air conditioning. The job is to push hot air out of the room, not to "freshen" it inside.
- Summer mode means fewer watts. An overclock stable at 68 °F is often unstable at 95 °F; cutting the power limit by 10–20% brings stability back at almost no revenue cost.
- Memory suffers first. GDDR6X runs close to its limit even in winter — in summer, watch it before anything else.
- Fall is inspection month. After summer: blow-out, fan check, and thermal pad replacement if needed.
The physics of heat: why a "summer" overclock is different
A card's temperature is the room temperature plus the cooler's "delta". A cooler does not create cold — it moves heat into the air, and the hotter the air, the worse the transfer. So an overclock that lived happily at 68 °F gets +27 °F on every sensor at 95 °F room temperature — and slides past the edge of stability.
GDDR6X memory takes the hit first: its headroom is small even in winter, and in summer it reaches the limit before anything else (details in the hot memory article). The second candidate is thermal throttling: the card drops clocks on its own to survive, and hashrate falls further than it would from a voluntary watt cut.
The counterintuitive conclusion: in summer, "trimming watts" is not giving up revenue — it is how you avoid losing it to throttling and crashes. The physics of temperature and GPU lifespan is covered in detail here.
Ventilation: exhaust, not cooling
A 1.5 kW rig emits exactly what a 1.5 kW heater does — literally all of its consumption becomes heat. An air conditioner grinds that heat down for extra kilowatts; an exhaust simply throws it outside. The order of priorities:
- a duct fan to the outside (through a window, into a vent) — the main summer investment. Hot air pools at the top, so the exhaust goes high;
- intake down low. An exhaust without intake works at half strength: give the air a path, "in at the bottom — out at the top";
- do not line rigs up like a train: the second rig sitting in the first one's hot exhaust lives in a sauna;
- air conditioning is the last resort. It works, but it turns heat into a double power bill. Exhaust and watts first.
The rig's summer mode
- Cut the power limit by 10–20%. The hashrate loss is usually half the wattage loss — and temperatures drop noticeably. Same logic as efficiency mode (the winter article covers the economics of it).
- Watch memory, not just the core. The core is still fine when the memory is already at the edge.
- Nights are cooler. If the rig lives at the limit, it is sometimes easier to sit out the hottest hours and run at night. The pros and cons of stopping — in the night article.
- Do not panic over "high" fan speeds. Fans at 80–90% in summer are normal; the worry sign is rising temperatures at already-maxed fans.
And the fall inspection: summer is a hard season for consumables. A blow-out, a fan check by RPM trend and a look at the thermal pads — on the regular maintenance schedule, just done right after the season.
Where automation fits
The summer switch is not "retune everything from scratch" — it is changing the mode. OCBot runs each card in a chosen style — compromise, efficiency or hashrate — and keeps stable presets per card. Moving to efficiency for the summer and back to winter presets in the fall is a couple of Telegram commands, not a new season of hand-tuning.
This article is general information summarizing public practice; actual temperatures depend on your hardware and room. OCBot is an independent project: not affiliated with NVIDIA, HiveOS or Hiveon. Use at your own risk — see the disclaimer.