⚡ recommended overclocks
midstate is a BLAKE3 sequential-time VDF — pure compute, memory speed is irrelevant. The kernel is power-bound, so the single best thermal lever is a core-clock cap: a small clock cut barely dents hashrate but drops temps a lot. Memory OC does nothing here.
⚠️ For card longevity, tune for ≤ 65 °C. Push fans high — fans are free cooling, zero hashrate cost. There is no forced cutoff; add
--temp-max 69 if you want the miner to auto-stop above 69 °C.| GPU | power limit | core clock cap | typical temp | note |
|---|---|---|---|---|
| RTX 3060 Ti | 100 W | 1350–1400 MHz | 57–63 °C | our rig setting |
| RTX 3060 (8/12 GB) | 120 W | 1500 MHz | ~50 °C | min PL ~123 W on some |
| RTX 3070 / 3070 Ti | 130 W | 1500 MHz | ~55 °C | — |
| RTX 3080 / 3080 Ti | 230 W | 1500 MHz | ~60 °C | — |
| RTX 4070 / 4070 Ti | 150 W | 2400 MHz | ~55 °C | Ada runs cool |
| RTX 4080 / 4090 | 250 W | 2500 MHz | ~60 °C | — |
| RTX 5060 / 5060 Ti | 125–150 W | stock | ~50 °C | Blackwell runs cool |
Apply (Linux, all GPUs):
sudo nvidia-smi -pl <watts> # power limit
sudo nvidia-smi -lgc 0,<clock> # cap core clock — the thermal lever
sudo nvidia-smi -rgc # revert clock cap
Persist across reboot — a oneshot systemd unit (the cap resets otherwise):
# /etc/systemd/system/gpu-oc.service
[Service]
Type=oneshot
RemainAfterExit=yes
ExecStartPre=/bin/sh -c 'until nvidia-smi >/dev/null 2>&1; do sleep 2; done'
ExecStart=/usr/bin/nvidia-smi -lgc 0,<clock>
[Install]
WantedBy=multi-user.target
Then
sudo systemctl enable --now gpu-oc. Lower the cap until your hottest GPU sits ≤ 65 °C. Need help? Ask on Discord.