GPU Health Summary Under Power Cap | Generated by AI
GPU is healthy. Quick read:
RTX 4070 (12 GB)
─────────────────────────────
Temp: 66 °C ✓ (throttle starts ~83 °C → 17 °C headroom)
Power: 208 / 215 W → at the power cap, normal for sustained training
Fan: 56 % ✓ moderate, room to spin up if temp rises
GPU-Util: 100 % ✓ pegged
Memory: 7.5 / 12.0 GiB (61 %) ✓ comfortable, no OOM risk
Perf state: P2 ✓ max performance
ECC errors: none reported
Training process (PID 1200744, python3.11) is the only meaningful workload — 7.36 GiB. Xorg + gnome-shell take <120 MiB combined, no issue.
Why MFU is “only” 12.86%: the card is hitting its 215W power cap, not thermal limits. That’s the bottleneck — there’s no fix without raising the cap (which on a 4070 isn’t meaningful) or undervolting. Step time of 11.11 s/iter is what this card can do at this power budget, and it’s been rock-stable for ~5 hours.
Bottom line: nothing concerning. Temps are good, power is steady, memory has headroom, no errors. Run can finish.