I'm running the devstral:24b model locally through Ollama, and HWiNFO reports that my Intel i5-13600KF is thermal throttling. The CPU reaches 100°C at around 5,088 MHz, with roughly 78% reported usage, 100% utility, and about 185 W package power. At idle, it sits around 42–46°C.
The CPU is cooled by a Corsair iCUE H150i 360mm AIO. I also replaced the included thermal paste with Noctua NT-H2, so I expected temperatures to be lower. During the workload, the coolant temperature stayed below 40°C, the radiator fans reached about 2,100 RPM, and the pump reached approximately 2,689 RPM. The radiator is mounted at the top of the case, and the pump is detected in the BIOS.
Is reaching 100°C and throttling normal for this kind of AI workload and power draw, or could there be an issue with the pump, mounting pressure, CPU contact, fan curve, or BIOS settings?
4 Answers
Check the BIOS power limits and motherboard enhancement settings. Some boards remove Intel’s stock limits or enable multi-core enhancement, which can let the 13600KF sustain higher voltage and power than expected. Setting Intel-recommended PL1 and PL2 values, or applying a modest undervolt, can reduce temperatures significantly with little performance loss. Also verify that the CPU fan or pump header is configured to run the pump continuously and that the radiator fans ramp up based on CPU temperature.
The coolant temperature staying under 40°C while the CPU reaches 100°C suggests the radiator is removing heat reasonably well. That makes a completely failed pump or radiator less likely, though poor cold-plate contact, uneven mounting pressure, an overly thick or poorly spread paste layer, or a protective film left on the block could still cause high core temperatures. It would be worth removing and remounting the block, checking the paste imprint, and confirming that the pump is connected to a constant full-speed header rather than a temperature-controlled fan profile.
The reported 100°C is not automatically proof that the cooler is defective. It means the CPU reached its thermal limit and reduced boosting to protect itself. If the clock speed remains stable and performance is acceptable, occasional throttling during a long all-core workload may simply be the processor managing its power. If it immediately jumps to 100°C, clocks drop heavily, or temperatures are much higher than expected compared with similar systems, then remounting the cooler and checking BIOS power settings should be the next steps.
The missing CPU model matters, but with an i5-13600KF, this behavior can be normal under a sustained 185 W workload. Intel CPUs are designed to boost aggressively until they reach their temperature or power limits, and a local AI model can keep the CPU heavily loaded even when the overall usage percentage does not look like 100%. A 360mm AIO is appropriate, but it cannot prevent the processor from reaching its configured thermal limit if the motherboard allows it to use that much power.
The radiator is mounted at the top of the case, and the pump is detected in the BIOS. I’ll still check the block mounting and cable connection to make sure everything is making proper contact.