Hardware monitoring software polls devices at a low priority so that your performance stays good on the tasks you need the CPU / GPU to work on, instead of pausing to check temperature. As a result, it may be pushed back or time out. When this happens, you get an anomaly like this. The bogus values can vary, sometimes bouncing back a maximum depending on the bit width of the sensor data path. For example, my SSDs with no temp sensor will always erroneously read 128c or 255c, and some fan RPM sensors will be 32k or 65k.