To counter the potentially erratic behavior coming from these high thermal ramp rates, we need either a cooling solution that will keep the CPU-die at a safe temperature at all times, no matter the heat load. Or an accurate means of detecting them on the CPU-die and taking precautions before anything happens.
The thermal management feature on the Pentium 4 therefore includes a feature called ‘Thermal Monitor’ which in essence is a highly accurate on-die temperature sensing circuit that has the ability to act quickly upon any thermal issues (~50ns). This circuitry keeps an eye on the most taxed areas of the CPU-die at all times and will quickly act upon temperatures going over the safety limits. Each Pentium 4 CPU includes this thermal management feature and is individually calibrated during manufacturing to eliminate any manufacturing variations.
Fig 6. A good example of a suitable Pentium 4 heatsink, featuring a copper base, folded fins and a powerful 60mm fan.
The thermal monitor’s control circuit, when active, lowers the CPU temperature by throttling the internal CPU clockspeed. This is done with a 50% duty-cycle which means that a 1.5GHz CPU will then effectively run at a 750MHz clockspeed. Due to the fast response time of the thermal monitor circuit (~50ns) the CPU will only be ‘throttled’ for a very brief period, once the CPU-die temperature is within safe operating limits again it’ll set back to the 1.5GHz clockspeed it originally operated at.
Although the thermal monitor feature is always enabled, the clock throttling portion is actually disabled by default. To maintain compatibility with previous generations of CPUs having no thermal management, the motherboard’s BIOS must support and enable clock throttling or a driver could initialize it after the system has booted the operating system. This means that the clock throttling part of thermal monitor is not enabled by default and can be dis/en-abled by either the BIOS or a driver.
Additionally the Pentium 4 includes a second on-die diode to monitor CPU-die temperature, this circuitry actually works independently of the thermal monitor. Its purpose is to be able to monitor the average CPU-die temperature and in the event of a cooling failure it will shut the processor down when the CPU-die temperature has reached approximately 135C. So at all times there’s two on-die diodes and their control circuitry safeguarding the CPU, either through overheating and erratic behavior due to high thermal ramp rates, or due to a cooling failure and an overall CPU-die temperature in excess of 135C.