The heat production is not just the results of high voltages, the frequency also plays a role there...
if they throttle Vcore without throttling frequency, we would get bluescreens. I think throttling does what it can while still continuing to operate.
Expect heat - watts of power - to go up proportional to frequency when voltage is constant. Power goes up proportional to voltage squared when frequency is constant.
The CPUs need for voltage at higher frequency is not linear, but a linear relationship sets a lower bound on the total power variation -- frequency cubed as we try to push higher overclocks by increasing volts.
Various benchmarks at the same total power, speed, voltage, etc. produce different temperatures depending on how the power is spread among the CPU cores. Realbench is a stringent test and does a very good job of spreading the power evenly and holding the power constant over the multi-threaded tests. OCCT distributes power evenly among the cores. Those two stress/benchmark programs show lower temperature than Aida64 at the same speed, voltage and using comparable stress test -- non AVX CPU and cache stress. Aida64 rotates the stress among the cores one after the other so that the cores being stressed are hotter than the others and monitored temperature appears higher. Single threaded benchmarks are the worst at uneven heat.
That's one of the 'features' we live with when using multi-core CPUs -- hot spots. We need to pay attention to the hottest core because that's what will cause throttling and threaten the health of the CPU.
Jeff
Expect heat - watts of power - to go up proportional to frequency when voltage is constant. Power goes up proportional to voltage squared when frequency is constant.
The CPUs need for voltage at higher frequency is not linear, but a linear relationship sets a lower bound on the total power variation -- frequency cubed as we try to push higher overclocks by increasing volts.
Various benchmarks at the same total power, speed, voltage, etc. produce different temperatures depending on how the power is spread among the CPU cores. Realbench is a stringent test and does a very good job of spreading the power evenly and holding the power constant over the multi-threaded tests. OCCT distributes power evenly among the cores. Those two stress/benchmark programs show lower temperature than Aida64 at the same speed, voltage and using comparable stress test -- non AVX CPU and cache stress. Aida64 rotates the stress among the cores one after the other so that the cores being stressed are hotter than the others and monitored temperature appears higher. Single threaded benchmarks are the worst at uneven heat.
That's one of the 'features' we live with when using multi-core CPUs -- hot spots. We need to pay attention to the hottest core because that's what will cause throttling and threaten the health of the CPU.
Jeff
