Raspberry Pi
Darren Yates explains why your CPU gets hot, and the thermals at play you may not be aware of.
How CPU coolers really work
CPU temps: how hot is too hot?
Ever since physicists William Shockley, John Bardeen and Walter Brattain invented the transistor in 1947, we’ve been making transistors bigger and smaller, faster and more efficient. Today, their form may be different but transistors still amplify the music in your phone, display the action of your movies and the games on your consoles. But no matter the application, the one thing every transistor does is dissipate excess power as heat and nowhere is this more crucial than in your PC. Pushed hard in gaming, encoding 8K video or mining cryptocurrency, the average PC CPU can dissipate enough power to drive six 15-watt LED light globes, so getting rid of that heat quickly enough to keep the billions of transistors inside your CPU happy and healthy is a science in itself.
Clock rate x voltage = heat
There are many factors affecting CPU temperature – from the size and number of transistors in the CPU, right down to the electrical characteristics of the transistors themselves. But in practice, the two key factors are the voltage the CPU operates at and the rate at which the transistors switch on and off – what we know as the ‘clock speed’ or ‘CPU frequency’.
An important power management technique developed early on and still used today is ‘dynamic voltage and frequency scaling’ or DVFS. Electrical power consumption in this context is a function of voltage, current and the CPU frequency. By throttling back the CPU frequency and operating voltage on lighter workloads, CPU makers can reduce the instantaneous power dissipation. However, to achieve maximum performance, the CPU needs this voltage and clock frequency at their peaks – and there’s where the heat is generated. If this heat isn’t mitigated, the transistors will overheat and the CPU will fail.
How hot is too hot?
Which brings us to the money question – when it comes to CPU temperature, how hot is too hot? It’s a common-enough question, but one that’s difficult to answer. That’s because like most things, it’s not all black-and-white.
All modern CPUs have two key temperature ratings – the first is the case temperature (Tcase), which is measured at the metal heat-spreader your CPU cooling sits on. The second is the junction temperature (Tjunction) – this is the temperature at the transistor or CPU die level. All mainstream CPUs also include a sensor to monitor this Tjunction temperature. However, as all CPU families are different, not just in brand or architecture, but in manufacture, each CPU type has its own maximum Tjunction rating.
To find these temps on Intel CPUs, head to http://ark.intel. com, look up your CPU model, look for ‘Package Specifications’, then Tjunction or Tcase. For example, the old Core i5-2300 only lists a Tcase temp of 72-degreesC, whereas the Core i5-9400 has a Tjunction max temp of 100-degrees C. AMD lists ‘Max Temps’ on each CPU’s individual spec page under ‘Specifications’. For the Ryzen Threadripper 3990X, it’s 95-degrees (tinyurl. com/1tfw1lu5). In general, Tjunction max temps are typically between 90 and 100-degrees C, but check your CPU manufacturer for your specific CPU. We’d also add the further below these temps your CPU stays, the better for its longer-term reliability.
How does TDP relate to temp?
TDP or ‘Thermal Design Power’ describes the power a CPU will need to dissipate when running at 100 percent load at its design clock frequency. For phone CPUs, this can be as low as a couple of watts, for some Intel multi-core desktop CPUs, it can reach 130-watts (130W) and AMD’s 64-core Ryzen Threadripper 39x0X CPUs can hit 280W. However, TDP itself is a poor alternative to determine maximum temperature because heat dissipation is also related to surface area.
This means a larger CPU can dissipate more power but at the same or very similar maximum junction temperature. A quick example – take Intel’s Core i5-9400 above with its 100-degree C Tjunction temperature. It has a TDP rating of 65-watts (65W). Compare that with the Core i7-8700K – it also has a 100-degrees C Tjunction max temp, but a TDP rating of 95W. The AMD Ryzen Threadripper 3990X has a 280W TDP rating but a 95-degree C max temp. So, TDP tells you the heat dissipation you can expect, but nothing about the maximum temp ratings.
How does overclocking affect temperature?
It should be pretty straightforward that if you increase the clock speed and maintain (or increase) the CPU voltage, you increase the power consumption and the heat to be dissipated. The key with TDP figures is that they’re the maximum-load power at the design clock speed. Overclock the CPU and you need to handle more power – this is why you might ditch the bundled cooler for something with more grunthandling capacity.
There are two thermal transfer processes in a CPU – the CPU die to the heat spreader and the heat spreader to the CPU cooler. The first happens internally; the second you take care of.
But if you think the CPU cooler is the most important component, I’m here to tell you ‘think again’. It doesn’t matter how good your cooler is, if you do not have a sufficient thermal transfer interface between the heatspreader and your cooler, your cooler won’t be working at its best. This ‘transfer interface’ is the thermal compound or paste you see either already attached as a pad to the CPU cooler, or you buy separately in a small syringe.
The CPU heat-spreader and the base of your CPU cooler might seem perfectly flat – but they’re not. The role of thermal paste is to ‘fill in the gaps’ to ensure a complete thermal connection between the two faces and minimise thermal resistance. So to get the most from your expensive cooler, don’t skimp on the quality of the thermal paste. And remember, just a thin film over the heat-spreader - you’re not slapping on sun-screen.
How do I know if my PC is running hot?
Intel lists a number of factors to look out for if you suspect your PC is overheating. These include the PC shutting down for no apparent reason, the cooling fan running continuously at high speed and the CPU clock rate being less than expected. DVFS is the CPU’s first line of defence to reduce power consumption and keep the Tjunction temp below maximum levels. If that fails, that’s when an apparently-random shut-down will kick in as the back-stop defence mechanism to prevent thermal damage.
Transistor power dissipation has been an issue since 1947 and the same principles also apply to CPU cooling. Removing heat as efficiently as possible will help ensure it gives you a long life of high-speed processing.