When I reviewed the Raspberry Pi 4 at launch, I highlighted its dramatically increased power draw and heat output compared with its predecessor the Raspberry Pi 3 Model B+. In the months since, the Raspberry Pi Foundation has been working to address the issue through a series of firmware updates – and, with assistance from Eben Upton and Tim Gover, my feature runs through each release and sees what difference it actually makes.
For the feature, I had to develop a method of benchmarking the Raspberry Pi. Power draw was relatively straightforward: the built-in current meter in a bench-top power supply is used to measure the minimal draw at idle and peak draw at load. For thermal performance, I wrote a custom benchmark which uses two open-source utilities – glxgears and stress-ng – to place a heavy load on both the CPU and the GPU while measuring the resulting temperature rise and the speed of the CPU, which throttles at 80°C to protect the silicon.
These measurements provided a graph of temperature rise and fall, the latter thanks to a five-minute cool-down period built into the benchmark, but for a more visual approach I also took thermal imagery of the board at idle and load to demonstrate which components are responsible for the heat output and better highlight the improvements made at each firmware revision. This was no small undertaking: the benchmarking and thermal imagery was completed for five firmware revisions, the last of which was not publicly available at the time of testing, plus a baseline Raspberry Pi 3B+ for comparison.
The feature also takes a look at a real-world workload, in which temperature and clock speed is measured while a four-worker compile of the Linux kernel is carried out. This revealed something which may come as a surprise to critics of the board: Using the latest firmware, the Raspberry Pi 4 did not throttle at all during the compilation – something that can’t be said for the Raspberry Pi 3B+, which throttled to 1.2GHz from 1.4GHz almost immediately. For the final bit of testing, there’s even a comparison of the Raspberry Pi 4 running sat flat on a desk and balanced vertically – at Upton’s suggestion – with a resulting dramatic impact on the throttle point and operating temperature.
Finally, firmware developer Tim Gover was kind enough to answer my questions on what the Raspberry Pi 4 firmware actually does, how it is developed, and how it can have such a dramatic impact on power usage – along with the confirmation that USB mass-storage booting and IPv6 network booting are on the to-do list for future releases.
The full feature, and plenty more beside, can be found at your local newsagent, supermarket, or downloaded at no cost in digital form under a Creative Commons licence from the official website.
One thought on “The MagPi, Issue 88”