The Thermal Performance Of NVIDIA's Jetson Nano $99 Developer Board
One of the exciting product launches for this month has been the introduction of the NVIDIA Jetson Nano as a $99 Arm developer board offering four Cortex-A57 cores that isn't too special itself but packing in a 128-core Maxwell NVIDIA GPU makes this board interesting for the price. Out-of-the-box the Jetson Nano is just passively cooled by a small aluminum heatsink, but does it work any better if actively cooled to avoid any potential thermal throttling? Here are some thermal benchmarks.
The Jetson Nano provides a lot of potential for under $100 when using software to leverage both the CPU and GPU with use-cases from building your own robot to DIY appliances or even having a nice hobbyist Arm Linux developer board with some "oomph" to it without spending much money. In time for the launch day earlier this month I didn't have the time to run any thermal tests, but here are those numbers. On launch day I only had a few days experience with the Jetson Nano but since then I have continued running performance benchmarks and it has been running great and without any issues -- also, no thermal issues to speak of, but decided to run some tests including with an active fan attached to see how that would perform.
For these complementary benchmarks today are the Jetson Nano benchmarks when running in its performance mode and being tested both with the "stock" cooling of just the passive heatsink and then "Fan Attached" with a small fan attached to the heatsink and connected to the PCB. Ubuntu 18.04 was running on the system with the Linux 4.9 Tegra kernel, as is the case right now for L4T/JetPack. A variety of workloads were tested to stress the system while looking to see if the performance changed at all with the fan running (indicating thermal throttling or different thermal behavior) as well as looking at the SoC temperature.