Having followed and forgotten about the UnixBench GitHub issue someone else raised a while ago on support for compiling on the RISC-V architecture I was pleasantly surprised to receive an email saying that support had been merged and approved. Now we have RISC-V UnixBench, let’s see how it goes!
Now, it’s not going to be groundbreaking, UnixBench is still a very old piece of software as far as things go and the tests within it aren’t the most representative but hey, some people still use it for comparisons so hopefully these numbers mean something to them. The usual disclaimers should be given that these types of test rely heavily on the software side of things too, so please do take everything with a pinch of salt, these aren’t meant to be scientific and super strict tests, more something that can be used to get a rough idea on performance.
Table of Contents
What are the MangoPi MQ Pro and StarFive VisionFive 2?
Getting to the testing though, I’ll be using the Allwinner D1-based MangoPi MQ Pro and the StarFive VisionFive 2, which runs the StarFive JH7110 SoC. The obvious differences here will be that the MQ Pro only has a single 1GHz CPU core, whilst the VisionFive 2 packs in 4 1.5GHz cores. The MQ Pro comes in a Raspberry Pi Zero form factor. The VisionFive 2 is a Pico-ITX (100mm x 72mm) board so they’re clearly targeting different use cases and the performance/power specifications reflect that.
It should be obvious given the above that there will be a big gulf in difference, though this is not meant to be a comparison per se, I’m just presenting the data in a single graph for ease.
Now we have support for RISC-V UnixBench, the installation process is much the same as you’d perform on any other platform. To download the software (you may need to install
git if you don’t have it already) you’ll need to clone the GitHub repository:
git clone https://github.com/kdlucas/byte-unixbench
To compile and run the test:
Benchmarking Results: MangoPi MQ Pro & StarFive VisionFive 2
Analysing the Results: Strengths and Weaknesses
As you can see, we have a clear difference between the 2 boards and that’s fine. The Allwinner D1 SoC was released in April 2021 and is running at 1GHz, which when compared to the 1.5GHz cores on the JH7110 which from what I can find, was first available in 2022.
If you’re looking for a low-cost (well, lower-cost) way to get into the RISC-V SBC space then the MangoPi MQ Pro is a solid option and at this point, has relatively “good” support and most of the features you’ll need to get started, such as WiFi. The VisionFive 2 does show that its 1.5GHz cores are in some areas, up to 10-15x “faster” but as I was unable to match the kernels and software versions exactly, some of this could be software differences but there’s no doubt that the larger, stronger JH7110 cores are pulling their weight here.
As I eluded to at the beginning, this was never meant to be a particularly scientific piece, but as RISC-V support was pushed to UnixBench directly, I thought it would be good to test the boards that I already had. If you’re interested in seeing how the MangoPi MQ Pro performs when compared to similar boards, then please check out my Great Pi Zero Showdown comparison. If you’re looking for more VisionFive 2 benchmarks then I uh, I don’t have any yet but I promise it’s coming, it’s just in my backlog.
If you have another RISC-V board, follow the steps above and let me know what numbers you’re seeing! RISC-V is exploding with popularity and is progressing at an incredible rate so it’s going to be incredibly interesting to see how the ecosystem involves in the coming years. Will it overtake ARM in the space? Let’s wait and see.