Milk-V Second RISC-V SBC – Physically Compatible with the Raspberry Pi 3 Model B
hackster.io
hackster.io
Proprietary drivers are available and work.
Open source drivers in the form of a mesa3d driver are being developed by a third party employed by Imagination Technology. We know about this effort since early last year.
I do not know the state of the effort.
It's a nice chip, but would have been SO MUCH better if it has been available to board-makers in September 2021, as originally planned for the BeagleV "Starlight", not in January 2023.
In 2021 it was clearly a much faster alternative to the Allwinner D1 with a single 1.0 GHz core, which had started arriving in customer hands in late June. The D1's C906 core had a vector unit, but that was the only real advantage.
Now in the first half of 2023 the Pi 3-class JH7110 is running up against the Pi 4-class TH1520.
The VisionFive 2 and Star64 started arriving in the hands of customers in February and April, while the considerably faster (higher MHz and OoO, plus vector units) TH1520-based LicheePi4A started arriving in May. There will no doubt be other TH1520 boards soon too.
So far the 8 GB JH7110 boards are roughly $40 cheaper than an equivalently outfitted 8 GB TH1520 board. Perhaps the effective difference is more than that as as VisionFive 2 and Star64 have M.2 or PCIe while the LP4A doesn't.
So maybe there is a place for both. Let's see what price the Mars will be.
Sipeed now has at two RISC-V "compute modules" that follow the concept but don't attempt to be Pi compatible:
- LicheeRV module, Allwinner D1, same as the R-01, but with two side by side M.2 connectors as the interface. They sell a LicheeRV Dock for it.
- Lichee Module 4A, THead TH1520 SoC. Currently only available as part of Lichee Pi 4A, but they plan to sell the module separately, as well has a motherboard holding a cluster of 7 modules, a tablet, a laptop, maybe a phone. Uses a SODIMM connector.
In addition, Sipeed have a number of other lower powered RISC-V "module" products using K210, BL808, BL702. Most with circuit boards with castellated edges as the connector, some with some kind of edge connector. And some ARM ones too.
I have a too big TODO list.
Over two years ago, 95% of Debian (largest Linux distribution in packages) already built for RISC-V.
The last significant exception I was aware of (dotnet) finally merged RISC-V support earlier this year.
Dotnet doesn't yet have a RISC-V JIT. Samsung are currently doing that. A couple of months ago they had it to the point of a command line Hello World working, but there's still probably a 1-year project for a team to finish everything to production quality.
There was a message from the maintainer this week that Luajit is getting close to being usable.
Java JVM has worked for a long time but without JIT. I think that is also being worked on but I'm not sure the status.
There was a blog post this week with the status of Box64 on RISC-V with some benchmark results. It seems to be basically fairly complete but not yet polished.
I am glad to see the ball moving forward for Risc-V, I do hope they avoid some of the vendor locked extensions that many ARM based SBCs seem to face. Feels like anything other than RPi so far has been an exercise in relative frustration. I have my SiFive 2 that I haven't played with yet... with the work I saw a few weeks ago with Haiku might be fun to play around with.
https://www.haiku-os.org/blog/kallisti5/2021-11-07_booting_o...
Citation needed.
The Cortex M23 and M33 seem to have made a good step forward in efficiency compared to RISC-V, but those are microcontrollers.
I haven't seen any indication of the Cortex A53/A55 being more efficient than the U74, or the A72 being more efficient than the C910. In fact the reverse.
Maybe the A510 / A515 / A520 is better, but you can't buy an SBC with those.
> the environmental impact for the existing box is already spent.
That is certainly true.