From the website:
Further more, a dedicated hardware NPU accelerator coming up next for ROCK Pi 4 will boosts complex Machine Learning algorithm and reduce the power.
So does the v1.4 board have an NPU or not?
If it does, does anyone have experience with it, running Debian or Ubuntu?
No, the RK3399 has no NPU, and there is no external one on the board. I believe they’re referring to possibly making a hat for it.
You are right.
But we did made a NPU HAT for ROCK Pi 4, which is not officially announced yet.
Sounds awesome, looking forward to that announcement. (AI isn’t really my field but it’s awesome to see what your team is coming up with!)
Thank for clearing that up! Can I ask, why a hat vs an AI edition using the pro chip? Will the hat be more powerful?
You mean this? https://forum.radxa.com/c/rockpin10
Well, the N10 is more than twice as big (100x100 vs 85x54) and one USB 3 port sticking out one side and the other one at the opposing end makes the total installation even bulkier. For that larger footprint there are a lot of good boards available, some of which perform considerably better than the RK3399 for not that much more.
For that larger footprint there are a lot of good boards available, some of which perform considerably better than the RK3399 for not that much more.
such as?
The HiKey 970 for example.
Performance comparison:
The HiKey 970 board is labelled as ‘ARMv8 Cortex-A73’.
What about the most important and expensive component? Software support.
That’s highly subjective and use case dependent. Also not really related to what was discussed (performance per footprint).
There is so many hardware on the market that looks cool, its great on paper, but you actually can’t use it for real, is driven by closed source, is supported only by vendor, etc. Everything matters.
Yes, its indeed case dependent. I strongly agree on that, but Antutu score is usually not exactly a use case.
Orangepi 4B is RK3399 with NPU on board.
And Debian and Ubuntu has nothing in common with hardware functions.