Note that this is not a microcontroller, but a full Linux-capable computer similar to Raspberry Pi etc.
TLDR:
The OrangePi RV2 a good board at a good price from a good (traditionally Arm) company, but some people will be better off with the OrangePi RV at the same price with the StarFive JH7110 SoC with SiFive U74 cores as it performs slightly better in the real-world if you don't need SIMD/vector.
----
I ordered an 8 GB on March 11 from the official OrangePi store on Aliexpress for US$49.90 which came to a total of US$67.20 including shipping to New Zealand and our 15% GST and received it on March 20:
https://www.reddit.com/r/RISCV/comments/1jeqt28/well_that_was_quick/https://www.aliexpress.com/item/1005008612193589.htmlDespite the different chip labelling, it is as expected the same as the Spacemit K1/M1 chips that have been in the Banana Pi BPI-F3 for a year (orders taken in March 2024, customers had them in May), followed by the LicheePi 3A (which I have), Milk-V Jupiter, MusePi, MuseBook laptop, DC-Roma II laptop etc.
> Let's hope linux patches will get upstream, in a timely matter.
The same Bianbu Linux (Ubuntu with manufacturer's patches) image works on all boards, has been regularly updated, and upstreaming is in progress.
Upstreaming always takes a while, so in the rapidly-moving RISC-V world (and to an only slightly lesser extent with Arm) if you wait until upstream works then the board is already obsolete.
Note that there is also the OrangePi RV with the StarFive JH7110 SoC with 4 SiFive U74 cores at the exact same price for the same amount of RAM i.e. also $49.90 for the 8 GB.
Note that the Spacemit boards are available with 16 GB RAM from other manufacturers (and my LicheePi 3A has 16 GB) but 8 GB is the maximum from OrangePI at present. The JH7110 is not available anywhere with more than 8 GB.
The OrangePi RV2 also has the advantage of having a newer RVA22-compliant instruction set, and the RVV 1.0 vector extension with 256 bit vectors. The OrangePi RV has only the original RV64GC plus Zba and Zbb (bit manipulation).
Despite the JH7110 having 4 U74 cores and the Spacemit 8 X60 cores, overall performance is very similar.
On micro-benchmarks the two cores perform very similarly:
// 14.685 sec Lichee Pi 3A SpacemiT X60 @1.6 GHz 214 bytes 23.5 billion clocks
// 14.885 sec VisionFive 2 U74 _zba_zbb @ 1.5 GHz 214 bytes 22.3 billion clocks
On tasks that generally fit into L1 cache the 8 X60 cores on the OrangePi RV2 will give an advantage over the 4 U74 cores in the OrangePi RV.
However the JH7110 gives 2 MB L2 cache, while the K1/M1 has only 512 KB L2 cache per four core cluster.
Doing "real world" tasks such as compiling a Linux kernel or gcc the end result is very similar:
Linux kernel:
42m12s EIC7700X Milk-V Megrez 16 GB ($199) with 4x SiFive P550 cores @ 1.8 GHz
67m35s JH7110 4x U74 @ 1.5 GHz with -j4
70m57s K1 8x X60 @ 1.8 GHz with -j8
97m59s TH1520 4x THead C910 @1.85 GHz with -j4
108m01s K1 with -j4
227m52s JH7110 with -j1
354m44s K1 with -j1
Building the GCC 9.2 toolchain supporting RVV 0.7.1 (snapshot on my github). This needs more than 8 GB RAM so I don't have VisionFive 2 figures.
68m56s EIC7700X
132m33s Spacemit K1
138m50s TH1520
See also my recent post on GNU MP library performance:
https://old.reddit.com/r/RISCV/comments/1jsnbdr/gnu_mp_bignum_library_test_riscv_vs_arm/