C906 vector unit: 128 bit registers, in-order single pipe, most instructions take LMUL3 clock cycles. C910 vector unit: OoO, two vector pipes, single-cycle exe

C906 vector unit: 128 bit registers, in-order single pipe, most instructions take LMUL3 clock cycles. C910 vector unit: OoO, two vector pipes, single-cycle execution, 128 bit registers but 256 bit ALU (I believe this means LMUL=1 and LMUL=2, for example, have the same execution time). The article tests C906, which has been out in cheap boards for 2 years -- I posted my first vector-accelerated benchmark of one in April 2021: https://hoult.org/d1_memcpy.txt https://hoult.org/d1_memcpy.txt https://hoult.org/d1_strcpy.txt https://hoult.org/d1_strcpy.txt A cheap ($99) quad 2.0 GHz C910 board is dropping, hopefully, in the next week or so: https://sipeed.com/licheepi4a/ https://sipeed.com/licheepi4a/ Even better (for those with $$), a *sixty four* core 2.0 GHz C910 chip (SG2042) and dev board is available for pre-order in China. I have ssh access to one and have just started running some tests. Looks good so far. If anyone has some relatively simple test I'm happy to run a few.