Re: still not getting it
Let's see. You know Gigadevice, GD32F103... the STM32 clone. This same company released a set of pin compatible and equipped with similar peripherals GD32FV103 which is a RISC-V, they're offering it at, not sure now, at one point i looked and it was like half the price, and it is a little faster and substantially more power efficient. However i am well aware that the improvement is IN SPITE of RISC-V rather than due to it.
Espressif is also going RISC-V in their new low cost offerings, nothing changed much on the power front, and the performance got a very modest nudge up compared to their older Tensilica Xtensa based offering. It just stands to reason that there is no inherent disadvantage for these from adopting RISC-V, both being of the MIPS legacy, but they got a better deal on the core IP.
The typical advantage of MIPS designs has been specifically that they have a coprocessor interface - that custom DSP instructions and such can be lower-latency since they don't need to traverse the memory bus at all, it's all just register transfers. Depending on core design, they can also take advantage of existing pipelining infrastructure etc. And these are never particularly wild things, maybe you just need to accelerate computing checksums or something of a kind and you want to be able to make do with the smallest and least complex addition to an existing design. Or look at PS1's GTE for a well-documented practical example.
Chips already exist to satisfy all necessary custom functions? That they kinda do, but they also get refreshed once in a while, for one reason or another, they won't be up to date forever. Say NVMe flash controllers will need to be updated for next PCIe standard and the chip companies will go core shopping again, and at least Samsung seems to have chosen a RISC-V implementation for something like that, for the next round.
Things that end up winning aren't necessarily the best things, the most sensible things. A lot has been said about inefficiency of x86 as an ISA, and yet on the high end, it's fine, it's not what's holding them back. And ARM? Nobody cared much about them at first, at least until they grew a Thumb; and at that point, they became Poor Man's SuperH, and that's when they started to matter. RISC-V is likely to have a bright future ahead of it as a Poor Man's architecture. It will be taken as an opportunity for new entrants to compete with the established players, like we're likely to see something like Poor Man's PSoC - patents are running out aren't they? Lack of instruction set license costs, a large competitive field of IP suppliers pushing IP costs down, lack of dependence on a single entity for the future of the product lineup, and the cores, they're good, they're fine.
Biting the hand that feeds IT
