The POWER ISA wasn't royalty free until 2019, over a decade after its decline. RISC-V was royalty free from the beginning.

But thats still 4 years of basically zero POWER adoption.

I don't want to hate on RISC-V, but I just don't see a line from here to where ARM is now, even with ARM's shenanigans.

Is there any actual POWER core ip out there that people can build on or experiment with? Yes, the ISA itself is open, but last I looked that was it. I think RV's availability of implemented cores (even if they aren't the most cutting edge or high performance) is helping it build some momentum. Whether that's actually "ARM-killer" momentum remains to be seen, and might well not be... but I don't think I'll write them off quite yet, just based on IBM's long history of failures to position and market POWER well.

Microwatt is the only one I know of. Built as a tiny fgpa core.

https://github.com/antonblanchard/microwatt

Yes, there are open core IPs for POWER:

- https://github.com/OpenPOWERFoundation/a2i

- https://github.com/OpenPOWERFoundation/a2o

- https://github.com/antonblanchard/microwatt

ah, interesting, will need to check those out

I get the impression that devs are happy about RISCV and willing to port things to it, not so much for POWER. Also China seems to be pursuing RISCV heavily.

But where the rubber hits the road seems to be in niche stuff like accelerators (which is indeed a great application for RISC-V) and low power embedded stuff.

Other countries (like Russia, and China's MIPs thing) have pursued novel architectures for higher power general purpose computing, but its mostly for government use I think. China pulling off a societal transition to RISC-V would be quite a coup.

To be fair, a lot of software already supports POWER right now, including the Linux kernel, compiler toolchains, the OpenJDK, various database engines, and so on. Don't forget that a lot of embedded devices used PowerPC in the past and it was easy to get PowerPC based hardware like e.g. old Mac computers or game consoles. Today you can get large POWER based servers which compete with big multisocket x86 servers.

I think the main problem of POWER today is that you can't get a "Raspi sized" development board, while you can easily get all kinds of ARM based SBCs as well as various RISC-V based SBCs. The smallest POWER system you can get right now is iirc a workstation from Raptor which essentially competes with Xeon workstations, if you don't want to play around with soft cores on FPGAs.

Four years is nothing in HW timescales.

> I just don't see a line from here to where ARM is now

I don't understand why. RISC-V is being used and pushed by 1000s of companies at this point. You can license cores from 100+ vendors or you can make your own.

There are lots of companies pushing it into every single part of the market. There are well kown cores for space travel, the next generation is RISC-V. RISC-V is already in plenty of low level cores, in things like SSDs, NICs and so on. Companies like Andes push it into billions of consumer devices.

In terms of the ISA itself it is already a match or better then ARM in most ways. And in 1-2 years with the next rounds of specs, it will be more advanced.

To me it seems like RISC-V is already winning by far and its just a question of time. This is more like Linux in the late 90s vs commercial Unix.

1990s sounds seem to be all the rage in pop music right now. But the only ‘90s revival I want is the POWER ISA. Sadly it seems more unlikely than Sinead O’Connor appearing in a Justin Bieber track.

POWER was fake open, while not actually being close to open for a long time. When RISC-V gained adoption they made a panic move and opened it up.

But there is basically no reason to pick POWER over RISC-V at this stage. Not unless you are already invested in the POWER architecture.

RISC-V has far more OpenSource work, far more companies involved, the spec is driven and improved by a lot of people and so on.

The reality is most work on RISC-V doesn't go into super high performance stuff and if its there, then usually along side high performance AI stuff. SiFive, Esperanto, Tenstorrent and so on.

> I'm not sure how RISC-V is supposed to surpass that from square one.

Buy having 100x more companies working on it.