Hacker News new | ask | show | jobs
by badgersnake 82 days ago
Power is open. But nobody wants to build power devices for some reason.
3 comments

They are very expensive. Cheapest Power9 system Raptor Systems has is $6,794.99 and it has only 4 cores and 8GB DDR4 RAM and 128GB SSD. Reminds me of Sun Sparc pricing.

https://www.raptorcs.com/content/BK1SD1/intro.html

Yeah, that’s because nobody is building them so they have to use IBM server chips.
Why is the motherboard so expensive?
Same problem, low volume
Power?
RISC-V is also open. That “some reason” is likely to be power/performance levels being quite far from ARM & Intel for consumer devices.
China is building out RISC-V, just like they are leading actually-open AI.

https://www.scmp.com/tech/big-tech/article/3347684/alibaba-d...

Weirdly, the authoritarian state is the one saving us from our own digital authoritarians.

> they are leading actually-open AI.

How are they leading? If I parse this correctly, "actually" open would mean fully open data training and weights? Then, by this definition, I'm only aware of Olmo (AllenAI - Seattle), Apertus (Swiss) and to some degree (unclear what data was actually published) Nemotron (Nvda, US). What are some examples of chinese similar models? (I'm not aware of any).

RISC-V is slow though
RISC-V is an ISA.

There's nothing inherently slow about it, anymore than there is anything inherently slow about x86 or ARM.

High performance microarchitecture implementations are definitely possible. Some of them are available for licensing.

At least one of them (Tenstorrent Ascalon) has been tapped out into a chip and will show up in development boards later this year.

If you can’t buy a fast one, then it’s slow.

It might not be slow forever, but it’s slow now. I’d love to see an open ISA that is fast. But I don’t understand why the industry decided to start over with RISC-V when the compilers and toolchains and chips already existed in power land.

That’s no guarantee that a Power implementation isn’t compromised.