Hacker News Reader

Comment by ladyanita22 | Hacker News Reader

Why the focus on Risc-V when the Power architecture already has high-performance cores? Genuine questions.

OpenPOWER was created to sell chips more then anything else. But many people don't just want some stanard 'high performance' chip. There are many different kinds of high-performance chip. A high-performance chip, for a car, a gaming computer, a AI chip or a cloud chip that runs VM are all different.

There is simply not that amount of options in the POWER market. Its a restrictive set of vendors.

Pretty much every new startup designing high performance chips picked RISC-V. And they optimize their chips for specific use-cases.

In addition, the software ecosystem for RISC-V is arguable already better then POWER. And many companies and universities are working together on making it first class. Android is going to RISC-V, it will never be on POWER.

So if you were to design something for consumers, not going with RISC-V would be crazy.

RISC-V is also evolving far more quickly. There are like 80 working groups working on improvements all in the open. The amount of companies working on POWER is tiny in comparison.

Didn't that Android port to R5 get partially walked back?

Power and RISC-V are going after different markets.

While Power is IRSC has a lot of legacy baggage that makes it significantly more complex that RISC-V (which is clean slate and purposely simplified.) It is also definitely not optimized for low power situations / embedded.

It is basically super computer chip.

Also unlike ARM and RISC-V it isn't designed to be extensible/customized by users. It is more monolithic.

I wouldn't say "super computer" chip (or at least "supercomputer")...

I've read the definition of a supercomputer as "some computer that takes specific domain, compute-bound problems and turns them into IO bound problems." (Implicitly, in that statement is the second statement of "and they have a ton of IO too). They're not really general purpose computers, and likely you'd be able to use risc-v or anything else, with specialized hardware, as the basis of a "supercomputer".

The power platform, on the other-hand (I'll create a new word here) is a foundation for a "SuperEnterprise(tm)" computer.

Power has insane / awesome things like "oh, you can use the ECC bits for ECC and hardware memory tagging[1]." Eventually such things may trickle down to things like ARM or RISC-V, but they're pioneered at the top of the enterprise mountain and trickle down...

[1]https://www.devever.net/~hl/ppcas

I guess I was referring to the fact that Power ISA chips were dominant at once time (prior to the rise of x86 chips) in the TOP 500 listings:

https://en.wikipedia.org/wiki/TOP500#/media/File:Processor_f...

They have faded from that over time from this niche. I do not know where they excel now.

If you're a student in a microarchitecture class who has to implement a core in a semester then the simplicity of RISC-V makes it a much better choice. And if you're a grad student trying to try out some idea in your thesis then you should also choose RISC-V for the same reason.

If you're creating a microcontroller for a specific product then RISC-V's ability to only implement the hardware features you need is a big benefit and its compressed instructions let you save money using fewer RAM chips.

The people I know using RISC-V don't want a high performance core. They want a very small area low power core for things like PCIE and DDR link training.

Reply on news.ycombinator.com