News: 1729704612

  ARM Give a man a fire and he's warm for a day, but set fire to him and he's warm for the rest of his life (Terry Pratchett, Jingo)

RISC-V reaches milestone with RVA23 profile ratification

(2024/10/23)


Comment The ratification of the RVA23 profile for RISC-V marks a monumental moment for the architecture, and anyone who's been following RISC-V knows that this isn't just a checkbox.

RVA23 is a long-overdue unification of the instruction set architecture (ISA) that effectively gives RISC-V the structure it needs to compete with giants like Arm and x86 - without the legacy bloat or licensing headaches.

This standard lays out a consistent set of ISA extensions that software developers can rely on across RISC-V hardware, which is no small feat considering RISC-V's open source DNA invites a potentially messy degree of fragmentation.

[1]

This is a big deal for RISC-V, but why should you care? Well, if you're a follower of what's happening (or even just a bystander with a slight interest) let's get into it: What the ratification of the RVA23 profile brings to the ecosystem is standardization of ISA across multiple implementations which brings a certain level of consistency that RISC-V has historically lacked: to deliver an ecosystem it requires hardware, software, and everything in between to be compatible with each other.

[2]

[3]

RVA23 brings features such as vector operations, floating-point and atomic instructions, which are all essential in - sorry here's a buzzword for you - AI. But outside of AI, it also brings capability for machine learning and other elements of high-performance computing which other ISAs bring to the table.

The vector extensions are especially crucial here. Without them, RISC-V would be nowhere in the world of AI, datacenters, or high-performance computing. Much to our delight, it seems as though RISC-V is moving from being a novelty, that different kind of architecture, an underdog, to a real player in parallel workloads that demand serious computational power.

[4]

The key here for RISC-V and what RVA23 offers is it eliminates that fragmentation within an ecosystem, where developers can actually develop for an instruction set, and the software itself works seamlessly across the architecture; or that's at least the goal here.

Ratification isn't just about having more instructions to play with; it's about achieving consistency. RVA23 means that whether you're coding for a RISC-V chip in a toaster, a router, or an HPC cluster, you're targeting the same basic set of features. That's how you scale an ecosystem, and it's how you stop developers from thinking such things as: "Hmm, maybe I'll just stick with Arm for now, the ecosystem works, it's easier to work with."

But where things get really interesting is the inclusion of hypervisor extensions. Virtualization is critical in modern compute environments, and it's one area where RISC-V has traditionally been behind. The RVA23 profile finally sets the stage for more efficient virtualization on RISC-V chips, and it's not just about running multiple OS instances. For datacenters, this is massive—consider the potential for RISC-V to undercut x86 and Arm in environments where licensing and power efficiency are key.

[5]Xen to RISC-V port progresses with foundational efforts

[6]Why RISC-V must get its messaging right on open standard vs open source

[7]US government reportedly ponders crimping China's use of RISC-V

[8]SiFive expands from RISC-V cores for AI chips to designing its own full-fat accelerator

So what's next? Well, RVA23 is like the foundation, but the next step is actual silicon that takes advantage of these standards. With the ecosystem now starting to look somewhat stable, or at least that's what this is meant to achieve, we're likely to see a wave of new RISC-V hardware that's no longer just about low-power IoT chips or achieving an exercise of academia.

For datacenter operators, AI researchers, and anyone looking for an alternative to the Arm/x86 duopoly, they should certainly be paying close attention, because RISC-V is no longer the underdog. This challenger is sharpening its teeth. RVA23 is like an aggressive growl that means RISC-V is moving forward, and not remaining static. ®

Get our [9]Tech Resources



[1] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2ZxlyBTK4FuHbq-6fef6JdwAAAMk&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0

[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44ZxlyBTK4FuHbq-6fef6JdwAAAMk&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0

[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33ZxlyBTK4FuHbq-6fef6JdwAAAMk&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0

[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_onprem/systems&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44ZxlyBTK4FuHbq-6fef6JdwAAAMk&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0

[5] https://www.theregister.com/2024/09/24/xen_risc_v_port_progress/

[6] https://www.theregister.com/2024/05/29/riscv_messsaging_struggle/

[7] https://www.theregister.com/2024/04/24/us_commerce_china_risc_v_investigation/

[8] https://www.theregister.com/2024/09/19/sifive_ai_accelerator/

[9] https://whitepapers.theregister.com/



I say this as a RISC-V fan...

3arn0wl

What RISC-V really needs now, is to be able to boot standard Linux images easily.

RISC-V is no longer the underdog(?)

Anonymous Coward

RISC-V is like the love handles of CPU archs ... it's great to have it around, but other parts perform better by design -- at least as far as Oops! and dynamic performance are concerned (eg. LISPs and Smalltalks were better served by RISC-III and RISC-IV).

But yes, if your computer is trying to be a large macro-GPU (for matrix-vector " AI ") then RISC-V will do just fine ... and in this case, might as well go all the way, hog wild, and weave those cores into a dataflow arch, for near-memory compute, IMHO, and make those cores sweat the good sweat!

Albert Einstein, when asked to describe radio, replied: "You see, wire
telegraph is a kind of a very, very long cat. You pull his tail in New
York and his head is meowing in Los Angeles. Do you understand this?
And radio operates exactly the same way: you send signals here, they
receive them there. The only difference is that there is no cat."