OpenWRT is remarkable in how many architectures it supports, with both ARM and RISC-V available. This just happens to be one board and it uses ARM, but it could easily be moved to an official RISC-V one when available with enough support to meet their demand.

P.S. you don’t get to start calling things “legacy” just because you would prefer something else. ARM is clearly at a scale right now where that’s the sweet spot for cost/performance/availability right now.

It works great on x86 as well.

I run OpenWRT in a few locations. One of them is VM on a x86_64 fanless (SuperMicro) server hypervisor. The other is a VM on a x86_64 fanless CompuLab FitPC hypervisor.

Along with OpenWRT, those systems also run a PiHole VM (https://pi-hole.net/), and other VMs such as FreePBX (https://www.freepbx.org/).

ARM is legacy because it predates the open ISA standard.

I am interested in this, even if it is based on legacy ISA ARM, as I would rather a hacker-friendly router designed for OpenWRT.

This is relative to some consumer device off the shelf which is hostile to OpenWRT, needing trickery to just install. That is what I had in my previous location, and I would prefer this instead.

Maybe but I think going with middling ARM chip is more the right call then trying to pursue a RISC-V based chip. The OpenWRT folks don't have the needed logistics chain to distribute this at all, and it sounds like they're leaning hard on both MediaTek and the Banana Pi team for this. I don't know of anyone else off the top of my head that would give a bunch of open source devs working on a niche distro that kind of support.

That said, the VisionFive 2 is available. There's some people on the Openwrt forums working on trying to get OpenWRT to work; might not be a bad idea to reach out to them and see how far they've gotten.

I have a VF2. It does have strong CPUs (relative to A53) and 2x GbE, solving half of the problem.

The other half (wireless) could be solved by using the PCIe lanes. To do it well at scale, we'd need a new board designed for the purpose.

The Pi4A guys meant to release such a router device, but seem to have elected to do other devices instead. Shrug.

That's a weird way if putting it. By that logic, a current Gen Intel CPU is also legacy. It might be carrying a lot of legacy cruft, but calling it a legacy architecture seems just wrong.

Let's not mix microarchitectures and ISAs.

It is very possible to make a great modern implementation of a bad ISA full of legacy cruft; Zen4 is a good example of that. Intel get a participation award for trying.

More architectures is good.

I don't want to replace ARM with RISC-V. I want ARM and RISC-V to compete with each other, micro-architectural progress to continue, and to be able to pick the best part for each task.

>More architectures is good.

I agree. This is why we need to support the new open standard architecture over the established proprietary one.

> a legacy (ARM) SoC.

That is not what that word means. There will be more new ARM CPUs shipped next year than RISC-V CPUs. It is actively developed and improved. And this trend of people calling existing ISAs "legacy" when they're not is really annoying.

>There will be more new ARM CPUs shipped next year than RISC-V CPUs.

Next year is 2025 for context. I wonder what data you're supporting this forecast on.

I meant in the next year, but it's probably true of 2025 as well.

My primary basis is continually trying to get a Linux-capable machine with a RISC-V CPU, which remains difficult. Look at the number of SBCs released each year for the last... I dunno, 5-10y? A bit of x86, a lot of ARM, and at the very end a single digit number of RISC-V boards. Which still are slow, expensive, or both. Even if RISC-V ships exponentially more boards each year it won't break even until after 2025. If you wish to offer data to the contrary feel free.

Now, there's a possible argument for number of non-Linux-capable chips, but I don't know how to gauge that; the... Cortex-M0? I think? Is still a popular option AIUI. Likewise, if you have actual evidence feel free to share.

(None of this should be read as dislike for RISC-V; I want to use it, but the market doesn't yet agree, and even when it does take off the competition won't magically become "legacy" at a result.)

Edit: Actually that last bit deserves more emphasis - while I believe what I wrote about numbers, it's secondary to the rest of the comment; RISC-V is great, but it in no way renders everything else "legacy" so long as they remain actively used and developed (the latter is important, but ARM continues to improve so the point stands).

>continually trying to get a Linux-capable machine with a RISC-V CPU, which remains difficult.

It wasn't hard for me to get my VisionFive 2. It shipped within two weeks in early 2023.

Have you heard the phrase "extraordinary claims demand extraordinary proof"?

You're asking yjftsjthsd-h to furnish data to make a slam dunk conclusion that ARM-licensed CPU designs will out-sell RISC-V.

You realize Apple Silicon is ARM, right? And 99% of all smartphones that have been sold in the last decade, and indeed in 2023, are based on ARM cores?

We are talking hundreds of millions of devices being built and sold each year, just in the phone+PC space. That's ignoring all other verticals where ARM designs can be scaled up/down to meet.

I mean, gosh, even if virtually EVERY company in the world, EXCEPT Apple, chose to drop all ARM IP tomorrow and switch to RISC-V, there'd still be a good chance the statement "There will be more new ARM CPUs shipped next year than RISC-V CPUs." would still hold true in 2025: Apple sell a lot of computers in various sizes, you know :)

> You're asking yjftsjthsd-h to furnish data to make a slam dunk conclusion that ARM-licensed CPU designs will out-sell RISC-V.

I think the statement that was made was slightly different: that there would be more newly shipping ARM parts than RISC-V parts (to show that ARM is still actively developed).

Which seems very likely, but is not quite a sure thing for 2025. (And it depends upon how we count: lots of RISC-V wins inside proprietary SoCs is probably not quite what we mean, but instead new RISC-V COTS parts we can put into our designs; but if you count the former RISC-V probably looks much better).

>> You're asking yjftsjthsd-h to furnish data to make a slam dunk conclusion that ARM-licensed CPU designs will out-sell RISC-V.

> I think the statement that was made was slightly different: that there would be more newly shipping ARM parts than RISC-V parts (to show that ARM is still actively developed).

I actually thought about it at the time and left it ambiguous because I believe both; I believe there are and will be more ARM designs, and I also believe there are and will be more physical ARM chips manufactured.

> I believe there are and will be more ARM designs

Just out of curiosity, what does ARM designs mean here?

- New distinct processor cores available?

- New distinct parts I can buy at DigiKey and through other channels that include an ARM core?

- New distinct chips taped out using an ARM core, including vendor specific chips?

The latter is where I expect RISC-V to surge first. Probably not passing ARM in 2024, but threatening to do so. It's pretty easy to get rid of a license fee at this point unless you need extreme performance or are really picky about tooling.

ARM may not win on the first measure, either, but that's largely the vagaries of how roadmap packs into a calendar year.

There's also the whole question of whether Cortex-M counts as ARM anymore, given that modern large ARMs no longer run the Thumb instruction set (and Cortex-M only runs Thumb).

I was thinking of tape outs when I wrote it, though again I think all those are true in the short term.

I guess we can argue semantics of what counts as "ARM", but Cortex-M sure isn't RISC-V so for the purposes of this conversation I'm inclined to unambiguously include it.

MIPS, Xtensa, and x86 aren't RISC-V, too, should we count them? ;)

It's hard to measure how big RISC-V is for vendor-specific SOC wins. There's clearly a whole lot of momentum. Performance and all of the new things ARM does aren't necessarily too relevant if you're designing a chip that goes inside a hard drive or whatever, but a few pennies a unit to ARM might be.

>You realize Apple Silicon is ARM, right? And 99% of all smartphones that have been sold in the last decade, and indeed in 2023, are based on ARM cores?

Emphasis on "cores". Relative to each one of these application processor cores, how many embedded smaller cores doing specialized tasks do you think there are?

Are you aware RISC-V already has penetration there? Do you know how e.g. the popular Qualcomm SoCs have shipped a bunch of RISC-V cores for a while already?

Are you aware the companies behind popular alternatives to ARM such as Arc or MIPS have already embraced RISC-V?

In short, what makes you think that in 2025, most of these cores will be ARM?

> Emphasis on "cores". Relative to each one of these application processor cores, how many embedded smaller cores doing specialized tasks do you think there are?

Fair point, but the context here is your original comment:

> Great idea, but not a fan of the timing yielding a legacy (ARM) SoC.

So you're talking about the main processing cores of the SoC, the one(s) that are responsible for running the Linux kernel + OpenWRT user space, right?. Not the random other 8/16/32-bit MCUs...

I've seen religious zealots who are less adamant than some of the RISC-V fans. Do people really self-identify with an instruction set architecture so strongly?

It is weird indeed.

But, irrational zealotry can be surprisingly powerful: Linux would not have happened without it.

So while it may seem ridiculous at times, sometimes great good (and great evil) comes from this disguised will to power, and it is part of human nature.

If you are going to be a zealot, at least be a zealot for something useful.

I think a lot of people are really fed up with the status quo.

Management engines, millions of lines of hdd controller code, bios/firmware blobs that are basically opaque, etc are annoying and prevent me from having control over something I own.

Rightly or wrongly a lot of hope is being put into RISC-V as a way of improving the situation. The open nature of it might make it easier to break monopolies, to get control back of the hardware they own.

It's not about instruction set - its about open source. No different than the people who didn't get that Linux was a big deal because they compared OS features as if that was the primary driver.

> Rightly or wrongly a lot of hope is being put into RISC-V as a way of improving the situation.

Those hopes are obviously misplaced. The ISA isn't why your hard drive's firmware is closed. Western Digital using RISC-V isn't going to make them more willing to open up about the non-CPU parts of their chips, where their competitive advantages reside. RISC-V isn't even going to make anyone change their decision about whether to design their own CPU core or license an off the shelf core. The only real impact RISC-V will have on the industry is to slightly improve the breadth and pricing of off the shelf CPU core IP available to ASIC designers—and that doesn't have any implications for the openness of products incorporating those CPU cores.

RISC-V cannot break a dam holding back companies from open-sourcing everything when that dam is entirely imaginary.

100% agree.

The fundamental problem here is the IP system; even open-source is just putting lipstick on a pig. The core problem is that if you pick any particular type of device, chances are that no company is building fully open-hardware devices; it's a systemic problem. What's worse is that this is a solved problem - patents originally solved this, except for some reason firmware have copyright protection despite being a functional design, and thus they are gifted the privilege of legal protection without the responsibility of publishing documentation.

What's more, firmware's copyright isn't tuned for the electronics industry - they get protection for 70+ years, when it's duration should be the minimum duration necessary to properly incentivize producing innovative firmware. That's a variable period so I don't expect it to be an easy problem, but frankly it's not even in the ballpark.

So hypothetically, we will be legally permitted to modify the firmware of today's routers in maybe 2090 - no source code though. In fact the source code might have been lost forever, because copyright was designed for books and hasn't been properly updated before being applied to software.

Patents tend to last 20 years, which IMO is a far more reasonable duration (if your 2004 router design still isn't profitable then sucks to be you, you had a very generous window) - IMO the firmware should require source code in escrow in order to gain copyright protection, and it's copyright duration shouldn't last longer than a patent would.

See also: the tragedy of the anti-commons.

Can you please stop posting repetitively about RISC-V? We've been getting complaints about it becoming a tedious pattern.

You've also posted some good unpredictable comments so I don't want to ban you, but basically anything that an account does that becomes predictable is what we're trying to avoid here.

One way to not be predictable is to include unique, relevant information in your comment, as opposed to posting something shallow. Another way is to just post on a variety of different things, without too much of a pattern.