Ask HN: Does anyone care about OpenPOWER?

75 points by sandwichbop ↗ HN
I see a lot of energy around RISC-V but I never see anything similar for OpenPOWER, on paper it seems like the dream machine with actual performance like the Talos Workstations albeit a bit expensive but this sounds incredible https://www.raptorcs.com/content/TL2PA1/intro.html >Designed with a fully owner-controlled CPU domain, you can audit and modify any portion of the open source firmware on the Talos™ II mainboard, all the way down to the CPU microcode.

is there something I'm missing? why does no one seem to care about OpenPOWER

112 comments

[ 3.0 ms ] story [ 193 ms ] thread
$6,804.74 for that system.

This is for people who want IBM + total auditability? Who have severe commitments to the ecosystem already?

it is expensive, as far as I understand it (I'm still just learning about it all), it is for those need total auditability, they mention EU a lot but also this seems catered to actual individuals who really want a truly open system that doesn't scarifice performance (RISC-V seems great for embedded but is still not there for high end workflows). To me, this sounds like a reasonable price for what they're providing but I understand it's definitely outside the price point of many but I must wonder if there's anything more besides price, is it really just that they're too expensive and this is why there's so little talk about it?
I have vague memories of PPC instructions sucking compared to alpha, sparc, and x86, 20 years or more ago. I've got no specific memories what my objection was; but I recall dropping my plans to support it as a platform after a very quick look at it.

Perhaps it didn't have atomics at that point? That would've killed it for that project.

PowerPC uses LL/SC for synchronisation, which is very different to the classic x86 approach. Maybe you didn't like that? But i think Alpha and ARM use that too.
It seems unlikely that Power completely lacked atomics. This is a very vague recollection since it's been a long time, but I feel like there were some differences. I think PowerPC used a load-linked/store-conditional with a stronger guarantee than a CAS (compare and swap) on x86. For example, if x = 1 and you CAS it to a new value, another thread might have changed it to 8 and then back to 1 in the meantime and it'll still succeed (the ABA problem: https://en.wikipedia.org/wiki/ABA_problem).

I think x86 also has a stronger memory model providing less flexibility, but greater consistency.

> Perhaps it didn't have atomics at that point? That would've killed it for that project.

That would have been PPC 601 and perhaps [entry level] 603 models. Motorola/IBM first released a single processor PPC, then they added support for multiprocessor systems which did have atomics – https://devblogs.microsoft.com/oldnewthing/20180814-00/?p=99...

Other than that, I immensely enjoyed hand writing PPC assembly – it is a very straightforward RISC design, the «rlwinm» / «rlwimi» instructions are fun – once you figure out how to use them.

It's crazy expensive, Raptor torched their own roadmap, and no one actually needs what they're selling. (People say they do, but 99% of those people haven't actually bought it.)
> Raptor torched their own roadmap

I don't think that's a fair summary; Raptor has a very specific product, more or less the selling point of which is 100% open and auditable systems, so when IBM came out with POWER10 that bakes blobs into the system they're stuck with completely switching course. That's not torching anything, it's them being forced to correct for external factors.

But they're free to design and build their own chips without such blobs, right?

Right?

<insert Padme meme>

I'm not sure I follow your point. They're a vendor using POWER chips to provide blob-free computers. I'm fairly confident that they could make their own chips, but given that they'd need a fab and call it a billion dollars that's an impractical step up from building machines around existing chips. None of this constitutes them having "torched their own roadmap", just being stuck between a rock (their product being decent performance blob-free computers) and a hard place (needing to source CPUs).
When someone says "build their own chip" they do not in general mean to extract their own silicon ingots from sand, or to build their own fab. They mean to "tape out" their own SoC design and send it to a fab owner such as TSMC, Samsung, Intel for the physical manufacturing step.

Apple does not own a fab. Arm does not own a fab. Qualcomm does not own a fab. I believe IBM does not own any modern fab (some built in early 2000s, yes).

Yes? When I said "they'd need a fab and call it a billion dollars" I meant they'd need the use of a fab, not that they'd have to own the factory themselves. I don't see how it matters anyways
I'm not sure where that billion dollars is going to go then.

SiFive has raised a total funding of $366M over 6 rounds and 9 years, which has been enough to give them leadership in the RISC-V ecosystem and a product range spanning virtually the entire range ARM's does -- from equivalents for Cortex-M0 (in fact lower) up to Cortex-X3 (and POWER9) for their latest core announced in October (P870, 18 SPECInt2l6/GHz).

Part of the problem was that IBM (for reasons that have not been made entirely clear to me ) made Power10 require not only a non-redistributable firmware blob on the chip, but also on the off-board memory controller, making unsellable by anyone not named IBM.
If you're not used to buying server-grade hardware, I know it seems expensive. I think the founder initially priced it too low in an effort to make it as affordable as possible, found himself giving tons of expensive support, and decided to raise prices to the market average.
But Raptor machines aren't server-grade. They're desktop motherboards with crippled Power9s at higher-than-x86-server prices.
I don't follow: how are they not server grade? How are they crippled Power9s? What x86 server are you comparing them to?

For example, the Talos II system has dual (server) CPU sockets with up to 24 cores/96 threads per processor; 16 ECC memory slots with 2TB capacity; SAS controller; out of band management and service processor (separate serial and ethernet ports for the BMC); a server form factor; better capacitors; no onboard audio and only rudimentary graphics; and a thick/quality PCB. It's also made in America. What else were you looking for?

There are some people who care, otherwise it wouldn't even be a thing... but as you say, it doesn't get a lot of attention. I think it would be interesting to hear why? Why do companies implementing RISC-V and actively working on developing it not choose OpenPOWER instead? For that matter, OpenSPARC? I am not well enough versed in instruction sets to really comment, but I would love to hear more from anyone in the know?

Perhaps it is because IBM and Oracle exercise too much control over their architectures making it hard for a community to develop around them? Perhaps it is something in the licensing? Perhaps more fundamental problems with the design of the other instruction sets, making RISC-V easier a better base to improve on?

I did some looking, and it appears that OpenSparc is copyleft (GPLv2) which, combined with Oracle Yuck, may be a huge reason it is held back from wider interest. However, OpenPower stuff seems to all be Apache licensed from their repos. I could not find the RISC-V license, but it appears to be extremely liberal as well.
OpenSPARC is apparently used as a base for the Elbrus CPU line in Russia

https://en.wikipedia.org/wiki/Elbrus-8S

That specific chip I linked to the article about has the ability to run x86 code via translation and they claim it can run Windows. Guess that solves the chicken and egg issue, but the biggest problem is who would be willing to fab it for them, sad that politics and war interrupted this.

Interesting! The article says windows XP and 7, which are pretty old and unsupported now. But it is a very interesting approach.
Elbrus are VLIW/EPIC, based on home-grown architecture.

The SPARC-based one is the old MCST R1000.

>who would be willing to fab it

Before the conflict, they used TSMC. Now, AFAIK, the only fab available is Mikron, located in Zelenograd near Moscow. And only 90nm process. There were some rumors about switching to SMIC, but I have not heard any updates for quite some time.

IBM is currently shipping Power10 while Raptor ships Power9 in the Talos II. If I understand correctly, they had issues getting to a fully open source firmware solution for a Talos III with a Power10 in it.

This actually made me hold off on spending $10k on a computer that's last-generation technology. I suspect I'm not alone - a lot of interest in OpenPOWER is probably waiting for Power11 and Talos III or whatever permutation that can ship a real product that isn't 5+ years old.

I think that POWER in gaming was drowned by AMD, and ARM suffocated it everywhere else.

Apple dumped it because the G5 was not going to work in a laptop, and if it couldn't beat x86 in that space, then the M1 has closed that door forever.

The Cell was an innovative design, but an AMD core and an ATI GPU on the same die was an onslaught that IBM wasn't going to survive.

ARM has been the top supercomputer, and it runs in tiny things. The pervasiveness that it has came at the expense of architectures that were not as flexible.

I do care about big iron, about open hardware, but... I do not have the financial firepower to help developing a computer architecture, and buying expensive iron with limited software support is not much interesting. At least if some major distro do support OpenPower as a tier-1 arch, meaning almost all packages are there, updated at the same speed (almost, at least) of amd64 well... I can buy a classic workstation and making that a main personal desktop. I can't do much more otherwise...

PS while I prize and want open hardware I really doubt it can really be auditable at hw level, at least for most owners, even if technically well skilled. Projects of a certain size can be known only if they are FLOSS from the first SLoC in a way a spread community born around them, knowing them from the start and passing knowledge.

The software support is good. I use it and don't lack for pre-built packages or updates. For example, it's supported by Ubuntu Pro.
Hardware Decimal floating point is very useful, both for legacy and new financial systems and some numerical problems you need improved stability on.
Although frustratingly loosing support with the new open xlc compiler!
Dropped in the latest Power chips I believe.

I had a friend who worked on it: he called it simply "decimal floating" because there was no point :)

The Talos stuff is cool but it's priced way too high for anyone who isn't taking a business expense tax deduction for it. Especially post 2020, the pricing for their products has gone way up.

It's what, $6000 for the 4-core, 8GB memory entry level model?

I know it isn't that insane considering what new and used POWER servers cost, but also is anyone using these who isn't locked into AIX or IBM i? Is there any real reason to use POWER when priced against commodity AMD64 machines?

I don't think you can run (legally) Aix on Talos. And I doubt you could run 'i' as it needs special firmware support.
I'm aware, I was more asking if anyone is invested in the ecosystem for reasons other than being locked into proprietary IBM operating systems.
If you're used to building your PCs, you can build an 8-core workstation with a bunch more RAM for about $4000. While it's about double what you might spend on a similarly capable commodity desktop, you're getting a Libre boot process, fewer blobs, and a different class of chips.
If a company is serious about making POWER based SBCs or motherboards that are affordable yet fast enough for enthusiasts (no ECC, commonly used components), there would be a lot more interest. The price is what would make me turn away from POWER, as much as I love it and the ability for it to be based on completely open hardware. Someone with a lot of new ideas and a lot of energy doing something crazy like x86 binary translation or a Cell-styled CPU would also bring more attention to the platform.

RISC-V built a lot of traction very fast and was affordable and is now starting to be competitive with ARM, so it has different circumstances around it.

There's a group developing a POWER based laptop with a quad-core NXP processor, I've been watching them since 2020 and they've made some pretty good progress. It even has an MXM3 slot for adding a dedicated video card.

https://www.powerpc-notebook.org/en/

This was definitely the problem. I was interested in OpenPOWER when it came out but there were almost no vendors in the space. Expensive Talos and IBM workstations and a handful of dev boards were pretty much it.
> There's a group developing a POWER based laptop with a quad-core NXP processor

What I've seen from that project in the past does not fill me with confidence, and nothing I've seen since has changed that impression. Even if the project results in working hardware (which is uncertain), its performance is unlikely to be on par with expectations.

Previous discussions:

https://news.ycombinator.com/item?id=23988511

https://news.ycombinator.com/item?id=28578021

I have felt the same way about it a bit, and I do think they need to change a few things if they plan to get anywhere with it. I do hope though that they're able to succeed, they aren't stuck with that design forever and what they learned could be applied to a successor model. They should really consider partnering or merging with Raptor Computing, the last thing the scarce POWER market needs is more fragmentation.
I don't think they have anything useful to contribute. The PowerPC Notebook project has had prototypes in hand since Dec 2022, and bringup progress is essentially zero. It took them until July 2023 to realize they needed a JTAG debugger, and as of their last update in December, they still didn't have U-Boot working. At the rate they're going, a lot of the parts they used are likely to be discontinued before the computer is usable.
> RISC-V built a lot of traction very fast and was affordable and is now starting to be competitive with ARM, so it has different circumstances around it.

And, N=1, for all that I'm default-interested in new / less common CPU options, RISC-V only recently became really interesting with the availability decent-enough Linux-capable hardware <$100. This especially matters when it's competing against a plethora of sub-$100 ARM SBCs.

>no ECC

Please no. ECC is a must.

Not having ECC being common is an abnormal, bad situation to fix, rather than preferred.

It's simply too expensive. People that want to play around with a non-x86_64 system can buy one of a thousand different ARM devices and get all sorts of software running on it. If you want POWER, you're buying old datacenter gear with all the downsides of that class of hardware - or you're buying a Raptor system where the price has skyrocketed in recent years for aging hardware.

I pre-ordered a Blackbird motherboard and 32 thread CPU and got it in 2019. I used it as my main workstation until 2022 and then decided I'd had enough fighting the software ecosystem. I still have the machine because I've regretted selling other odd hardware in the past ... especially my dual 133mhz BeBox.

This. There are no inexpensive Power SBCs like we have for ARM and RISC-V. Heck, even MIPS has ways to buy and run inexpensively (I run NetBSD on Ubiquiti Edgerouter hardware, for instance).

If we could have a low end, quad core 2 GHz Power SBC with 4 gigs for $80 and a microATX motherboard with 16 cores at, say, around 4.5 GHz for under $500, the ecosystem would be VASTLY different.

Right now, all of my PowerPC work is on an old 1.5 GHz PowerPC G4 Mac mini and on an even older first generation iMac upgraded with a 600 MHz PowerPC G3. It'd be nice to have new hardware that didn't cost more than a decked out Ryzen 7950X3D system.

I would completely seriously buy that hypothetical POWER SBC immediately.
(comment deleted)
This. And that extends to the datacenter too.

So, why spend all the time improving compiler outputs for a platform that doesn't have that much traction or perceived potential?

>and then decided I'd had enough fighting the software ecosystem.

And by all indications, such ecosystem is now never going to take off.

In contrast, RISC-V is rapidly growing the strongest ecosystem.

(comment deleted)
I also have a Blackbird, and haven't had any issues with getting precompiled software or compiling software for it. What issues have you had?

I find it more performant than the kind of ARM hardware I think you're referring to, and more satisfying to use. I'd probably buy it again if mine died.

A few major items were pain points for me:

* Firefox didn't have a Javascript JIT. That made using Grafana, GCP Console, etc, basically impossible in my browser of choice.

* Closed source Electron applications (Teams, Slack, etc) weren't able to run natively, and the browser solution worked poorly for me.

* I was using Void Linux PPC, a one-man fork of Void Linux. I decided that the bus factor was too high, and since I wasn't willing to switch distributions, I had to switch back to x86_64. This proved to be a good decision, because shortly after I switched, the maintainer of Void Linux PPC announced that he was stopping maintenance to work on Chimera Linux.

There were other little things that accumulated to create enough mental pressure to abandon it as a daily driver. It's fantastically neat hardware, and in my experience utterly reliable. I can't think of a single machine crash I had in the entire three years I used it daily. It can and does work well for plenty of people - it just wasn't quite right for me.

I suspect IBM opened up the ISA (and some hardware designs) because they were quickly losing market relevance. Apple and gaming have moved on, new projects will use Arm or RISC-V, so the only markets I can think of are:

* automotive and other legacy embedded applications

* data centers with existing POWER applications

* niche workstations like the Talos

I do enjoy alternative ISAs, so I'd love to be wrong on this.

Gaming still may be interested if there's compelling enough options to get a leg up over the competition. The PS4/XBONE and PS5/XSX were both identical hardware, but that's starting to drift away with the PS5 Pro and the recently leaked portable Xbox. Who knows, x86 performance may hit another dead end for a while like it did the early 2010s and ARM may not fit their performance profile. I would mention Nintendo but sadly I don't see them making another non-portable console.
I won't deny that it's possible, but I think it's unlikely. The introduction of AMD SoCs like the Steam Deck and Z1 suggest that a portable Xbox is possible with much less overhead than switching ISA (again)

I'm not even sure if the performance race is going to endure - Sony continues to hold the home console crown, Nintendo prints money with their hugely outdated Switch, and Microsoft appear to be pivoting to publishing more than a console exclusivity race they will probably never win ("going Sega")

As you said, the future of x86 isn't guaranteed, but if I were calling shots I'd look at Arm designs with embedded graphics like Exynos and Snapdragon before trying to build a fully custom design based on POWER and graphics from elsewhere (probably still AMD, and that single-source has been compelling enough to adopt even while the CPU performance was lackluster)

I think a relatively substantial issue with swapping ISA is the software library, which is what's kept Intel in business all these years. Older/retro games is a large part of gaming, and being able to play 15+ year old PC games easily with a single download and no kajoogling is a major 1-up for the Steam Deck, especially when some games have better compatibility under Wine than Windows 11.

The Xbox Series consoles are capable of playing a select portion OG Xbox and Xbox 360 games almost natively, while presumably playing Xbox 360 games via an emulator, but they stopped adding new games to their backwards compatibility program for some god unknown reason despite the fact they were some of the best selling on the platform, but to me it seems like Microsoft wants to shift literally everything to the cloud, starting from Xbox and eventually getting to Windows somewhere down the line, which scares me honestly. Sun said the network is the computer, but I'm sure they meant a network you controlled, not goddamn Microsoft.

I believe the issue with backwards compatibility was licensing, especially for games like JSRF that Phil Spencer explicitly wanted to add.

The 360 games I've played BC on a Bone have downloaded the whole game onto disk, even with the original DVD in the drive so I'm not sure how much they're changing behind the scenes.

Regardless, I share the sentiment that running everything on Microsoft's infrastructure is a troubling trend. I like xCloud for managing JRPG inventory on the bus, or demoing a game before I commit to the full download, but it's no replacement for local gaming and it shouldn't be pushed as such.

Iirc OpenPOWER predates RISC-V by two years
The OpenPOWER Consortium was created in August 2013, more than three years after the RISC-V design effort started, but yes two years before the RISC-V Foundation was created and the then current ISA draft was widely publicised.

As far as I understand, at that point your only freedom was to license IBMs core or chip designs.

IBM released their first open source POWER core, the "Microwatt", in August 2019, a month after the base RISC-V 32/64 IMAFDC ISA was ratified (frozen forever and published)

The cores used in the highest performance RISC-V SoCs currently available were announced in October 2018 (SiFive U74) and July 2019 (THead C910).

I think OpenPOWER had some initial patent grant or license issues that wasn't ready until a few years after, after RISC-V came out. By then it was a little too late.

May be blame it on IBM again. Personally I quite like OpenPOWER.

OpenPOWER has no advantages over RISC-V. The performance is going to be the same given the same level of investment into the silicon design. These are just ISAs, and a lot more goes into CPU design to make them fast.
It's possible that OpenPOWER has no advantages today, but I remain curious how we got here; my understanding is that POWER and SPARC both had open source versions before RISC-V was created, so why didn't those get picked up but RISC-V did?
SPARC is a neat ISA, but the concept at the core: register windows, did not take off.

OpenPOWER: It may have to do with the size of the ISA when it started to be open.

RISC-V, when announced was a small ISA. That means making chips is easier.

It could just be the new hawtness effect. Look at how many people use NoSQL when SQL would work just fine ;).

> RISC-V, when announced was a small ISA

It still is today, comparatively.

RV32I/RV64I are very very small ISAs. RV64G is pretty small -- comparable to, say, MC68010, and a lot simpler than i386.

These small subsets will be supported by the ecosystem forever, for those who want to use them. They will also remain, unchanged, at the heart of more complex standards such as RVA23 and successors, and make up most of the instructions in programs.

RISC-V's success has been mostly in small embedded systems — where a POWER chip would be overkill or unsuitable. The core ISA is very small, thus cheap, and then extensible. The design even encourages proprietary extensions. Then it has grown from there as ISA extensions have been ratified.
Depends on your market. If you want a high end server/workstation class processor then openPOWER absolutely has an advantage because they have a shipping product that is (somewhat) competitive in that arena and despite the risc-v promises the products look like early 2010 era arm server machines. OpenPOWER have 30+ years of instruction set and architectural tuning to fall back on, and that is nothing to sneeze at if you actually want good vector perf, a stable toolchain, or a roadmap that doesn't depend on VC's staying in the game.

On the other hand if you looking for a small embedded one off solution then sure, risc-v. But unless your already sold on risc-v, there are a lot of other competitors in that space which have such a huge breadth of product offerings that it might be hard to justify a risc-v solution solely based on the processor architecture.

One very interesting project built on OpenPOWER, but with unique vector extensions, is LibreSOC, https://libre-soc.org/. It promises to be fully open hardware as well.
I have very low confidence that it will ever be anything but vapour, given the project leader's previous record e.g. this project fully funded in August 2016 and not yet delivered 7 1/2 years later.

https://www.crowdsupply.com/eoma68/micro-desktop

The project updates (last one in 2020) are painful to read.

A competent outfit such as Pine64 or Sipeed can knock a project like this out in 6 months. They could have been subcontracted given the $234k raised.

Before the micro-desktop, that person was also involved in some KDE/Plasma tablet, which also never shipped.
> is there something I'm missing?

Affordable, mature hardware. The Talos systems are starting at 3k for a quad core CPU on a micro ATX board. Same thing happened to MIPS and Sparc. Performance and technical merit mean nothing vs cheap and ubiquitous hardware. It's a lot of cash for a what amounts to an experimental toy. They are also a bit finicky as a friend bought one from another dev that refuses to post for unknown reasons. So there is risk involved too, no one else is making these boards.

The performance gaps and architectural features that made these chips matter 20 years ago have been closed by commodity off the shelf x86 hardware and various Arm CPU's are eating everything.

The only reason Risc-V matters is that no one has to pay for licenses.

> The only reason Risc-V matters is that no one has to pay for licenses.

Well it’s also a clean design that learns from the past 30 users and has got some traction with lota of material available and a rapidly growing ecosystem.

For the sake of my own curiosity, what is unclean about or in the POWER ISA?
I always assumed that Google was putting just enough money into OpenPOWER to keep the vendors interested, as a means of gaining leverage when negotiating with Intel.

I haven't kept up, but competition from AMD, ARM and RISC-V all probably fit that need now.

OpenPOWER happened a few years after Apple finished the transition to Intel.

The whole architecture is niche since then.

Linus Torvalds has a very interesting POV why x86 won. "Develop at home" issues. You are going to deploy to a system that is similar to what you built on. If you run x86 then you'll deploy to x86. And he points to that as the reason for x86 servers.

Home is today a x86 or arm computer (arm if you like Apple), perhaps some SBCs (usually arm, perhaps some mips), and some IOT (often esps, so xtensa / risc-v) plus some router/wifi device (arm).

RISC-V is scaling up on that axis. It is killing other ecosystems for embedded/iot. It's becoming useful for SBCs and low end desktop boards are on the horizon.

That's the scaling path that works. You are $20 away from trying it out. And it can scale all the way to an affordable desktop soon (Milk-V).

It's IMHO not "a lot of energy on RISC-V", it's a quickly growing user base. OpenPOWER lost that.

Today we deploy to the cloud and work on Arm/x86

Do most web developers do things that have issues being cross platform?

Even modern C and C++ code generally ports just fine as long as you avoid things with unspecified behavior like exotic casts and weird pointer tricks. Have to watch endianness too but that doesn’t actually come up all that often. Also AFAIK there are no longer any big-endian architectures in common use.

Newer languages are even easier. Rust and Go almost always port with zero issues. Obviously the same goes for scripting and VM based languages like Java.

The architecture matters less than it used to.

If you’re higher up the stack, sure. If you want performance, you want to keep the dev/prod environment alike so that you can properly optimize
Sure if you are counting cache misses or optimizing for a specific compute unit. Few people do that outside of AI, crypto, AV codecs, simulation, etc. If you are doing that you may need to specialize per arch anyway because you’re probably using vector units.

I was speaking to whether the code will work. The answer is almost always yes. Optimizations at a higher level such as algorithm choice or where allocations are performed also are architecture neutral. This is most optimization.

s390x (IBM Z) is a big endian holdout, with IBM in the wings still quietly paying for s390x ports of various open source software.
Are you aware that x86 and ARM/POWER/RISCV memory consistency model are really different? You can encounter very sneaky multitreading bug when running on ARM/POWER/RISCV a program that you have only tested on x86.

Apple has actually put a lot of effort to make the x86 to ARM transition as smooth as possible regarding memory consistency model, this is a strong indication that it's not as trivial as you seem to think.

You have to be doing pretty strange things with threads, without using mutexes or atomic instructions, to have a problem with this.

RISC-V has a standardised (optional) TSO memory model mode, plus "fence.tso" instruction that works in normal mode, even on CPUs that predate it (it defaults to the stronger "fence rw,rw" in that case). Well, assuming the CPU core designer read the spec carefully and doesn't trap on unknown "fence" instructions (looking at you, THead...)

There are _so many_ deploy issues surfaced by eg developing on a mac and deploying to linux/aarch64 (or, in my case, developing on x86 linux). Anytime you add a dependency on a package with a native component you're likely to have something to fix.
This right here, it's just crazy. On one project we invested in a private cache of prebuilt/binary pip wheels, as much as it complicated the build and maintenance, it saved us so much time and effort in the long run, and was ultimately a productivity boost.

I really like it that the x86 monoculture is being disrupted, and enjoy the benefits of low-wattage high-performance systems, but we still have a long way to go.

For the vast, vast majority of cases a VM can solve the issue of developing for linux/arm on an arm Mac.

It doesn't help so much if your Mac is still Intel based, admittedly.

In the .NET world we're developing on Windows/x64 and deploying on Linux/ARM64 en masse at this point. I haven't personally seen an issue yet where I had to debug a Linux- or ARM64-specific issue. I think different language ecosystems have wildly differing levels of readiness for this.
The platform of course matters less and less as you move into higher-level stuff. The necessary abstractions have already been built for you.

Though even in .NET you can still get bitten by assumptions about endianness, for example.

(comment deleted)
x64 and ARM64 are both little endian on Linux and Windows; there are never endianness issues when moving from Windows/x64 to Linux/ARM64. You can freely assume little endian and that's fine in this combination. It's one of many benefits ARM64 has going for it that IBM POWER does not.
The commercial Linux world (see https://github.com/open-mpi/ompi/issues/4349) and other open source OSes (eg FreeBSD) seem to have lined up behind little-endian PowerPC. IBM still has a big-endian problem with AIX, IBM i, and Linux on Z.
(comment deleted)
(comment deleted)
(comment deleted)
I've encountered a couple of NPM packages with native components which didn't work on a M series MacBook. Languages that uses native libs as part of their own packages are still a risk to do multi arch development/deployment.
Unfortunately so, native modules for Node are starting to be written more and more in rust rollup and parcel are two examples. Unlike previous modules written in C++ if there's no prebuilt binary for your platform, they won't even try to build locally, just flat-out wont install despite building fine if you do it all manually.
$9 RISC-V Linux board: Milk-V Duo

https://www.youtube.com/watch?v=YqUtGk0DHbQ

And that's a genuine $9, as in you can buy 10 or 1000 if you want, unlike the "$5" Pi Zero which was one per order, plus shipping. (It has a genuine, higher, price recently, and the Pi Zero 2 is a great deal if you want a low end Arm board).

And yes, the Oasis looks very promising, hopefully this year.

I did, until RISC-V started emerging.
Nobody cares about it because unless you use IBM, or use it in some embedded use cases it’s irrelevant.

It’s irrelevant because while they would openly license it, it wasn’t open source as you think about it traditionally and would require you to cross license derived technology to the consortium, which was effectively Freescale and IBM if I recall correctly.

IIRC they have only recently open sourced and freely licensed the ISA. Even then, there isn’t much interest in it outside big iron or niche platforms.

The ones, who should care to avoid US backdoors, built their own CPU's instead. And then came RISC-V
I care deeply, and was one of the early adopters of the Talos. I do my graphics and design work on an iMac G5, still. Power is the hill I'm willing to die on, technologically speaking. The bi-endianness matters a lot to me. I prefer to do my development on BE, and I get hardware accelerated LE VMs for free on Power.

But the rest of this thread is correct: right now it is a huge cost sink. I recommend people buy Apple Silicon Macs for new hardware unless they really need the owner controlled firmware of a Talos. There's just no denying the M2/M3 spank the Power9 core in every bench, single and multi thread.

I'm eternally optimistic - I was told I was crazy in the dark days of P7 and P8, then the Talos came. Maybe the LibreSoC or PPC Notebook projects deliver? Maybe Talos 3 isn't stupid high cost? I hope, but I don't hold my breath.

I'd prefer to buy a theoretical open-source R8000-R12000 based board*, but the only 'open source MIPS' stuff I ever see are low end R4000 and R5000 microcontrollers.

* Yours sincerely one of the few people that isn't offended by delay slots.

edit: and yes, Loongson is on my radar, has been for a while, but again only ever see microcontrollers rather than general purpose computing motherboards.

another edit: looks like loongson-3 3A5000 and 7A2000 boards and mini-PCs are starting to become reasonably affordable, so my dream may be reality in the nearish future.