[cinntaile]

They're preparing to IPO in 2023. It's a shame the Nvidia sale didn't go through, now they'll likely die a slow death because they don't have enough R&D resources to compete with the big boys who are just going to design their own CPUs instead of buying the design.

[swamp40]

Small ARM Cortex's have taken over the microcontroller market. Billions made each year from every large microcontroller manufacturer. They aren't going anywhere. It's hard to believe ARM wouldn't make enough from licensing fees and from each one sold to make an easy profit. Maybe the license doesn't account for individual sales, IDK?

New secure ARM Cortex's are launching everywhere too, so they are very much alive and well, for the next 20 years minimum.

[fundmondawyaya]

Small RV32/64 cores could supplant them.

For example, GigaDevices makes an STM32F1 clone which comes in two flavors with the same peripheral memory map: ARM Cortex-M3, or RV32IMAC.

[swamp40]

Lots of things could supplant them, but the base is too large now and it has grown to cover the entire ecosystem.

They spent the last 10 years beating out PICs, 8051s, etc. Because they were BETTER.

And now a million engineers now how to use them, and the tooling from 1000+ companies is all in place to support them.

They aren't going anywhere.

[fundmondawyaya]

I dunno if Cortex-M chips will be better for long, though. Some of the RISC-V specifications like interrupt controllers need a bit more time, but they have very good support in mainline compilers like the GNU toolchain.

My experience writing C for RISC-V MCUs is about the same as doing so for Cortex-M platforms. They're a bit low-volume/expensive atm, but the core is very appealing to chip makers because it has solid software support and no royalties.

I also wouldn't write ARM off, though. Their 'coming-soon' MCU neural accelerators could open a lot of doors if it works well on a low power budget.

[chmod600]

"They aren't going anywhere."

That seems to be said of a lot of things that do end up disappearing. Hard to tell whether there are real roots in the market or if it's going to evaporate like so many other things.

[humanwhosits]

Bill Of Materials.

ARM will have to lower prices, as if you can save even just 1 cent per-device by using RISC-V in your microcontroller..

[ksec]

>compete with the big boys

Which big boys are designing their own CPU instead of buying from ARM? Apart from Apple and arguably Qualcomm with Nuvia?

[mschuster91]

The "big boys" you mentioned are all holders of ARM Architectural Licenses - there are only 15 of them in total and they likely pay a shitload of money for the privilege [1].

And they will need to hold on to the ARM architecture and thus the license payments for quite the time... I bet Apple could go and develop a completely new ISA from scratch, but that would mean a third infrastructure switch in two decades on the PC and the first one on everything mobile - a lot of work for everyone involved with no real benefit, and especially after they spent multiple billions on getting their ARM stuff to the performance (in compute power and power efficiency) point they need.

Even the "biggest boys" aka nation states fail at making decent alternatives to the x86/64 and ARM duopoly - the Russian Elbrus and Indian SHAKTI CPUs are utter exotics and Chinese Loongson are a plain fork of MIPS (and again, completely exotic).

[1]: https://www.anandtech.com/show/7112/the-arm-diaries-part-1-h...

[Kon-Peki]

That's an interesting idea.

The Apple M1 lineup, for example. The M1 standard, pro, max, and ultra are not differentiated by their ARM cores other than the quantity of them. It's really the GPU cores and media encoding components. And that Neural Engine thing is really important for many apps and OS features. At what point do the general purpose CPU cores on a modern chip become minor players that have very little impact on overall chip performance? I don't think we're there yet, but it certainly could happen.

EDIT - I'm saying M1 vs M1 Pro vs M1 Max vs M1 Ultra. Not M1 vs Intel i7

[yywwbbn]

That not really true though. M1 had considerable better single core performance at the time of release than anything else with a comparable TDP. And the Neural Engine is only useful is fairly narrow use cases.

[mkr-hn]

I wonder if there were similar conversations about math coprocessors when they joined the CPU. What did people say about increasing specialization of cores then? The closest recent analogue is GPUs joining the CPU on mobile devices, but it was only recently that those weren't complete garbage.

[kcb]

Apples ARM cores have been significantly ahead of off the shelf ARM designs for years now.