back to article 5% of the cloud now runs on Arm as chip designer plans 2023 IPO

Three major hyperscalers – AWS, Microsoft Azure and Google Cloud – deliver as much as five percent of their compute services using Arm CPUs. This revelation was disclosed on Tuesday amid SoftBank's Q3 2023 earnings call (SoftBank owns Arm). "In the space of cloud, initially, the share was zero percent. But we're working with …

  1. Randy Hudson

    The death of CISC?

    Many will say that the demise of CISC was pretty obvious 30 years ago, but what has taken so long? IMO, we have Apple to thank for speeding things up. The benefit of developers having the same architecture sitting in front of them can't be ignored.

    1. Aitor 1

      Re: The death of CISC?

      Software blocked its death.

      The platform does not matter. Software and the ecosystem is what matters. But now as software is running behind closed doors on walled gardens, they can use something they control and not pay royalties and huge margins to others.

      It is also terrible for the small players, as we are moving from open systems to everything proprietary, including hardware.

    2. Yet Another Anonymous coward Silver badge

      Re: The death of CISC?

      >Many will say that the demise of CISC was pretty obvious 30 years ago, but what has taken so long?

      Because since the Archimedes you couldn't, and still can't, buy a desktop ARM cpu that compares with Intel/AMD.

      So I can't develop software on it.

      So there is no market for a desktop ARM machine

      Because there is no software for it

      1. Flocke Kroes Silver badge

        Re: The death of CISC?

        The bottom end of X86 is fast enough for all but a small minority. Mid-range ARMs are more than fast enough for most people's use of a desktop and are fine for software development too. There is plenty of software available for common tasks. The mass market are staying with X86 for their legacy software and their phones for everything else. That is the barrier to sales of an ARM desktop (and people being able to earn a living from selling binaries compiled for one.)

        1. Yet Another Anonymous coward Silver badge

          Re: The death of CISC?

          >Mid-range ARMs are more than fast enough for most people's use of a desktop

          Yes but (at least until the Pi4) I couldn't go to Bob's Big Box Computer Store and buy a beige box with a latest and greatest 64bit ARM

          I can get a bunch of dev-kits with the right promise of building a product and after signing an infinite number of NDAs - but that's not exactly consumer friendly

      2. werdsmith Silver badge

        Re: The death of CISC?

        So there is no market for a desktop ARM machine

        What am I missing here? Apple had no problem shifting shitloads of Max Minis with M1/M2 ARM based architectures. And because they are shifting ARM machines, software vendors are shipping native versions of their software. The Rosetta thing helped,’of course, but most popular software had universal versions out or have betas out. Market drove it.

    3. Flocke Kroes Silver badge

      Re: The death of CISC?

      I have been programming ARM for a very long time. I have actually used a couple of Apple computers. One had a 6502 CPU and the other a 68008. I have never done any ARM targeted development on an Apple computer. It used to be test and debug on x86 then use a cross compiler. Later ARMs became fast enough to host a compiler directly so I could use x86 as a terminal and ssh into the ARM system. Lack of an ARM CPU on the host system was not a barrier before Apple eventually switched to ARM. When that happened it was a complete non-event for me programming ARM.

      The first barrier to ARM growing up was most people wanted to run legacy DOS/Windows binaries. Although that is still a significant requirement for some people a larger market opened up: people wanted to run software for which they had the source code. With the source code they could compile it for whatever architecture was most cost effective. The next barrier was boot code: ARM has a long history of requiring a boot loader created specially for every micro-controller and the board it is soldered onto. To some extent it is now possible to have a BIOS that will work with more than one specific variant of ARM chip and so have an ARM mother board that could in theory be upgraded by replacing the CPU.

      The thing getting to x86 now is the thing that made x86 dominant decades ago: price. Intel used to be the cheap and nasty CPU. There were much better architectures around decades ago, but you paid for that clean efficient design because it performed so much better than Intel. As a result, Intel got used whenever price mattered. Intel got the big (cheap) market. Their R&D costs got divided by a much bigger market so they had the budget to improve. The others then had their R&D costs divided by a shrinking market and their designs stagnated.

      Intel became an effective monopoly and priced to match. They created a vicious competitor for their own low-end chips: themselves. Any cheap CPU sold took away an opportunity to sell a high margin CPU. Any factory time spend making cheap CPUs was factory time not spent on making high margin CPUs. This left the bottom end of the market to others. It has taken decades, but the high volume of cheap ARM CPUs funded development of slightly better ARM CPUs. This repeated until ARMs became fast enough for more and more demanding tasks.

      The performance/£ and performance per Watt gained such an advantage over Intel that Intel's biggest customers decided to solve ARM's problems for themselves. Apple have solved ARM's problems for Apple. Google, AWS and some others took a more community based approach. Apple is happy to use BSD licensed software but much prefer to create their own than share their work as would be required if using GPL software. The others can work with or around GPL: If the software never leaves their servers they haven't distributed it so they can keep improvements to the source code to themselves.

      I would say that ARM getting to servers has been more to do with everyone but Apple deciding to work to some extent with each other.

    4. jotheberlock

      Re: The death of CISC?

      It did die, pretty much, a modern desktop x86 CPU is RISC-like internally (and despite being a bit ugly, x86 isn't as CISC as say VAX anyway - importantly, for example, you don't get multiple memory operands in an instruction, or some of the wacky stuff like being able to do *******< some pointer> in hardware, in one instruction, which plays hob with modern caches/memory .

      (On the other hand, fixed length 32 bit instruction sets /also/ didn't entirely stick around - hence ARM Thumb and MIPS16 - because once you get to a certain point instruction decode becomes much less of a bottleneck anyway)

  2. Old Hand

    Worth noting that arm originally stood for Acorn Risc Machine. and Risc stood for Reduced Instruction Set. But that is lost in history now . . . LvS

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like