Most of the world’s smartphones are powered by processors based on designs from ARM, as are a growing number of other devices including servers, personal computers, IoT and smart home devices, and automotive systems.
While ARM releases new chip designs regularly, most modern ARM-based processors are based on ARMv8 architecture which was first introduced a decade ago.
Now ARM is introducing its successor. The company says the new ARMv9 architecture will enable next-gen chips with better security, enhanced artificial intelligence features, and CPU performance improvements.
ARM says it expects CPU performance to increase by 30 percent or more over the next two generations of mobile chip designs. But total performance gains may actually be higher if chip makers adopt multi-core designs and use specialized cores for graphics, AI, and other functions.
Speaking of AI, ARM says it’s partnered with Fujitsu to develop the SVE2 instruction set to bring improved processing power for virtual reality, augmented reality, 5G systems, and machine learning tasks.
For security, ARM says it’s introducing ARM Confidential Compute Architecture (CCA) to provide a hardware-based secure environment that allows applications to reside in a protected area of memory where they’re isolated from everything else. Among other things, this means that you could do secure banking without worrying that other apps on your phone would be able to access your financial information. ARM says the data can’t even be accessed by software with escalated privileges.
ARM says the new architecture backward-compatible with ARMv8, building upon the AArch64 instruction set. But the company says the new security and AI extensions should help the company design chips that will meet the needs of the coming decade with the goal of delivering “the next 300 billion ARM-based chips.”
Keep in mind that ARM designs chips, but the company doesn’t make them. The designs are licensed, and sometimes adapted, by companies like Qualcomm, Samsung, MediaTek, NXP, and Apple. We will most likely begin seeing new chips from those companies and others that are based on ARMv9 architecture starting early next year.
ARM also hinted at some other new technologies it’s working on, including next-gen Mali graphics processor designs with support for features including variable rate shading and ray tracing – something that so far has only been available to consumers with the purchase of high-end desktop or laptop GPUs from NVIDIA or AMD. It’s unclear when ARM’s new GPUs.
they would correct the ipc
What does that mean?
Are you disappointed with the Instructions Per Clock of the ARM X1 or Cortex A78? It sounds like it, and to be honest I’m a little disappointed too.
In any case, we saw a pretty huge upgrade to performance early on, as ARM’s little cores were catching up to x86 in many fronts. Then another substantial leap transitioning to their middle core designs (ie/ 2014’s 32bit Cortex-A17 on 24nm… to 2016’s 64bit Cortex-A73 on 16nm). And a modest leap to Cortex-A76 and now.
Let’s hope the first generation gets here in 2022 delivers as promised, and we get a substantial leap from the 6nm Cortex-A78… to the 4nm Cortex-A81, similar to the transition in 2016.
It was actually A15 and Krait (QC custom) to A57.
A73 was the 3rd gen ARMv8 core while A57 was the 1st.
A17 was more of a supplemental release at the end of the ARMv7 era that sort of matched A15 performance at a lower power draw, otherwise it added nothing and had no place in any flagship processors from the major named SoC designers like Samsung or Qualcomm.
Of course Qualcomm lacked the experience integrating ARM licensed cores as Samsung did when they designed the SD 810 with A57/A53 big Little clusters, so the release of A57 was a bit of a mess on Qualcomm’s side.
Samsung’s own Eyxnos A57 chip was fine having gained the experience with the A15/A7 chip beforehand.
One thing you are right about though is that the big cores of the next ‘Matterhorn’ CPU generation will come from the same design team at ARM Sophia Antipolis that created A17 and A73.
The Krait S3 was a very customized version of the Cortex-A9. The Krait S4 was a unique design on the evolution of the S3. The Krait S4.5 was a small refinement of the S4 (QSD 800 vs QSD 805), it was a very customized version of the Cortex-A15 but it wasn’t a very successful design.
The last, most advanced, and the best of the 32bit ARM cores was the Cortex-A17. Hence why I nominated it.
The Cortex-A57 was never a smartphone processor, as it was actually designed years prior, and was ARM Company’s first attempt to get into the lucrative Server Market. Obviously it sucked. They even advanced the idea of big.LITTLE for this reason, in order, to recycle this core design. The Kryo-100 and the Exynos 7420 was a more advanced and customized version of the Cortex-A57, to the point they were competing with the latest Cortex-A72. I also didn’t mention those cores for a reason, nor the Cortex-A72, which is a larger, wider, thirstier core than the A73.
The reason I specified the Cortex-A73 was because it is the smallest core ARM makes in 64bit that is Out-of-Order. You can find it in something like the QSD 632 and QSD 636.
The gist of my analogy is something along the lines of imagine you had to compare two phones, with everything being equal except the processor-software. So just think of the 2014 flagship phones, from smallest to largest: Sony Z3c, HTC M8, Samsung Note4. Think of them all stuck on Android 4.4, 32bit, with Cortex-A17 on 24nm. Now imagine if they had the exact variant but it ran Android 5.1, 64bit, with Cortex-A73 on 16nm.
At the time, in 2014, the difference might feel insignificant. But after a year or more, the difference and longevity would become very pronounced. That’s what I feel like is going to occur with anyone buying the lackluster QSD 888+ in 2021…. when ARMv9 designs might be hitting the market in 2022. The differences will feel small initially, but the software roadmap is going to become pronounced after a year or so.
Or to put it another way, so that spyware can harvest data from your device without software, even with elevated privileges, being able to see that it’s there.
If it is a hardware based setting that can be controlled locally by the user then I see that as a step forward… like pressing a button to allow software updates. Even then that would not stop a SolarWinds attack.