As Arm Rivals Cook Up Custom Silicon, Mediatek Sticks To Tried-and-true Cortex Recipe

Interview Arm Holdings has long been the primary architecture for mobile chips since the advent of modern smartphones – its Cortex is quietly humming away inside almost every phone or tablet you can think of. However, with Apple and Qualcomm producing their own custom silicon designs, Arm's market dominance appears less secure.

Arm cooking up powerful Cortex-X CPU to beat iPhone performance, says industry watcher

READ MORE

Apple is currently using its optimized M3 chips, having moved away from the traditional licensing model to develop proprietary designs. The company is also planning to include its next-generation M4 chip, already in the iPad Pro, in its upcoming Mac, MacBook Pro, and Mac Mini. Similarly, Qualcomm has begun shipping its Nuvia-powered Snapdragon processors featuring Oryon cores. These are not minor modifications of Arm's Cortex designs; through its purchase of Nuvia, Qualcomm started developing custom cores aimed at devices requiring advanced AI and machine learning capabilities. This development challenges Arm's position in the premium mobile chip market as standard Armv9 cores face competition from custom silicon solutions.

However, it is premature to discount Arm's influence. MediaTek recently announced it is continuing with the Armv9 architecture for its latest Dimensity 9400 chips, which demonstrates that Arm's Cortex-X4 and A720 cores remain competitive. The M5 chip is also under development, focusing further on AI and machine learning features, although Apple has yet to disclose specific details about its next-generation tech.

Apple's approach to custom silicon differs from that of Arm, which is providing a large ecosystem with the Instruction Set Architecture (ISA). Apple, by contrast, is controlling both the hardware and the software ecosystem within – with its classic iEverything approach. Although it does the fundamental silicon designs in-house, as we've seen from the last three generations of M-series chip, Apple outsources the fabrication to TSMC.

In 2021, Qualcomm acquired Nuvia, a company specializing in custom silicon founded by former Apple engineers. The objective was to develop custom Snapdragon processors with reduced dependence on Arm's ISA. The Oryon CPU cores within Qualcomm's Snapdragon X Elite are a result of buying Nuvia; they were initially referred to as "Phoenix" before being renamed to Oryon. These cores are based on Arm's v8.7-A ISA, similar to previous designs before Nuvia's integration into Qualcomm. While Arm's ISA remains a component, these cores exemplify custom designs, and Qualcomm intends to use them as part of its strategy.

The reason? To build its own customised Snapdragon cores and start weaning itself off its reliance on Arm's pre-built Cortex designs, although there is still Arm ISA within the overall architectural design. This move gives Qualcomm the chance to develop cores that specifically match what it needs. The Oryon cores that came out of this deal are Qualcomm's way into the custom silicon arena with the likes of Apple, and it is being positioned to go head-to-head with Apple's M-series, as well as Intel and AMD in the notebook space.

While all this may sound like bad news for Arm, the company remains integral to the industry. Arm's ISA continues to be the foundation of Qualcomm's current chips, ensuring its relevance as an intellectual property provider.

Competition is intensifying, particularly with MediaTek's recent announcement of its fourth-generation Dimensity 9400 series. The Cortex-X4 and A720 cores are central to its design, and MediaTek heavily relies on standard Arm cores to meet performance requirements. The Dimensity 9400, built using TSMC's 3nm process, is engineered to handle all the demands of modern flagship devices, including AI and machine learning tasks.

AI – how important is it?

AI might be the big buzzword of the moment, but Arm is clear that it isn't chasing after dedicated AI silicon such as an NPU, unlike Apple or Qualcomm… or everyone else really, for that matter. Instead, Arm's strategy is to make sure its underlying CPU cores with the Armv9 architecture at the heart of the SoC design can handle AI workloads themselves, leaving it up to its chip customers to decide if they want to strap an NPU in or not.

Arm is so, so over this AI accelerator unit craze

READ MORE

Arm's big smoking gun here – and something that doesn't fall prey to the pitfalls of exclusivity when it comes to having a custom NPU – is Armv9, which is geared towards the wider ecosystem for developers. Arm's Kleidi AI library, which comes as part of the Armv9 architecture, ensures that tasks get spread across the cores, with Arm making the CPU cores do the heavy lifting, at least 80 to 90 percent or around there, we've been told.

But let's not pretend everything is just about AI, even though the massive hype machine might want you to believe it is. Another key element within Armv9 is Scalable Vector Extension 2, or SVE2 for short. This isn't just about boosting, but it's designed to provide a solid boost to workloads and tasks, data processing, and all the usual high-performance grinds using CPU-level acceleration, but without the need for specialized accelerators such as an NPU. Arm prefers the holistic approach when it comes to optimizations, as the software developers are optimizing and writing code for a broad ecosystem and not just one specific accelerator.

Given these developments, we had the opportunity to speak with Christopher Bergey, senior vice president and general manager of the Client Line of Business at Arm. We discussed topics such as Arm's perspective on its position relative to competitors and the innovations required to stay abreast of the emerging custom silicon landscape. We also asked about how the next generation of Arm ISA (Armv10) is shaping up and Arm's future direction in both the mobile SoC and PC markets.

***

AI and machine learning are dominating chip designs at the moment, but what's Arm actually doing to stay competitive in that space?

Well, you're totally right that AI workloads are front and center, and will likely stay that way for years. But the way we think about AI – and I think this is the right approach – is that AI will run everywhere. So, when you consider where AI should run, the CPU plays a huge role. We've said for years that about 70 percent of third-party AI applications run only on the CPU because it's so widely used. Developers are targeting multiple platforms, which is why we focus on making AI run efficiently on our CPUs. That's where Kleidi and the v9 architecture with SVE2 come into play.

As the v9 architecture evolves – whether in MediaTek products or other phones that use v9 – a lot of AI is running on the CPU. The GPU also plays a significant role, especially in camera pipelines and advanced effects, so there's quite a bit of GPU computing happening too. And with NPUs (Neural Processing Units) now coming into the mix, we think of it as heterogeneous computing. AI isn't just about NPUs; it's about creating a balanced system. Many devices, especially in the broader Android ecosystem, don't have NPUs, but AI still runs well on Arm platforms. I think we're in the early days of AI, and we're just starting to see some really interesting use cases emerge.

We've seen a bit of buzz around "agentic AI," and MediaTek has been using it in its marketing for the Dimensity 9400. What's your take on that?

Agentic AI, in my view, is a class of applications where AI acts as a smart assistant or agent that adds intelligence to the platform. It's closely tied to hardware because we want to accelerate that experience. Depending on the workload – whether it's a long-running background task or something requiring a quick response – different parts of the hardware (NPU, GPU, or CPU) will be used.

There are still some growing pains with AI stacks – like where they should run and how portable they are. NPUs are useful, but sometimes developers choose to use GPUs or CPUs for certain tasks. OpenCL programming for GPUs, for example, is more widely used. We estimate there are over 15 million developers on Arm, and what they really want is flexibility. They want a broad ecosystem so they can make choices, and that's what we're enabling with things like Kleidi. It allows developers to write code once and take advantage of the hardware automatically.

We're also working to take AI to the next level with intelligent schedulers that better utilize NPUs, CPUs, and GPUs. The v9 architecture includes an AI-ready instruction set, and we're making sure the hardware keeps evolving... with technologies like SVE2 already bringing AI functions to the CPU. And there's more coming in the future.

With Armv9 Gen4 in full swing, what's Armv10 going to offer to keep mobile SoC flagships moving forward? Can you share anything about Armv10?

We haven't said anything about v10 yet, so I can't comment on that specifically. But we're committed to Arm being the most ubiquitous architecture. Over the past 30 years, we've consistently innovated and anticipated future computing challenges, such as power efficiency, which is crucial in everything from cars to datacenters. Power efficiency is always top of mind, and that's where Arm's strengths lie. We'll continue evolving the Arm architecture, and there's more to come.

• Intel, AMD team with tech titans for x86 ISA overhaul
• AMD pumps Epyc core count to 192, clocks up to 5 GHz with Turin debut
• MediaTek enters the 4th Dimensity with 3nm octa-core 9400 smartphone brains
• Fujitsu teams up with Supermicro on Arm-based server CPU

There's been some news that broke recently about Intel and AMD forming an x86 advisory group. Reading between the lines, it seems like this is aimed at the Windows on Arm push, driven by Microsoft and Arm. Can you comment?

Honestly, I haven't had a chance to read about that yet, but it does sound interesting. I'll have to take a look.

It seems like Microsoft's push for Windows on Arm devices is a key focus for them right now.

What Microsoft wants is diversity of choice. For a while, AMD has been strong in certain areas, but leveraging Arm's mobile ecosystem – along with partners like Nvidia, even in datacenters – gives Microsoft more options. If you're a company like Microsoft, you want a strong platform, and x86 has served them well. But Arm is also a very important platform, and they want to make sure they're not solely reliant on legacy systems. That's why you see such a strong commitment from them.

I agree! From our past talks, you know I'm certainly interested in this.

Absolutely. We've been working with Microsoft for a long time, and they've been collaborating with our ecosystem partners too. It's a smart move. Arm's mobile ecosystem, with its focus on things like battery life, thin-and-light devices, and great camera experiences, has become central to smartphones. It's interesting to see x86 now aiming to compete with Arm in areas that have traditionally been Arm's strengths.

At the end of the day, this competition benefits consumers. They get better products that focus on real-world use cases, not just benchmarks. And it's a huge opportunity for Arm to expand into markets where we haven't historically been as strong. So, we're excited.

***

As the industry stands, Arm faces significant competition – not only from Apple and Qualcomm but also from Intel and AMD, which have recently joined forces to create their x86 Ecosystem Advisory Group. This alliance between two x86 giants suggests the competitive landscape is intensifying. While AMD and Intel appear to be uniting to defend the x86 architecture, the momentum of Windows on Arm, coupled with advancements in Arm's ISA and Qualcomm's designs, is making an impact. Intel and AMD seem poised to respond vigorously, especially since the PC market has long been their stronghold.

Arm isn't just dodging punches from Apple and Qualcomm anymore, now they have Intel and AMD to contend with in a move that looks suspiciously like the old guard closing rank. Officially, it's all about improving the x86 platform, but the timing of this alliance comes right as Windows on Arm starts making real waves, and Qualcomm's Snapdragon X Elite has made the headlines with its Arm-powered laptops. Intel and AMD have had the PC market on lock for decades, so seeing Arm muscle in with Microsoft's backing clearly has them rattled. For Arm, it's not just about outgunning Apple and Qualcomm's custom silicon anymore — it's about holding ground as the x86 titans scramble to protect their territory.

Although Arm could not comment directly on Qualcomm and Nuvia's designs, it remains focused developing ISAs with an emphasis on power efficiency – a critical factor for smaller devices like smartphones and ultra-thin laptops. Notably, Arm does not incorporate a traditional Neural Processing Unit (NPU); instead, it relies on its cores to handle AI workloads. To ensure accessibility for developers, the Kleidi AI library is integrated into all of its Armv9 IP and designs. While NPUs have their advantages, they require additional frameworks and coding to manage workloads, and they are not the sole components capable of performing AI tasks.

For Arm, the markets are rapidly evolving from all directions, and the rise of custom silicon is increasing the pressure on its Cortex cores to deliver better performance. MediaTek continues to support Arm with its Dimensity 9400 series using the Armv9 architecture – and with more designs expected from other manufacturers in the coming months – it is evident that the real threat to Arm's business comes from custom-built designs like Apple's M-series and Qualcomm's Oryon cores.

Although Arm declined to discuss its next generation, Armv10, it will undoubtedly need to improve all metrics. One area where Arm's ISA particularly outperforms Intel or AMD's x86 architecture is power efficiency, delivering significant performance within a small power envelope. The future of Windows on Arm remains unclear, especially as Microsoft is heavily promoting it.

A spokesperson from Arm told The Register, "it's interesting that the x86 ecosystem is just now looking closely at ensuring software compatibility. For more than 30 years, collaboration and standardizing across a common and consistent platform has been the foundation of the Arm ecosystem, and today we have 20 million software developers taking advantage of that consistency and compatibility."

Regardless of how the industry evolves – whether it involves Arm ISA versus custom designs or x86 versus x64 in the PC space – the coming years promise to be interesting to watch. It will be worth observing how developments unfold with the introduction of more custom designs, Arm's next-generation ISA, and the impact of the unprecedented alliance between AMD and Intel in defending x86. This is especially significant as AI increasingly becomes the new normal in silicon designs. ®

Editor's note: This story was amended post-publication with comment from Arm.