Arm aims to challenge x86 dominance with faster CPUs and AI-enhanced GPUs

[zohaibahd]

The big picture: When you think of Arm's processor designs, power efficiency is likely one of the first attributes that come to mind. The British chip firm's RISC architecture has long been praised for its efficiency-focused approach, especially when compared to x86 chips from AMD and Intel. However, according to Arm's client business lead, Chris Bergey, that narrative is expected to shift a bit this year.

In a sit-down with PCWorld at CES, Bergey revealed that after achieving a lead in IPC (instructions per clock), Arm is now aggressively targeting higher operating frequencies. While some Arm products excel in IPC, their lower clock speeds compared to rivals have held back overall performance.

Another major priority is optimizing Arm's CPU and GPU designs for AI acceleration. For CPUs, this involves adding new instruction set capabilities beyond Neon, Scalable Vector Extensions (SVEs), and SVE2 to better handle AI workloads.

Arm's GPU business is also being enhanced with an AI/ML focus. Bergey highlighted techniques like rendering at lower resolutions and then using AI to upscale to higher quality, similar to Nvidia's DLSS. This approach allows for better visuals with less GPU power required.

He provided an example of a mobile device rendering natively at 1080p 60Hz or using AI to interpolate from a 540p 30Hz rendering to improve efficiency. He added that the company aims to "be a leader" in enabling total processing on GPUs within a mobile environment.

These AI-accelerated CPU and GPU advancements could be integrated into Arm's next-generation "Arm CSS for Client" compute platform, set to debut in 2025.

However, pursuing higher operating frequencies and AI acceleration are not Arm's only ambitions. The company is also focusing on optimizing CPU designs to achieve maximum clock speeds on the latest semiconductor process nodes.

As Bergey explained, Arm plans to provide "that recipe for some of the latest [manufacturing] nodes" to help vendors "maximize that frequency and get to a four-gigahertz design."

Regarding the ongoing legal dispute between Arm and Qualcomm, the saga appears far from resolved. Bergey noted that despite a recent court ruling, significant issues remain "unresolved" between the two companies. The future of the case remains uncertain and will require further negotiation.

Permalink to story:

Arm aims to challenge x86 dominance with faster CPUs and AI-enhanced GPUs

 

[yRaz]

ARMs dreams of overtaking X86 is going to disappear if they don't stop asking for unreasonable license fees. RISC-V is going to get popular really quickly if the cost of developing a compatibility layer is cheaper than on-going license fees and litigation costs.

 

[Kashim]

ARMs dreams of overtaking X86 is going to disappear if they don't stop asking for unreasonable license fees. RISC-V is going to get popular really quickly if the cost of developing a compatibility layer is cheaper than on-going license fees and litigation costs.

ARM is not a threat to x86 in the short term, but the long term might be a different story. RISC-V is a threat to no one. Not now, not any time in the foreseeable future. RISC-V is as much of a threat to x86 as Linux is a threat to Windows desktop; not at all. Linux is free if your time is worthless and the same goes for RISC-V, yeah it's license free, but you get what you pay for (trash).

 

[human7]

ARMs dreams of overtaking X86 is going to disappear if they don't stop asking for unreasonable license fees. RISC-V is going to get popular really quickly if the cost of developing a compatibility layer is cheaper than on-going license fees and litigation costs.

Perhaps, though in AWS, the latest Graviton (arm) instances are the cheapest and actually quite fast. I benchmarked all the latest instances for work and, except for a couple matrix factorization workloads, was impressed by the results.

In the broader ecosystem, I'm a little skeptical that Arm's IPC claim is going to make the gigahertz factor enough to compete with x86. As a Risc platform, an instruction would typically do less work than a CISC instruction might accomplish, but ARM isn't strictly RISC anymore, either.

There's a lot of hype around RISC-V, but I feel like the champions who say "year of RISC-V" are the same type of people who say "year of the Linux desktop". It's taken a long time for AMD to steal significant market share from Intel, and that's all within the same architecture. Arm has a significant leg up over RISC-V not only in terms of available hardware, but on the software side. I expect it will take many years for RISC-V to be something that businesses and hyperscalers would somewhat routinely have as a legitimate option.

I'm hopeful that RISC-V will shake up the market some, I just don't see it happening outside of places like China, where there are geopolitical factors that would push for RISC-Vs adoption outside of just economic ones.

 

[bviktor]

ARMs dreams of overtaking X86 is going to disappear if they don't stop asking for unreasonable license fees. RISC-V is going to get popular really quickly if the cost of developing a compatibility layer is cheaper than on-going license fees and litigation costs.

It’s not an “if”. The damage is already done. And so is ARM.

 

[bviktor]

Perhaps, though in AWS, the latest Graviton (arm) instances are the cheapest and actually quite fast. I benchmarked all the latest instances for work and, except for a couple matrix factorization workloads, was impressed by the results.

In the broader ecosystem, I'm a little skeptical that Arm's IPC claim is going to make the gigahertz factor enough to compete with x86. As a Risc platform, an instruction would typically do less work than a CISC instruction might accomplish, but ARM isn't strictly RISC anymore, either.

There's a lot of hype around RISC-V, but I feel like the champions who say "year of RISC-V" are the same type of people who say "year of the Linux desktop". It's taken a long time for AMD to steal significant market share from Intel, and that's all within the same architecture. Arm has a significant leg up over RISC-V not only in terms of available hardware, but on the software side. I expect it will take many years for RISC-V to be something that businesses and hyperscalers would somewhat routinely have as a legitimate option.

I'm hopeful that RISC-V will shake up the market some, I just don't see it happening outside of places like China, where there are geopolitical factors that would push for RISC-Vs adoption outside of just economic ones.

Tell that to Qualcomm, that American company

 

[techfactor]

ARM is not a threat to x86 in the short term

Meanwhile Apple M4 Max has the best single core performance in THE WORLD period and best multi core performance in consumer CPUs only server CPUs beat it while being 2-3x more efficient than garbage x86.

 

[Axeia]

Meanwhile Apple M4 Max has the best single core performance in THE WORLD period and best multi core performance in consumer CPUs only server CPUs beat it while being 2-3x more efficient than garbage x86.

And how much of that is thanks to ARM vs having a node advantage and sticking everything on one chip?
Spoiler: surprisingly little.

Also the Ryzen 9 9950X definitely has better multi core performance and isn't far behind in single core (if it had node parity it might even surpass it).

The M4 is definitely impressive, especially for its power budget and when its accelerators can fit into your workflow but it's not miles ahead

 

[HardReset]

Meanwhile Apple M4 Max has the best single core performance in THE WORLD period and best multi core performance in consumer CPUs only server CPUs beat it while being 2-3x more efficient than garbage x86.

What Apple fanboys always seem to forget is that Apple chips have no expandable memory. That makes comparson invalid

And like said countless times: ARM is NOT more efficient than x86.

 

[yRaz]

ARM is not a threat to x86 in the short term, but the long term might be a different story. RISC-V is a threat to no one. Not now, not any time in the foreseeable future. RISC-V is as much of a threat to x86 as Linux is a threat to Windows desktop; not at all. Linux is free if your time is worthless and the same goes for RISC-V, yeah it's license free, but you get what you pay for (trash).

Perhaps, though in AWS, the latest Graviton (arm) instances are the cheapest and actually quite fast. I benchmarked all the latest instances for work and, except for a couple matrix factorization workloads, was impressed by the results.

In the broader ecosystem, I'm a little skeptical that Arm's IPC claim is going to make the gigahertz factor enough to compete with x86. As a Risc platform, an instruction would typically do less work than a CISC instruction might accomplish, but ARM isn't strictly RISC anymore, either.

There's a lot of hype around RISC-V, but I feel like the champions who say "year of RISC-V" are the same type of people who say "year of the Linux desktop". It's taken a long time for AMD to steal significant market share from Intel, and that's all within the same architecture. Arm has a significant leg up over RISC-V not only in terms of available hardware, but on the software side. I expect it will take many years for RISC-V to be something that businesses and hyperscalers would somewhat routinely have as a legitimate option.

I'm hopeful that RISC-V will shake up the market some, I just don't see it happening outside of places like China, where there are geopolitical factors that would push for RISC-Vs adoption outside of just economic ones.

you ARM guys are too easy.

 

[human7]

you ARM guys are too easy.

How do you mean?

Tell that to Qualcomm, that American company

They're adopting RISC-V for one platform, probably as a pilot, but I wonder how much of that was a stunt due to their ongoing litigation with ARM.

I'm all for the ISA to be a new competitor on the block, but that requires a lot of good software support in addition to good hardware. Apple showed the world how to have a successful ISA transition, where as Microsoft showed the world how to fail at it. To be fair, Microsoft has a bigger task to pull off than Apple. They learned from that mistake and have released Prism, but it remains to be seen how effective it is in general, though, or if it will matter for the gaming community. In any case, it's one thing for a couple niche platforms to switch to a different ISA, it's another to move industries and laptops to it.

 

[yRaz]

How do you mean?

The simple mention of RISC-V overtaking ARM is like putting a drop of blood in a tank full of piranhas.

While I am serious about how ARMs aggressive stance towards the industry, I don't think RISC-V is the option. All they're going to do is hinder their own progress. Shooting themselves in the foot, so to speaker.

But the simple mention of RISC-V is some of the most potent troll bait for ARM guys.

 

[GoldenGoat]

ARM is not a threat to x86 in the short term, but the long term might be a different story. RISC-V is a threat to no one. Not now, not any time in the foreseeable future. RISC-V is as much of a threat to x86 as Linux is a threat to Windows desktop; not at all. Linux is free if your time is worthless and the same goes for RISC-V, yeah it's license free, but you get what you pay for (trash).

Linux dominates servers, cloud, and has half of smart phones. Even Microsoft Azure is running on Linux.

 

[M25M25]

ARM is not a threat to x86 in the short term, but the long term might be a different story. RISC-V is a threat to no one. Not now, not any time in the foreseeable future. RISC-V is as much of a threat to x86 as Linux is a threat to Windows desktop; not at all. Linux is free if your time is worthless and the same goes for RISC-V, yeah it's license free, but you get what you pay for (trash).

Totally my vision as well