The end of a “myth”: Intel wants to prove that ARM is not superior to x86

Deal Score0
Deal Score0

It’s a cliché that comes up regularly: x86 processors would be threatened by the ARM architecture, “better and more efficient”. Used by Intel and AMD, the x86 instruction set is a priori at the heart of your processor if you're running Windows, and still powers a vast majority of data centers and supercomputers around the world.
But as many players – like Qualcomm in the general public, but also Ampere and Nvidia in data centers – begin to question the quasi-monopoly of Intel and AMD chips, it was time to look into this affirmation.

Apple M1 reveal, not the Snapdragon

Arik Gihon, in charge of the development of the Lunar Lake SoC at Intel

Arik Gihon, in charge of the development of the Lunar Lake SoC at Intel.

© Adrian Branco for Les Numériques

Advertising, your content continues below

Given the success of the Snapdragon at the Computex show in Taipei – more than 22 designs from the majority of manufacturers – some commentators claim that Lunar Lake is a reaction to this ARM chip which promises mountains and wonders. These comments seem to ignore the development time of the components. Arik Gihon, engineer at Intel in charge of SoC design, explains: “Discussions about Lunar Lake's design took place long before Qualcomm announced development of its chip. The real trigger was Apple's M1.

Structural details of the Apple M1 chip.

Structural details of the Apple M1 chip.

Launched at the end of 2020, Apple's in-house chip ended up completely ousting Intel from its computers. Based on the ARM instruction set, its microarchitecture was entirely developed in-house. Understand that instead of buying generic core plans designed by ARM for its customers, Apple acquired a license allowing it to dive into the very heart of the language to organize its chip as it wanted, in fact creating its own microarchitecture, which he then integrated into an SoC.

Advertising, your content continues below

Instruction set ≠ microarchitecture ≠ SoC ≠ burning node

Intel highlights a

Intel highlights a “major breakthrough” of its x86 microarchitecture in terms of performance/watt ratio.

©Intel

There is a lot of semantic confusion in the terms surrounding CPUs, particularly the often misused word architecture. To understand, we need to dive a little deeper into our Central Processing Units, the CPUs, also called microprocessors in French. Responsible for managing the complete system – unlike a GPU or an NPU for example, which are more intensive computing brutes or accelerators – the CPU has basic instructions as its fundamental building block, which are gathered in a sort of dictionary, called Instruction Set Architecture in English (ISA): the set of instructions.

These instructions are a bit of the basic vocabulary of the chip. But it is not because the words in a work by Camus or a political tract all come from the French dictionary that the two texts are similar! Just as these two examples are different in substance and form, two chips designed with the same instruction set have different mechanisms by design. Logical design on the one hand – the interactions between the different calculation units – and physical design on the other hand. This organization of instruction sets, how they are related, and how information flows within them is called microarchitecture. It is from this microarchitecture that many CPU characteristics arise: target frequencies (minimum and maximum), number of cores, etc.

Advertising, your content continues below

Only the elements denoted P-Core and E-Core are CPU cores, the entire rest of the chip is devoted to other elements.

Only the elements denoted P-Core and E-Core are CPU cores, the entire rest of the chip is devoted to other elements.

©Intel

This microarchitecture therefore does not represent the final component. Once this has been determined, you must size the CPU according to your needs. A CPU which today is only a small part of the chips: you only need to look at a chip diagram to realize that the CPU part we are talking about no longer occupies only a minority part of the processors in our computers . Processors whose real name is now system on a chip, for system on a chip (SoC). Mapping an SoC makes it possible to identify the CPU part(s), the GPU, now the NPU, the parts dedicated to data exchange in and out of the chip (cache memory, fabric, etc.).

And finally, a strong argument has been brandished in favor of ARM chips recently: the fineness of the engraving. When Apple launches its M1 at the end of 2020, in front of it, Intel has 11th generation mobile Core chips. But while the M1 is engraved in 5 nm, Intel's chips are engraved in 10 nm. However, the fineness of the engraving is the main tool for reducing energy consumption. Clearly, although x86 is constantly singled out for its lower energy efficiency, this is actually not the case. And it’s not just Intel that says it…

Advertising, your content continues below

When a competitor's brain validates Intel's words

Gérard Williams, designer of the CPU microarchitectures of the Apple M1 and Snapdragon X Elite

Gérard Williams, designer of the CPU microarchitectures of the Apple M1 and Snapdragon X Elite.

© Adrian Branco for Les Numériques

For Arik Gihon of Intel, “x86 can be as efficient as ARM if used well”. Words that could be interpreted as a simple defense on the part of an industrialist who has recently been technically overwhelmed. But apart from the fact that this message is sent not by yet another marketer, but by one of the stars of chip design at Intel, these words were validated last year… by the competition!

A competition embodied by Gérard Williams, father of the Apple M1 and Snapdragon X Elite CPUs. The chief engineer, whom we met last October at the Snapdragon Summit, actually contradicted us when we mentioned the possible superiority of ARM over x86. “Any ISA is good”, he assures. “Once again, it all depends on what your starting point is and, for x86, it's in watts not milliwatts.” It is this starting point for the design of microarchitectures that G. Williams had at the time identified as the Achilles heel in relation to ARM, and not its DNA.

After confronting him with these comments that we collected last year, Arik Gihon agrees: “That's perfectly correct. The ISA is absolutely not at stake, it is the microarchitecture, its implementation and the burning node”. Recognizing that “Apple has raised the bar in many areas, including energy”, Arik Gihon assures that “competition is good for Intel.” The first tests of Meteor Lake, planned for the end of the year, will therefore show whether Intel has been able to meet the challenge and prove that x86 can indeed be as efficient as ARM… To be continued.

Advertising, your content continues below

More Info

We will be happy to hear your thoughts

Leave a reply

Bonplans French
Logo