Intel vs. AMD, a war over mere nanometers

Intel vs. AMD, a war over mere nanometers

Intel and AMD are two perpetual enemies locked in a struggle for dominance in gaming and productivity and the past few years have seen some of the most even battles we've seen in over a decade. In 2020, AMD finally eclipsed AMD's single threaded performance, stealing the gaming crown from team blue for the first time in many generations. A major reason for that is the enhanced process node that its Ryzen CPUs are built on — 7nm, vs the 14nm node Intel is still using after many years and iterations.

Intel is slated to finally move to a smaller node for its desktop chips in 2021 with its Alder Lake chips, which will have an impact for sure, but is that the whole story? And is 7nm the only reason AMD has pulled ahead?

Let's take a look at the nanometer war that's been raging these past few years and what kind of impact its had on the state of CPUs today.

What are the nanometres a measurement of?

Although typically used to describe the size of the process node and architecture that a CPU is based on, the nanometre figure is actually the measurement of space between the transistors within a processor die. It's not entirely accurate either, with some wiggle room either way. Other factors can play into it too, with some suggestion that Intel's 10nm process is closer to TSMC's 7nm node (used in AMD chips), while Samsung's 8nm node is actually based on its 10nm process, but design changes make it perform closer to what it would if it were 8nm.

It's a messy way to measure things that makes it difficult to quantify exactly what effect it has on real world performance, which is what's ultimately most important.

Typically though, the smaller the process node the denser transistors can be packed onto a CPU die, ultimately leading to greater efficiency or performance, depending on how the manufacturer wants to leverage that additional power. Usually, smaller process nodes, and therefore denser and more numerous transistors, means greater performance, but that's not always so cut and dry.

Smaller is winning, but that's not always the case

Right now, Intel's Comet Lake CPUs are based on a 14nm process which has been in use since its 2015 Skylake generation. AMD is using an advanced 7nm node developed by TSMC and it's definitively faster in just about every way than Intel's best options, offering more cores and better performance in single threaded and multi-threaded tasks throughout the pricing spectrum.

This wasn't always the case though. AMD used a 12nm node with its Zen+ Ryzen 2000 generation, and that couldn't catch Intel in most tasks, despite being a more refined process. Its 7nm Zen 2 CPUs still fell behind in gaming performance, even though they were on a much smaller node. It was only with the architectural improvements of Zen 3 that it finally stole the gaming crown, and even then it's typically only a few percent ahead of Intel's best options.

Intel spent many years resting on its laurels with 14nm architectures but maintained its lead and has stayed competitive most recently, thanks to architectural improvements which kept raising clock speeds and improving thermal performance. Although it's now fallen behind, and looks set to sacrifice multithreaded performance in favor of regaining the gaming crown with the upcoming Rocket Lake CPUs, it has achieved the impossible with its aged 14nm process.

It really is time to move on from it though, which is why the upcoming 10nm Alder Lake is so important.

What's taken Intel so long?

Intel originally planned to release 10nm desktop CPUs in 2015. Then 2016, 2017, 2018, and every year since. The problem was with the 10nm process node its fabricators were trying to manufacture. Yield was a major issue, with Intel simply not able to fabricate enough 10nm chips on scale to make it financially viable. It started to release some 10nm chips en masse with Ice Lake in 2019, but performance was hardly stellar.

Upcoming Rocket Lake chips will use a middle ground, backporting a 10nm architecture at 14nm to offer greater performance and efficiency, despite maxing out at 8 cores. Late-2021 Alder Lake will finally move Intel's desktop CPUs to 10nm, moving on from 14nm and reiterated Skylake designs for the first time in half a decade. Whether Intel will be able to maintain volume production is yet to be seen, and even then, the timing may not be great.

Too little too late?

AMD is still working on getting its Zen 3 CPUs at volume to meet the extreme demand of the new-generation chips, so there isn't much discussion about its long-planned Zen 4 CPUs. When that starts to pick up speed though, expect to hear a lot about them, because they're going to be based on TSMC's next-generation 5nm process node.

They'll probably debut towards the end of 2021 too. Right when Intel is trying to get people excited about 10nm.

How the two chip lines compare will depend on architecture and clever technologies like AMD's Infinity Cache, as much as the process node, but it seems likely that AMD will maintain a performance lead with these new generation chips, with such a big node advantage.

If Intel has truly cracked 10nm though, expect it to jump down the node sizes much quicker than we've seen in recent years, though. AMD has held a process advantage for a number of years and likely will do for a while longer, but Intel is a sleeping giant of expertise and capital for R&D investment. It will make a comeback, and if it's anything like the early-2000s Conroe release, we could see something quite special from team blue in the near to medium future.

Image source: Fritzchens Fritz/Flickr