Intel's upcoming Meteor Lake chips represent a pretty decent boost from the company's last-generation Raptor Lake processors: a 20% boost to efficiency (which is 20% higher performance at the same power), faster graphics, and a built-in AI processor. Those are just the major things Meteor Lake brings to the table, and there are lots of smaller things that, altogether, help Meteor Lake level up even more. However, Meteor Lake's impact as a product is undermined by the fact that it's skipping the desktop, making it clear that it's not good enough to replace Raptor Lake in pricing and/or performance.
Regardless, Meteor Lake is extremely important to Intel's future thanks to all the technology it introduces to the consumer market. Meteor Lake could be Intel's own AMD Zen moment and a chance to reverse the company's declining fortunes. If Intel wants to regain its titanic status, then it starts here with Meteor Lake.
The tile system is proving to be successful
The single thing that arguably lifted AMD from underdog to on par with Intel was the chiplet. In case you don't know, a chiplet is basically a piece of silicon that has only part of the functionality of a full processor. Think about how a Core i9-13900K has the CPU cores, integrated graphics, memory controllers, and a whole lot of other stuff on a single silicon chip. By contrast, the Ryzen 9 7950X has two chips with CPU cores (and only CPU cores) as well as a chip with memory controllers, I/O functions, and integrated graphics.
Intel is going into chiplets (which Intel calls tiles) with products like Meteor Lake, and while it's not the first Intel processor to use tiles or chiplets (Ponte Vecchio and technically Sapphire Rapids did it first), Meteor Lake marks the first time Intel CPUs with this technology will reach consumers. Tiles and chiplets have lots of benefits: more efficient production, the ability to make much larger and more powerful chips than normal, and customizability to name the primary ones.
But this isn't simply Intel catching up to AMD, as the company's strategy with tiles is a little different. Whereas AMD just makes CPU and I/O chiplets, and adds more CPU chiplets to increase performance, Intel is making a tile each for CPU cores, GPU cores, SoC functions, and I/O. One of the key advantages of this is that it allows Intel to squeeze as much vitality out of each tile until they absolutely need to be changed out. This means less money being spent on new designs and more money saved on production, since Intel could produce the same SoC and I/O tiles for a long time.
The time it takes to launch new products is also seemingly decreasing thanks to tiles. Take for instance Meteor Lake, which launches this December. Its successor, Arrow Lake, which will feature a new CPU tile and probably reuse all the other tiles, will launch in 2024 or in less than a year. Arrow Lake's new CPU tile isn't just a refresh of Meteor Lake's or an upgrade like Raptor Lake was to Alder Lake, but a brand-new CPU tile featuring the new 20A process, and perhaps better and more plentiful cores.
Tiles also offer much more specialization than AMD-style chiplets. AMD still relies on monolithic (non-chiplet) processors in the form of its classic APUs, but Intel is making the full transition to tiles. It's possible for Intel to do this since it's investing in making a wider variety of tiles, while AMD's chiplets are geared towards server chips, and although they work well for midrange and high-end desktops too, they're not very suitable for much else. Intel could even make a purpose-built chip out of tiles it already developed for use in other processors, at least in theory.
Intel's reinvigorated process strategy makes tiles even more powerful
But there's also another dimension to tiles: its synergy with Intel's process strategy. One of Intel's biggest failures within the past decade was its 10nm node (now the Intel 7 node), which Intel had intended to make a multi-generational leap with and ended up having to delay it so many times that Intel is now actually about a generation behind. Intel got way too cocky with 10nm, and the company has clearly learned its lesson.
That's not to say that Intel hasn't gotten less ambitious, however, as its new plan will theoretically put Intel back in the lead perhaps as early as the second half of 2024. That might sound exactly like what Intel did with 10nm, but the difference is that Intel is spreading out its progress throughout several nodes, which is a much more conventional strategy, though the speed at which Intel is finishing these nodes is unprecedented. The tile system not only reaps more reward from this revitalized process roadmap, but also helps mitigate any potential disasters.
The most obvious benefit with tiles is exactly what I already mentioned with Arrow Lake. Normally, launching so many nodes in such a small amount of time would be difficult to take advantage of with monolithic chips, but Intel only needs to design and manufacture a new CPU tile to take advantage of a new node to improve CPU performance. Not only is this cheaper, but development should also be less time-consuming, meaning Intel can launch new products more quickly than the company has been able to in recent years.
Historically, the introduction of new nodes meant older nodes were ultimately on the way out. Intel's fabs (short for fabrication plants, the facilities that make chips) would need to retool for newer nodes eventually, an expensive and time-consuming process. However, tiles give new life to older processes, which are suitable for SoC tiles, I/O tiles, and cache tiles since newer nodes don't bring much benefit to these use cases. Plus, Intel could reuse older CPU and GPU tiles for cases where more performance is unnecessary.
However, one caveat here is that the GPU, SoC, and I/O tiles in Meteor Lake are fabbed at TSMC, not Intel, undermining this potential advantage. In the future though, it's possible that Intel will introduce newer versions of these tiles manufactured by Intel fabs, which would then validate keeping older Intel nodes around. After all, Intel likely chose TSMC's 6nm for its SoC and I/O tiles because Intel's 14nm was too old and its 10nm was too expensive. Intel's GPUs however might be harder to switch to internal fabs.
Tiles can also help mitigate defects that occur within chip production. The productivity of new nodes can be seriously undermined by a large number of defects, but small chips can essentially contain defects and lower the amount of silicon they damage. Tiles can't make a broken process functional, but they can make a good but defect-prone node more productive, and this is something Intel has struggled with before. This could even help Intel launch cutting-edge nodes sooner than the company normally would be able to, and perhaps that's partly why Intel's process roadmap is so aggressive.
Even if Meteor Lake is underwhelming, its impact will be felt for generations
Even if Meteor Lake made it to the desktop, I wouldn't expect it to be a massive improvement over Raptor Lake. Intel's official efficiency improvement figures for Meteor Lake is 20%, which is going to be pretty significant for laptops but won't matter too much for desktop users, especially if that efficiency improvement doesn't mean an equally large performance bump. As a product, it's better, but it won't radically transform Intel the same way 12th-generation CPUs did. It's probable such a small performance bump combined with a likely higher price tag would have been poorly received.
Meteor Lake's single most important purpose has been getting tiles to the consumer segment, and it seems like it's mission accomplished for Intel. The SoC and I/O tiles should have lots of life left in them, as could the GPU tile. Fortunately for Intel, getting Meteor Lake out the door was the hardest and most expensive part, and now that it's clear that its tile-based chips can launch mostly on time and perform well across laptops and desktops, the path has become much easier for the company. Meteor Lake is really a pipe cleaner for far more impactful, tile-based CPUs that will launch in the coming years.