I’m not going to dissect the cancelation theories since won’t really know until Intel confirms more. I’m more interested in the discourse about Meteor Lake’s performance, which has been cited as a potential reason for Intel’s cutting-edge chip to skip the desktop. Meteor Lake’s core count reduction probably isn’t a mistake, nor a consideration made solely for laptops. Rather, it plays into Intel’s strengths on both the desktop and in laptops.
Hybrid architecture and the problem with P-cores
Most Alder Lake and Raptor Lake CPUs have something called “hybrid architecture,” which is what Intel calls using two different kinds of cores in a single CPU. If you’ve ever heard of Arm’s big.LITTLE technology, then you’ll be familiar with this concept. Intel uses performance cores (P-cores) and efficiency cores (E-cores). Despite a few bumps in the road when Intel first launched Alder Lake back in 2021, this design has proven to be quite potent and has been instrumental in Intel’s comeback.
Alder Lake and Raptor Lake aren’t perfect, but it’s not because of E-cores, which are often ridiculed for being individually weak. In fact, E-cores are great and Raptor Lake’s liberal usage of them proves it. It’s actually the P-cores that have been the biggest problem for Alder Lake and Raptor Lake CPUs because they consume tons of power. In its review of the Core i9-12900K, Anandtech found that in a single-threaded workload, one P-core consumed 78W while one E-core consumed 15W, which means a P-core needs to be at least five times faster to meet the efficiency of an E-core, and usually P-cores fall well below that target.
To make matters worse, P-cores also take up lots of space. A single Raptor Lake P-core is roughly the same size as three E-cores, meaning an all-P-core version of the Core i9-13900K would realistically only have 12 of them but also undoubtedly perform worse at the 13900K’s 253W TDP. It’s no wonder Intel wants to use E-cores when P-cores seem to only be useful for providing good single-thread performance in applications that don’t need tons of cores.
Efficiency gains are performance gains
Power consumption is definitely the biggest weakness of Alder Lake and Raptor Lake. It’s why P-cores are doled out in smaller quantities than E-cores and hybrid chips made specifically for laptops cap out at six P-cores rather than the eight we see on desktop models. Meteor Lake is certainly an attempt to address and fix those problems, but removing two P-cores theoretically won’t do any favors for Meteor Lake’s performance.
The thing is, two P-cores probably won’t make or break Meteor Lake performance. With the 13900K, Intel has basically reached the limit of how much power a mainstream CPU can consume. The 253W peak is already quite a high TDP, but even at stock settings, a 13900K can boost well beyond 300W. Intel is basically power limited at this point and can’t improve performance without achieving higher efficiency. Obviously, P-cores aren’t as efficient as E-cores, so it makes a ton of sense to get rid of a couple, especially since it only impacts multi-core performance and won’t reduce single-core performance at all.
We don’t know how much more efficient Meteor Lake will be compared to Raptor Lake, but one rumor claims Intel is targeting a 50% or higher efficiency gain over Raptor Lake at the same core count. Since Meteor Lake’s highest-end chip doesn’t have as many cores as the 13900K, we know the rumor can’t be referring to the flagship models, but it’s hard to imagine the top-end Meteor Lake CPU not being more efficient than the 13900K. Even a 20% efficiency improvement would mean 20% more performance at the same power consumption.
Unless desktop Meteor Lake chips are limited to sub-200W TDPs (which would limit top-end performance), concerns about Meteor Lake’s top-end performance seem unfounded. It’s on Intel’s 7nm node (officially called Intel 4), has a new architecture, and uses the new tile design. A 50% improvement in efficiency is well within reason, and it’s what Intel needs the most right now since increasing power consumption doesn’t seem to be an option anymore. Shedding a couple of P-cores seems like the right move for both desktop and laptop segments of the market.
Core count is not Meteor Lake’s biggest weakness
If anything brings down Meteor Lake, it certainly won’t be its core count. The new process alone can either improve frequencies by 20% without increasing power or reduce power by 40% at the same clock speed compared to Intel’s 10nm node. That’s the best-case scenario, but since Meteor Lake also comes with architectural improvements, we can believe Intel won’t have too many problems with improving performance and efficiency on some of its best CPUs.
What Intel might have a problem with is actually wrangling together all the parts of Meteor Lake, making it work, and getting it out on the market. The way Intel is going about chiplets (or tiles, as the company calls them) is deeply concerning. Where AMD develops a few different chips and uses many of them to target the performance it wants, Intel designs several different, specialized chips that all have different manufacturing considerations. For Intel, this means higher development costs, less flexibility in using its tiles, and, most crucially, an increased risk for delays. A single tile could hold up an entire segment if it’s not ready.
The 12900K and 13900K were great when they came out, but 10nm CPUs were delayed for years and years, and it was only Intel’s fourth try at 10nm that resulted in truly good CPUs. Imagine 12th Gen had launched in 2018 or 2019 instead of 2021; that’s what the 10nm delays cost Intel. It seems weird to worry about cores when Meteor Lake hasn’t even reached the finish line yet.
We’ll know soon if Intel’s 7nm process is going to start out as poorly as 10nm, and if it’s true that Intel has canned the desktop version of Meteor Lake, that’s a really bad sign. 10nm was exclusive to laptops for over three years because the 10nm node wasn’t ready for high-end CPUs with lots of cores and high power draw. It seems Intel’s 7nm is capable of making big CPUs, but if Meteor Lake chips are power-limited due to technical problems, then that’s a much bigger issue than a couple of missing cores.