- Intel's next-generation Meteor Lake chips, set to arrive in laptops in 2024, are an exciting development that will pave the way for CPUs across various segments.
- Meteor Lake will use Intel's new tile technology, with four tiles dedicated to specific functions: Compute tile with CPU cores, SoC tile with display connectivity and AI processor, GPU tile with graphics cores, and IO tile with important connections.
- Intel's process strategy aims to catch up with TSMC by releasing Intel 4 (twice the density of Intel 7) and Intel 3, followed by the launch of 20A in 2025, which the company believes will bring victory after a challenging decade.
At this year's Intel Innovation event, Intel released some key (though not completely unexpected) details about its next-generation Meteor Lake chips, which are coming to next-generation laptops in 2024. While Intel isn't bringing these to the desktop, this is one of the most exciting generations of Intel hardware in recent years. In fact, Meteor Lake is the pipe cleaner that will make the way for Intel's CPUs across several segments, from laptops to desktop to the datacenter to AI.
Meteor Lake specifications and Intel's new tile system for building CPUs
We've already known some key info about Meteor Lake for a while now: it will be built using Intel's brand-new tile technology, which is very similar to AMD's chiplets. However, the key difference between Intel's tiles and AMD's is that Intel is dedicating each tile with a specific function, and Meteor Lake will come with four tiles: a Compute tile with CPU cores; an SoC tile with display connectivity, wireless technology, and an AI processor called a Neural Processing Unit (or NPU); a GPU tile with graphics cores; and an IO tile that contains important connections for PCIe and Thunderbolt.
Although Intel probably has lots of different tiles in the works today, the Meteor Lake chip Intel has revealed so far will have a core configuration of 6-performance cores and 8-efficiency cores, two less performance cores than the top-end Raptor Lake chip the Core i9-13900K. However, since Meteor Lake is going to appear in laptops only, lacking two P-cores won't be such a big problem, since laptops have very limited power consumption anyway. Meteor Lake's Compute tile is built on the Intel 4 node, formerly known as Intel 7nm.
The P-cores and E-cores are also brand-new, with performance enhancements for multi-threading and increased bandwidth per core, in the case of P-cores. For E-cores, Intel is also delivering improvements in instructions per cycle (IPC), and they also come with an improved Thread Director technology. This should allows the processsor to make smarter decisions about which cores to use in specific situations, delivering better performance and power efficiency when needed.
We don't know too much about the Xe-LPG architecture that will power Meteor Lake's Intel Arc graphics chip, but Intel says it has twice the frequency than current Xe-LP integrated graphics, which could mean double the performance. Despite this massive clock speed boost, the graphics chip is more efficient than the one present in current Intel CPUs, and AI helped it get there. Intel also claims support for features like ray tracing and variable rate shading. The display engine also has native support for HDMI 2.1 and DisplayPort 2.1, and the media engine is now separated form the graphics die, along with adding support for things like AV1 encoding.
On the topic of AI, Intel is also betting hard on Meteor Lake's capabilities. Both the CPU, GPU, and NPU in Meteor Lake processors can be used for AI workloads, excelling at different kinds of work. The GPU is ideal for AI workloads in media and 3D rendering, while the CPU is ideal for light AI tasks. The NPU is dedicated to AI, offering the best performance for sustained AI workloads and for offloading AI work.
The SoC and IO tiles are pretty standard and contain what you'd expect: connectivity for important technologies like USB, PCIe, Bluetooth, Wi-Fi, and so on, but there are some exciting points here. Meteor Lake will have Wi-Fi 7, Bluetooth 5.4, and Intel Unison screen-sharing technology via the SoC tile, which also has E-cores of its own, though Intel didn't say exactly how many. The IO tile just has PCIe, Thunderbolt 4, USB, and some other connectivity technologies. Combined, the SoC and IO tiles have very similar functionality to AMD's IO die.
While it's easy to call Intel's tiles just like AMD's chiplets, it's important to know that Intel's strategy here is pretty different. AMD only makes an IO die, a cache die, and a compute die with the CPU cores, and then uses multiple of those compute dies for increased performance. Intel on the other hand seems like it won't be using multiple compute dies with CPU cores except perhaps for server chips, like Sapphire Rapids. Both strategies have advantages and disadvantages, and Intel is banking on its design philosophy being more successful in the long run.
Intel's process strategy going forward: leadership by 2025
Intel's 10nm node (now rebranded to Intel 7) was an unmitigated disaster for the company, leading to several years of 14nm CPUs that slowly saw Intel lose its advantage. Intel is now well behind its chief node rival, TSMC, and Intel's got a plan to catch up and then return to first place. Step one is to get Intel 4 (which is what Meteor Lake's Compute tile is manufactured on) out the door, which has twice the density of Intel 7 (formerly 10nm and the process of Raptor Lake. Intel 4 will be followed by Intel 3, an improved version of Intel 4. These two nodes will be what Intel uses for its best CPUs throughout 2024.
2025 will see the launch of Intel's 20A process, and this is what Intel believes will deliver victory after a pretty rough decade for the company. 20A will be followed up by 18A, and it seems like an improved version of 20A much like how Intel 3 is an improved version of Intel 4. From previous presentations, Intel will launch upcoming desktop and laptop CPUs Arrow Lake and Lunar Lake on 20A and 18A, which were originally slated for 2024 and later but seem to be firmly in 2025 based on this latest info from Intel.
Of course, one might remember that Intel trying to make lots of progress within a short period of time is what caused the 10nm debacle in the first place, though this time Intel seems to be content to make progress across multiple nodes rather than one single node, which may prevent what happened with 10nm. We will have to see whether this pans out for Intel in the coming years.