After months of rumors, leaks, and occasional backups from the company itself, Intel’s Meteor Lake architecture for laptops is now officialbecause the company presented it at the ongoing Intel Innovation, which we were invited to and which we will share more information about soon, since it is one of the most important technology meetings of the year.

As we expected, This year’s generational replacement will arrive in various flavors for desktops and laptops. So the first will receive the Raptor Lake Refresh, which, as its name suggests, is nothing more than an update of the 13th generation of Intel Core desktop chips, but with somewhat higher operating frequencies (which also translates into higher performance values), higher consumption. However, the generation of integrated notebooks is much more interesting because it shows important differences compared to its predecessor.

So below we will provide a complete overview of everything they told us about Intel Meteor Lake, all its new features, we will check what each of them means, so in the end you can deduce for yourself how revolutionary the platform is when it comes to Raptor Lake for laptops. And I’m already telling you, though there are many new thingsit is evident that Intel has focused a lot on artificial intelligence.

The process of reformulating integrated devices for PCs, in this case for notebooks, begins with a decision made several years ago. that Meteor Lake is the first generation manufactured on Intel 4node 7 nanometers companies. Since you are probably wondering why it is called 4 even though it is manufactured on a 7nm scale, in this article we will explain more about Intel nodes, the reasons for their names and what we can expect from them.

This node uses extreme ultraviolet (EUV) lithographycompared to Enhanced FinFET, which is the technology used in Intel 7, the node responsible for Alder Lake, Raptor Lake and the future Raptor Lake Refresh. This change translates into higher energy efficiency, which has been estimated at 20% compared to chips from Intel 7, a difference that becomes even more valuable when we talk about laptops.

Another big change is that Intel has finally made the leap to the MCM architecture, leaving behind the monolithic model it has followed until now. So we find a chiplet type design formed by four tiles (some rumors talked about three, but in the end there are four), between which all the elements that make up these new integrations are divided. They are the following:

• Calculation: This tile integrates powerful cores (P-cores) and efficient cores (E-cores) from Intel 4.
• SoC: is responsible for hosting, among other things, the NPU (neural processing unit) that we already told you about a few months ago, which will allow systems based on Meteor Lake to perform certain AI-related tasks much more. more efficient than before. Inside we will also find low consumption (E) cores that will contribute to improving the energy profile of the integrated unit. And it also integrates WiFi (up to 6E), Bluetooth, HDMI 2.1 and Display Port 2.1 management elements.
• GPU: As you may have already guessed, in this tile we find an integrated Meteor Lake graphics adapter, which in this generation includes Intel Arc technology.
• Input output: This chip is responsible for providing connectivity with PCIe Gen 5.0 and Thunderbolt 4 in addition, of course, to its predecessors.

To integrate all components and group tiles, Intel Foveros technology plays a key roleintroduced by Intel in 2020, a technology that allows stacking (3D) elements that make up the integrated, solving the problem caused by the need to integrate more elements, even more complex ones, in a monolithic design of the SoC type.

Intel Foveros which debuts in the PC segment with Meteor Lakeprovides an encapsulation system that allows different elements, which may originate from different nodes, to be integrated and interconnected in a very efficient way, while maintaining a closed size, high bandwidth and relatively low power consumption.

Cores

They come from the hands of Meteor Lake new Redwood Cove performance cores, responsible for taking over from Golden Cove, which debuted with the arrival of hybrid architecture at Intel in the twelfth generation Core (Alder Lake). Thus, in this generation we will find IPC-related improvements such as increased simultaneous multithreading (SMT) performance, more bandwidth per core for both caches and memory, and microarchitecture resource improvements. New improvements have also been added to the performance monitor unit

Intel also integrates new Cresmont efficiency cores, successors to Gracemont (also present in Alder Lake and Raptor Lake), which provide better support for CPU-based AI acceleration for vector neural network instructions (VNNI), among other new features. As we’ll see a bit later, AI is very present in Meteor Lake. As we told you a few weeks ago, this architecture will also be the basis of future Intel Sierra Forests.

It has undoubtedly been a key part of integrated devices since Intel made the jump to hybrid architecture the orchestrator, that is, the one responsible for distributing the load between power cores and performance cores. So we also find in this generation important improvements in it, which is reflected both in predicting the needs of each process, as well as in performance and efficiency. Improvements were also made to this end at Redwood Cove and Cresmont.

GPU

The graphics part also benefits from the MCM Meteor Lake architecture, because the encapsulation based on Foveros technology, by allowing the stacking of elements, makes it easier to devote more space to it, which is key in a component with the importance of a graphics adapter, which also plays a key role in some related tasks. to AI. In addition, Intel has separated the Xe Media Engine, which would traditionally be part of the GPU, and placed it in the SoC tile, which translates into more autonomy when performing certain types of tasks, such as AV1-related tasks.

As a result, in addition to higher performance than the previous generation of GPU APUs, the Meteor Lake GPU is able to offer Up to twice the graphics performance and power per watt. In terms of natively supported features, we find hardware-accelerated ray tracing, variable-rate shading, and feedback sampling that provide a new level of DirectX 12 support.

NPU and AI in Meteor Lake

As I mentioned several times and it was expected a few months ago, artificial intelligence at the client (i.e. those in which computing tasks are performed on the PC itself, not in a cloud service) It is one of the points, if not the main point, for this new generation from Intel. Of course, there was an interesting change, because both at the announcement and at the presentation of the first integrated device, Intel told us about the VPU (Visual Process Unit) integrated into Meteor Lake. But now we see that the stakes are much wider.

So Intel states in the technical specifications of this platform three encapsulation elements to provide different types of functionality, in response to their different needs:

• GPU: offers a high level of parallelization and performance, making it ideal for AI applications applied to media, 3D applications and rendering.
• NPU (neural processing unit:) is a dedicated low-power AI engine for AI.
• processor– Has a fast response ideal for lightweight, single-inference, low-latency AI tasks.

Without a doubt, this is the most revolutionary thing we’re talking about at this point Intel’s first package for this market segment to have dedicated integrationspecifically on AI. This is something we’ve been seeing in SoCs for smartphones and other mobile devices for a while now, so we’ve been looking forward to seeing it come to PC.

To maximize its integration with the system, Intel worked with Microsoft (in fact, the first announcement was made during Microsoft Build 2023) to ensure full compatibility with Microsoft’s new driver model, called the Microsoft compute driver model (MCDM). This translates not only into better utilization, but also greater security around the world Workflow processes in which NPU comes into play.

Another very interesting aspect of NPU Meteor Lake is that It has an architecture that integrates two engines, which, depending on the needs of each task, can work in a combined mode for a single workload or can handle independent tasks to manage two tasks simultaneously. So inside we find a pipeline dedicated specifically to inference (a fundamental task in AI) together with the VLIW DSP (Very Long Instruction Word/Digital Signal Processor), whose workload is managed by the DMA Engine, which emphasizes optimization, both performance and energy efficiency.

So, as you can see, Intel Meteor Lake represents a major technological leap compared to Raptor Lake. Of course, we will have to wait until we know the integrated systems based on this new generation of architecture, as well as the first device that integrates them.