Intel’s greatest laptop computer CPU replace in years is a big departure from previous designs – Ars Technica

Intel’s next-generation Meteor Lake laptop computer processors are nearing launch—the corporate introduced this week that the primary processors will launch on December 14. It’s unclear whether or not precise Core and Core Extremely methods will probably be in the stores on that date, however at a naked minimal, the official announcement will pave the way in which to many laptop computer bulletins at CES in January.

We already know numerous primary details about Meteor Lake; it makes use of a mixture of chiplets manufactured by each Intel and TSMC reasonably than a single monolithic die, and it’ll mark the retirement of Intel’s “nth-generation” and i3/i5/i7/i9 branding. We additionally know that it received’t be prepared for desktops and that the following spherical of Core desktop CPUs will probably be similar to the Twelfth- and Thirteenth-generation chips.

However at Intel’s Innovation occasion this week, the corporate dove a little bit deeper into a few of Meteor Lake’s developments, describing extra about how the chips would steadiness E-cores and P-cores and asserting its most substantial built-in GPU improve in years. We’ll hit some highlights under, although it’s price watching or studying the total presentation to seek out out extra.

Tiles for miles

Meteor Lake will probably be Intel’s first client processor to maneuver to a chiplet-based design—as an alternative of being one monolithic die that incorporates the CPU, GPU, and all the opposite bits and items wanted for a contemporary laptop computer CPU, Meteor Lake is break up into 4 “tiles” which might be joined collectively by a fifth base tile that enables them to speak with one another. The method of stacking the chiplets on prime of the bottom tile is a packaging expertise that Intel calls Foveros.

Here is a primary breakdown of what’s in every of these 4 tiles:

• • The compute tile is the place a lot of the precise CPU is. Intel’s present renderings present a chip with six high-performance P-cores primarily based on the Redwood Cove structure, and eight high-efficiency E-cores primarily based on the Crestmont structure.
  • The graphics tile is the place most graphics processing occurs, although a number of particular capabilities that you simply’d often discover in a GPU have been moved to different tiles. Meteor Lake’s built-in GPU is usually simply an built-in model of an Intel Arc devoted GPU, full with {hardware} ray-tracing acceleration.
  • The I tile handles most exterior connectivity, together with PCI Specific 5.0 lanes and Thunderbolt 4 assist (Thunderbolt 5 might want to wait).
  • The SoC tile might be probably the most fascinating of the 4. It consists of two further Crestmont E-cores, the media encoding and decoding engine that may usually be situated within the GPU, and the neural processing unit (NPU) used for accelerating AI and machine studying workloads. It additionally handles Wi-Fi and Bluetooth connectivity and connecting to exterior shows over HDMI 2.1 and DisplayPort 2.1.

One of many notable issues about Meteor Lake is that not all the tiles are being manufactured by Intel. The compute tile, which homes the precise P-cores and a lot of the E-cores, makes use of the brand new Intel 4 course of, an improve from the Intel 7 course of used for many present Core chips. However the graphics tile is being made on a 5 nm TSMC course of, whereas the IO tile and SoC tile are made on a 6 nm TSMC course of.

Intel has additionally used TSMC’s manufacturing to make its Arc GPUs, so it is not the primary time we have seen these two chipmaking frenemies work collectively. However Intel is making an attempt to meet up with TSMC’s manufacturing, and Intel sees its foundry operations as key to its future progress. I would not be stunned if shifting again to all-Intel-made tiles is the eventual purpose.

E-ven extra E-cores

Intel has stated that Meteor Lake’s P-cores do not change a lot in comparison with those utilized in Twelfth- and Thirteenth-generation Alder Lake and Raptor Lake CPUs—we may even see greater clock speeds, however not a lot has modified by way of instructions-per-clock or instruction set. The E-cores do get some enhancements, although.

Meteor Lake really consists of two completely different sorts of E-cores. There are two low-power (LP) E-cores within the SoC tile, and the up to date model of Intel’s Thread Director will try to make use of these E-cores for as many duties as attainable. Intel calls this chunk of the SoC tile the “low-power island” as a result of the thought is to permit the compute tile and the graphics tile to utterly energy down as a lot as attainable to avoid wasting energy.

When duties require extra efficiency than the LP E-cores can present, Thread Director will kick them to the compute tile—both to the principle E-core clusters, that are tuned to deal with low-impact multithreaded workloads, or to the P-cores, that are used for single-threaded duties and any multithreaded work the E-cores cannot deal with. This can be a change from how Thread Director works in Twelfth- and Thirteenth-generation processors, the place high-priority duties would head straight to the P-cores with out making an attempt the E-cores first. (Although it stays to be seen whether or not the Thread Director modifications will lead to any sort of user-noticeable delays for high-performance duties.)

Additionally price noting: Crestmont E-cores might be added to processors in teams of two, the place previous-generation Gracement E-cores may solely be added in teams of 4. This would possibly make it simpler for Intel to justify bringing small teams of E-cores to lower-end processors that did not have them earlier than. All E-cores stay single-threaded, whereas P-cores nonetheless assist two threads per core.

The brand new E-cores additionally embrace a number of different niceties—VNNI directions for accelerating AI workloads, and even AVX10, which brings lots of the advantages of Intel’s AVX-512 directions with out requiring 512-bit registers. The Twelfth- and Thirteenth-generation Core processors completely shut off AVX-512 assist as a result of the E-cores did not assist it, despite the fact that assist was current within the P-cores. This has created an ungainly state of affairs the place AMD’s newest Zen 4 chips assist the AVX-512 directions that Intel invented and promoted, the place Intel’s newest client chips do not.

Built-in Intel Arc: Twice the efficiency per watt

Intel hasn’t considerably improved its built-in graphics efficiency since late 2020 when it launched its Iris Xe built-in GPU in Eleventh-generation laptop computer CPUs. The Twelfth- and Thirteenth-generation processors used basically the identical GPU, so graphics efficiency on Intel ultraportables has stayed nonetheless whereas AMD has launched new RDNA 2 and RDNA 3-based built-in GPUs into Ryzen laptops.

Meteor Lake will change this. Intel says the brand new graphics tile will supply as much as twice as a lot efficiency per watt because the Iris Xe GPU in present Intel processors. The last decade-old Iris title additionally appears to be going away, to get replaced by the identical “Arc” branding Intel makes use of for devoted GPUs. Meteor Lake additionally helps the identical video encoding and decoding acceleration as Arc—together with hardware-accelerated AV1 video encoding—however the media engine is constructed into the SoC tile reasonably than the GPU.

The GPU is predicated on an structure known as Xe-LPG (low-power gaming) that’s principally similar to the Xe-HPG (high-performance gaming) structure in Arc A-series GPUs, changing the older Xe-LP structure in Iris Xe GPUs.

In comparison with Xe-LP, Xe-LPG sports activities greater clock speeds, extra GPU {hardware}, and architectural enhancements. A totally enabled Xe-LPG GPU will embrace 128 of Intel’s Vector Engines (previously Execution Models, or EUs), up from a most of 96 in Xe-LP. Intel did not say something about lower than absolutely enabled Xe-LPG GPUs, however the firm ships a model of Iris Xe with solely 80 of the EUs enabled, so a cut-down model of the Arc-integrated GPU appears possible, too.

Whereas Xe-LPG has practically every part {that a} devoted Arc GPU consists of, it’s lacking one factor price mentioning: Intel Matrix Extensions, or XMX. In Arc GPUs, these are used to enhance the efficiency and look of Intel’s XeSS upscaling in video games that assist it. Since these upscaling algorithms are fairly useful to have once you’re making an attempt to get a recreation operating at 60 FPS on an built-in GPU, it is too unhealthy that Xe-LPG solely helps the less-capable GPU-agnostic model that makes use of DP4a directions as an alternative of XMX.

We’ll want to attend to see how an Xe-LPG GPU performs in comparison with an Iris Xe GPU—doubling the efficiency per watt doesn’t suggest efficiency has been doubled—however Meteor Lake ought to have probably the most fascinating built-in GPU we have seen from Intel in fairly some time.