With Meteor Lake, its new processor architecture, Intel announces a leap forward in terms of energy efficiency and enables, for example, the playback of a video with the CPU and GPU switched off.
Intel today formalized Meteor Lake, its future computer processor architecture. As the company awaits commercial launch, it details what it calls “the biggest architectural change in 40 years.” New manufacturing processes, new layout, new features… Here’s a summary of seven hours of intensive technical conferences on what exactly will change Meteor Lake for the general public.
Photo credit: Romain Heuillard
This article was written as part of the Intel Tech Tour Malaysia 2023 press tour, which Frandroid was invited to.
Intel 4: better energy efficiency
Before getting into the logical part, Meteor Lake breaks with multiple generations of processors by bringing innovations into the “client” space in terms of manufacturing processes that were previously reserved for server-grade processors.
First, Meteor Lake inaugurates both the client and server series combined with the “Intel 4” manufacturing node. After Alder Lake and Rocket Lake are made in Intel 7, we’re going from 10 to 7 nanometers, among other things. This new process node alone promises to double density and increase energy efficiency by 20%. A new simplified EUV (extreme ultraviolet) lithography process promises better yields, i.e. better yields in the production of chips, thus increasing production capacity and hopefully reducing costs.
Foveros: chip stacking
Importantly, Meteor Lake also offers client processors (as opposed to server processors) its advanced Foveros packaging process, which revolutionizes processor design. What we commonly refer to as a “processor” performs many functions that were once dedicated to separate chips on the motherboard or on daughter cards. We can speak of SoC (System-on-Chip), i.e. a more or less complete system on a single chip.
Previously, Intel offered two types of SoCs to the consumer market. On the one hand, monolithic SoCs, which consist of a single chip that fulfills all functions. On the other hand, heterogeneous SoCs, which consist of several chips connected to each other via the substrate.
Foveros combines the advantages of both methods with almost no disadvantages. The various dies now lie next to each other and are connected to each other by an interposer, which is itself an underlying die on which they are stacked. The different chips, so-called tiles, are connected to each other almost as efficiently as in a monolithic SoC. At the same time, this packaging method makes it possible to use different manufacturing processes for each tile and thus use the optimal process for each function. It also simplifies the creation of different processors.
Tiles: a sliced processor
Above all, Foveros also enables significant increases in energy efficiency. The tiled design promises to significantly reduce processor power consumption by turning off individual tiles when they are no longer needed.
Meteor Lake consists of a Compute tile, a Graphics tile, an IO tile, and a SoC tile. The latter integrates some very low-consumption “E” CPU cores, video decoding functions and display functions, in addition to vital functions (power management, RAM controller, etc.). This allows the processor to turn off the computing and graphics tiles that consume the most energy when you are just watching a video or when the computer is idle (displaying a web page, etc.). Sleeping tiles are activated immediately at the slightest request (loading a webpage, etc.).
The Intel Thread Director communicates with Windows or the operating system to contain the CPU loads on the SoC’s Low Power Efficiency cores if possible, or vice versa to collect the loads on the performance cores of the Compute Tile instead of distributing them Optimization of energy efficiency.
With the gains made at other levels, Intel promises to generalize laptop PCs that can only be charged once a day. The founder is reacting to Apple, because Apple Silicon chips are known for their performance per watt.
Integrated artificial intelligence
Meteor Lake’s second major innovation after the new manufacturing processes is the integration of an NPU (Neural Processing Unit), a unit dedicated to artificial intelligence.
PCs haven’t waited to integrate an NPU to handle artificial intelligence. They have relied on their CPU and GPU and will continue to do so. But a specialized unit makes it possible to process certain artificial intelligence tasks much more energy-efficiently. Intel gives the example of generating an image with Stable Diffusion, which would be 7.8 times more efficient with the NPU than with the CPU and 2.6 times more efficient than with the GPU. The NPU is therefore integrated into the SoC tile so that it can be used even when the CPU and GPU tile are switched off. However, the CPU remains more efficient and is still used for certain simple tasks, the GPU for certain massively parallelized tasks.
Derived from a VPU and previously presented as a VPU (Vision Processing Unit), the Meteor Lake NPU will support functions such as background blurring, automatic framing or noise reduction during video conferences. More broadly, it will be possible to bring what was previously processed on servers in data centers back to PCs at the edge, with gains in latency, bandwidth and confidentiality.
Developments at all levels
Meteor Lake also brings welcome iterative improvements, but these are more typical of a “simple” new generation of processors.
New CPU microarchitectures
First of all, Meteor Lake brings new CPU microarchitectures with new Redwood Cove P-Cores and new Crestmont E-Cores and LP E-Cores. Intel is not currently quantifying the expected performance increases; it promises, above all, “great progress” in terms of energy efficiency. Remember that Intel CPUs have been using a hybrid architecture for several generations, similar to the big.LITTLE architecture of the ARM processors in our smartphones, with cores dedicated to performance and others to efficiency.
Ray tracing for the GPU
Otherwise, Meteor Lake is introducing a new Xe-LPG GPU that Intel promises will be twice as efficient as the previous generation Xe-LP and twice as efficient. This GPU increases frequencies, lowers voltages and increases the number of vector units, geometric units, etc. Above all, it inherits certain advantages of the Arc-A series dedicated GPUs, in particular by integrating ray tracing units and providing DX12 Ultimate -Compatibility.
AV1, 8K HDR, HDMI 2.1
As we have seen, the media and display functions have been extracted from the GPU, and in this case the graphics tile, to be transferred to the SoC and IO (input/output) tiles. Since the low power CPU cores are also integrated into the SoC tile, the processor can therefore handle certain tasks such as: B. displaying a web page or a video, even with the calculation and graphics tiles turned off.
The Xe Media Engine offers hardware encoding and decoding, up to 8K 10-bit HDR, from H.264 (AVC), HEVC (H.265), VP9 and AV1, including all codecs enforced for the next few years.
The outputs HDMI 2.1, DisplayPort 2.1 and eDP 1.4. This “display engine” can also ignore repetitive images and work in bursts to save power.
PCIe 5, Thunderbolt 4 and Wi-Fi 7!
Finally, the IO Tile, i.e. the tile responsible for inputs/outputs, supports PCIe 5.0, Thunderbolt 4 (but no longer the official Thunderbolt 5) as well as Bluetooth 5.4 and Wi-Fi 7. The latter increases the maximum throughput by 2 .4 to 5.8 Gbit/s by transitioning from 1024 QAM to 4096 QAM modulation and a bandwidth from 160 to 320 MHz. During the demonstration with a TP-Link Wi-Fi 7 access point, we observed 4.2 Gbit/s when sending and 2.8 Gbit/s when receiving.
First processors on December 14th
At the end of this technical presentation of the main principles of Meteor Lake, it seems that this new generation of processors implements several innovations to achieve further advances in both performance and energy efficiency. Efficiency has indeed become an important criterion as the share of laptops in the computer market is constantly growing and expectations for them are increasing.
To know the details of the range, that is, the references of the different processors and their composition (number of cores, frequencies, etc.), you will have to wait for the commercial launch scheduled for December 14, 2023. Only Intel says these processors will adopt the new high-end Core Ultra brand.
We can conclude that these “Core Ultra” based on the Meteor Lake architecture will coexist with “Core” based on an earlier architecture, Raptor Lake or Raptor Lake Refresh. Above all, it is certain that Meteor Lake will, at least initially, be reserved for mobile chips for laptops, mini PCs (like NUCs), portable consoles, etc. Desktop PCs would be eligible for Raptor Lake Refresh, an evolution of the Raptor Lake architecture of the 13th generation Core processors, but Intel is refusing to comment on the matter. We will know more in the coming months!