New Intel Core Ultra (Meteor Lake) processors coming on December 14 with AI module – Intel

Intel’s vision includes technologies that provide easier access to cloud-based AI solutions, a better performance/price ratio for Intel AI accelerators for the data center market than competitors, and tens of millions of processors with AI accelerators in PCs as of the beginning of 2024. In addition to Tools aimed at secure application of AI at the “edge of the network.”

Aside, at least for now, from the business applications of AI, Intel’s 2023 Intel Innovation Conference gave analysts and journalists in attendance a chance to learn a lot of details about the upcoming processors codenamed “Meteor Lake” that are expected to be available. On laptops on the market as of December 14. As part of its new naming scheme, Intel is calling its new processors “Core Ultra” and their processors will be the first to be manufactured using the “Intel 4” (7nm) process. It is also the first processor to include a neural processing unit (NPU) to accelerate artificial intelligence applications. In almost every way, Core Ultra processors essentially reveal where Intel is headed: efficient and powerful processors that can keep up with the demands of modern AI.

Currently, with the Core Ultra processors, the company seems to be fully focused on the laptop market. Rumor has it that Intel will soon announce “Raptor Lake-S Refresh” processors for desktop computers, so the corresponding version of Core Ultra for this market will be postponed for about a year. While it may be surprising that Intel decided to prioritize laptop processors over desktop processors, its strategy makes sense. The new laptops are supposed to provide long battery life, sufficient processing power, and great capabilities in the field of artificial intelligence, and this is what the company wanted to promote at the Intel Innovation 2023 conference, so it decided to separate the two families of fourteenth generation processors, as they will instead come in contrast to What desktop users really want: as much power as possible.

Core Ultra

The new Core Ultra processors incorporate the latest innovations from Intel, with the first and best FOVEROS 3D package. It also includes all-new P (Performance) and E, (Efficiency) cores with the company having a strong focus on power efficiency, and has an “integrated graphics” (iGPU) that delivers up to 2x the performance/watt compared to the past while in addition to the NPU Integrated, they can take advantage of the CPU and GPU cores for other AI-related tasks. Unfortunately, the new Core Ultra processors will not support the company’s recently announced Thunderbolt 5 standard, but will support Thunderbolt 4, DDR5 memory, and PCIe Gen 5.

As mentioned in the text above, the Meteor Lake processor includes three logical units, the central processing unit (GPU), the integrated graphics processing unit (iGPU), and the neural processing unit (NPU). Each of the above modules is located in a separate matrix (die) and is very bandwidth demanding. Using the new FOVEROS 3D packaging technology, the company decided to use four different boxes to build its processors, namely Compute, Graphics, SoC and I/O.

count – count

The compute tile contains the processor cores (CPU) or rather its main computing engine. More specifically, it includes P cores and E cores manufactured using Intel’s 4 (7nm) process, latent level 3 (L3) memory as well as the popular Ringbus hyperlink. The new P core is called “Redwood Cove” and provides higher IPC (instructions per clock), faster clock speeds and greater ISA capabilities than the previous generation “Raptor Cove”. In the case of the new Core Ultra (Meteor Lake) processors for laptops, we do not expect to see models with more than 6 P cores.

In addition to the 6 P cores, the Compute chip includes up to two cores each with four E cores. The new E core is codenamed “Crestmont” and offers a significantly improved performance/watt ratio but also a significant increase in IPC in terms of E basic today codenamed “Gracemont”.

Of course the “Redwood Cove” cores have dedicated Level 2 (L2) latency, while each of the two “Crestmont” core clusters has an L2 cache shared between the four cores that make them up. On the other hand, E-core and P-core clusters share a unified third-level (L3) cache.

SOC

The SoC includes both a Neural Processing Unit (NPU, essentially the “hardware backend” for Intel’s AI Boost innovation) and the processor’s Media Accelerator as well as a Display Controller.

Memory controllers as well as a PCIe root collector are also integrated. Another element of the SoC – which is somewhat surprising – is a module consisting of a quartet of electronic cores, which the company refers to as a “low power island”. The cores are based on the same “Crestmont” architecture as the electronic cores on the compute chip, but they are not part of Ringbus and do not share the L3 cache.

This specific core group has its own L2 cache which is shared between the four cores. The Low Power Island array’s electronic cores are part of the processor core where they are exposed as logical processing units to the operating system and users can monitor them in the Windows Task Manager. They have the same ISA as the electronic cores of the compute board. The reasons for having this extra core set are two-fold: first, the company places the x86 core close to other controllers on the die, most of which are close to the NPU (possibly to reduce latency), and second, it allows Intel to disable the compute system at will. Tiles when not needed to achieve significant energy savings.

As we said, the SoC also has an NPU to accelerate high-performance inference. The NPU has four basic hardware-level components, a cross-bar called Global Control, along with a Memory Mapping Unit (MMU) and a DMA, a small amount of Random Access Memory (RAM), and two NCEs (Neural Computing Engines). Each NCE contains a programmable DSP and inference pipeline with the latter including a MAC array, the main analysis resource.

Graphics

A graphics card, as the name suggests, includes an integrated graphics processing unit (iGPU) and more specifically any components necessary to display and compute the graphics but without the media accelerator and display engine/display controller.

Through “Meteor Lake” processors, Intel plans to double the performance of its graphics processing unit compared to the last generation. The iGPU is divided into three tiles, the graphics tile which contains the main hardware for the graphics rendering process which is distributed in the front end of the iGPU and the Xe Render chips which integrate the Xe cores (Xe Cores). The SoC contains multimedia accelerators and a display engine. The I/O board contains the physical interfaces for the monitor/display which include DisplayPort and HDMI.

The Xe-LPG architecture is an evolution of the Xe-LP architecture, which Intel’s latest generations of iGPUs are based on. However, it is still “less dynamic” than the Xe-HPG architecture that the GPUs in the Arc discrete graphics cards we know under the “Alchemist” codename are based on. With Xe-LPG, Meteor Lake iGPUs gain full DirectX 12 Ultimate capabilities, including real-time ray tracing acceleration. At the top end, the iGPU has 8 Xe Cores and 2 Xe Render Slices which equates to 128 EU which in turn translates to 1024 uniform shaders. The iGPU also contains 2 engineering pipelines (equivalent to 8 ROPs), 8 samplers, 4 pixel backends, and 8 ray tracing units.

I/O

Although the input/output (I/O) board is physically different from the SoC, it is an extension of the SoC and contains all of the processor’s interfaces at the physical level.

The I/O board contains all the physical layer interfaces such as the processor’s PCIe interface as well as other interfaces such as Thunderbolt and USB4. The I/O board is basically an extension of the SoC. Intel decided it was better to create a separate I/O board for space reasons as it could use different sized I/O boards to build different processor models to serve different needs. The I/O board in the presentation was the first, with the most PCIe lanes, and the most USB4 and Thunderbolt connections.

As mentioned, Intel’s new Core Ultra processors are expected to be released starting December 14th inside several new laptops.