Intel’s Meteor Lake chips, set to launch on December 14th, promise to be the company’s most power-efficient client processors ever. They feature distinct chiplets for each component, operate on the Intel 4 process node, include a dedicated AI coprocessor, and offer up to twice the graphics performance. Additionally, they incorporate a “low power island” for independent task execution and integrate with Microsoft Windows for intelligent chip control.
It consists of modular components
Intel is adopting a Lego-like approach with Meteor Lake, building computer chips using separate pieces of silicon for components like the CPU and GPU, printed at different sizes and then combined. While this approach is already utilized by AMD and Qualcomm, it’s relatively new for Intel. Notably, not all parts of Intel’s chip use their cutting-edge Intel 4 process; the graphics use TSMC’s 5nm process, and the I/O and the new “SoC Tile” rely on TSMC N6. This approach allows for greater flexibility in selecting the best building blocks and selectively controlling power usage.
Intel aims to significantly reduce power consumption right from the start by implementing chiplets. Instead of a single CPU or display region, each Meteor Lake chip has two: one on a “low power island” that can potentially operate independently. This island includes its own efficient CPU core, NPU AI coprocessor, media engine, and memory.
The rest of the processor cores are divided into different components: the “Compute Tile” on Intel 4, which houses the P (Performance) and E (Efficiency) cores known as Redwood Cove and Crestmont, respectively, and a separate Graphics Tile on TSMC N5. For most users, the Compute Tile will be the primary focus. However, Intel aims to minimize chip heat generation by employing an improved “Thread Director.” This directs tasks to the lowest-power cores before utilizing higher-power ones. This approach also extends to the AI coprocessor, which is visible and monitorable in Windows Task Manager.
With the separation of “Media” from “Graphics,” Meteor Lake enables integrated video encoding and decoding without necessarily utilizing the graphics tile. For instance, it includes hardware support for AV1 film grain features, which were previously handled by GPU shaders.
Moreover, the SoC tile offers native support for AV1 video, HDMI 2.1, DisplayPort 2.1, and the capability to handle either 8K HDR video or connect up to four 4K monitors simultaneously. Additionally, it features Bluetooth 5.4 and Wi-Fi 7, which theoretically provides transfer speeds of up to 5.76Gbps when utilizing a 320MHz channel in your workplace or neighborhood.
Is it capable of gaming?
However, even though various graphics-like tasks are handled separately from the GPU, it doesn’t mean the GPU is merely decorative. Intel asserts that Meteor Lake can incorporate its Intel Arc graphics, now featuring dedicated ray-tracing units and up to 8 Xe cores directly on the chip.
Intel claims that the “Xe LPG” GPU delivers up to double the performance per watt compared to Xe LP, which themselves offered twice the performance per watt compared to Intel’s 11th Gen Intel UHD integrated graphics. Furthermore, these graphics also support Intel’s XeSS, an intelligent upscaling technology similar to Nvidia’s DLSS and AMD’s FSR, making it available for the first time on integrated graphics. This could potentially lead to further improvements in frame rates.
Intel has announced that its new GPU offers improved performance compared to previous integrated GPUs, boasting the ability to operate at lower voltage levels while achieving clock speeds exceeding 2GHz. Additionally, Intel has introduced its patented cost-effective vapor chamber cooling solution to support gaming and creator laptops featuring Meteor Lake processors.
However, it’s important to note that not all Meteor Lake/Core Ultra processors will include the advanced GPU. Specific MTL processor-powered systems with dual-channel memory will be the only ones equipped with Intel Arc graphics.
Is the presence of AI assured?
The NPU, which is the AI coprocessor, is expected to be a standard component in every chip. According to Intel’s Tim Wilson, VP of architecture, “The NPU will be available across the full product stack of Meteor Lake.” This aligns with Intel CEO Pat Gelsinger’s statement in a July earnings call, where he mentioned, “We’re going to build AI into every platform we build.” However, it hasn’t been confirmed if Meteor Lake will be available outside of the Core Ultra, so choosing a non-Ultra chip might not include the NPU for now.
Intel isn’t implying that this small NPU will replace the need for large cloud servers equipped with Nvidia H100 chips for generative AI tasks. Nor is the company suggesting that this single NPU can handle all AI tasks effortlessly. Instead, Intel is presenting options: it’s more efficient to have the NPU handle specific tasks like running the image generator Stable Diffusion, and it’s faster and still reasonably efficient to have both the GPU and NPU collaborate on such tasks.
Intel didn’t provide many specific examples of how AI will revolutionize tasks on Windows, as coined by Microsoft’s outgoing Windows chief, Panos Panay. Their examples were somewhat reminiscent of what Qualcomm’s NPUs can already accomplish with Windows Studio Effects in Microsoft Teams.
Nonetheless, Intel is offering tools like the OpenVINO Inference Engine to facilitate communication with applications and direct data processing to the NPU.
Intel is using Meteor Lake to showcase its commitment to executing effectively, as part of CEO Gelsinger’s ambitious turnaround strategy for the company. Intel claims that the yields of the Intel 4 components of Meteor Lake, particularly the Compute Tile, surpass not only their previous 14nm Broadwell and 10nm Ice Lake introductions but also their refinements within the “tick-tock” strategy with Skylake and Tiger Lake, respectively.
The actual performance and battery life of these components are crucial factors, but Intel has not provided specific details or answers to questions regarding these aspects. During a conference call, representatives repeatedly apologized and mentioned that they would address such questions closer to the launch date.
Regarding CPU performance, the only hints provided are that the new Crestmont E-cores exhibit unspecified “IPC gains over previous E-cores,” although the fine print notes that results may vary. Additionally, the new Redwood Cove P-cores are said to offer “improved performance efficiency.”
However, Intel emphasizes significant investments in this future model. They are allocating $3.5 billion to New Mexico through the next year and $7 billion to Penang, Malaysia, over the next decade to boost capacity for Meteor Lake and other Foveros-style chips. These chips involve three-dimensional stacking and assembling different building block chips onto a single package. Furthermore, Intel is progressing towards direct copper-to-copper bonding, referred to as Foveros Direct, and intends to place these chips on glass by the end of the decade.
