Intel's Lunar Lake Architecture: The Next Generation Of Core Ultra 200V Processors

Intel has recently unveiled its next-generation Core Ultra 200V processors, known as Lunar Lake, during an event in Berlin ahead of IFA. This new architecture aims to deliver competitive performance while operating at ultra-low power levels, specifically designed for thin and compact laptops. With Qualcomm's ARM-based Snapdragon X Elite already making waves in the Windows PC ecosystem for its efficiency, Intel's Lunar Lake architecture is poised to make significant strides in the market.

Intel Lunar Lake Architecture Overview

Last year, Intel shifted from its traditional monolithic design to a tile-based chip design with its Meteor Lake processors. Lunar Lake takes this concept further by integrating not just the CPU and cache within the Compute tile but also the GPU and NPU, making it a comprehensive processing unit. The result is that the Compute tile is now the largest tile on the die.

This year, the Lunar Lake architecture is being manufactured using TSMC’s N3B process node. Although the yield of N3B is lower compared to the newer N3E node, this transition marks a significant move for Intel away from its foundry to TSMC's advanced 3nm technology, which is an encouraging sign for performance improvement.

The platform controller tile, responsible for I/O and connectivity, continues to utilize TSMC’s 6nm (N6) process, similar to the previous Meteor Lake design. This is also the first instance where Intel designs its processors while TSMC handles the manufacturing. The entire chipset is packaged using Intel's innovative Foveros 3D technology.

Memory Integration and Performance Enhancements

In a notable upgrade, Intel has incorporated memory directly into the processor. This means that Lunar Lake chips will support unified memory, akin to Apple’s M-series chips. The on-package LPDDR5X-8533 RAM is available in 16GB or 32GB configurations, boasting memory speeds of up to 8,533 MT/s. This integration is expected to enhance efficiency and reduce power consumption significantly.

Overall, the Lunar Lake architecture represents a marked improvement over its predecessor, with enhanced CPU, GPU, NPU, and cache integration within the Compute tile. By using TSMC’s advanced 3nm process node, Intel aims to boost efficiency while also reducing power consumption and enhancing bandwidth.

During the Computex event, Intel's executive VP and GM, Michelle Holthaus, stated, “We’re going to bust the myth that [x86] can’t be as efficient.” Intel claims that the x86-based Lunar Lake processors achieve a remarkable 50% reduction in power consumption when compared to the previous Meteor Lake chips.

Key Features of the Lunar Lake CPU

All Lunar Lake processors come equipped with eight CPU cores—four performance (P) cores dubbed Lion Cove and four efficiency (E) cores named Skymont. Intel asserts that the P-core Lion Cove yields a 14% improvement in IPC (instructions per cycle) compared to the Redwood Cove P core used in Meteor Lake.

Interestingly, Intel has taken a bold step by completely eliminating Simultaneous Multi-Threading (SMT)—better known as HyperThreading—from its processors for the first time in over two decades. This decision is aimed at improving performance-per-watt by approximately 15%. Without HyperThreading, Lunar Lake processors focus on executing more instructions per cycle, enhancing single-threaded task performance.

On the efficiency side, the E core Skymont is particularly impressive, offering a substantial 68% IPC improvement over the Crestmont E core found in Meteor Lake. The Skymont cluster operates in a 'Low Power Island,' allowing it to consume just one-third of the power when reaching peak performance, effectively delivering twice the performance in single-threaded tasks compared to its predecessor.

Granular Clock Speed Control and Competitive Battery Life

Another innovative feature of the Lunar Lake architecture is its granularity in clock speed boosts. Instead of increasing clock speed in larger increments that typically consume more power, this architecture allows for finer adjustments—incrementing clock speeds by just 16.67MHz to better manage power budgets.

Intel claims that this approach enables the Lunar Lake CPU to achieve single-threaded performance levels comparable to Meteor Lake at just half the power consumption, a significant achievement in efficiency.

In terms of competition, Intel asserts that Lunar Lake processors not only match the performance per watt of Snapdragon X Elite but exceed it by 20% within the same power envelope. This translates to an impressive battery life estimate of around 14 hours for office productivity tasks, in contrast to the Snapdragon X Elite's 9.5 hours and AMD HX 370's 10.1 hours.

Geekbench Scores and Performance Expectations

While the Lunar Lake processors are set to launch on September 24, early Geekbench scores have surfaced. The entry-level Core Ultra 5 228V has demonstrated a single-core score of 2,621 and a multi-core score of 10,072, with a maximum clock speed of 4.5GHz and a default TDP of 17W.

Conversely, the high-end Core Ultra 9 288V achieves a single-core score of 2,790 and a multi-core score of 11,048. In some instances, it even surpassed the 2,900 mark in single-threaded tasks, with the SKU capable of reaching up to 5.1GHz and a default TDP of 30W.

Advancements in Graphics Performance

The integrated GPU within the Lunar Lake architecture is built on Battlemage graphics architecture, featuring eight second-gen Intel Xe cores. This new GPU design incorporates larger ray tracing units, enhancing gaming performance and real-time ray tracing capabilities. For AI-related tasks, the Lunar Lake GPU can achieve up to 67 trillion operations per second (TOPS), providing a substantial boost in performance.

When compared to the GPU in Meteor Lake, the Lunar Lake GPU is 31% faster while consuming half the power, and it also offers up to 60% faster XeSS AI upscaling. Notably, the Lunar Lake GPU outperforms Qualcomm's Adreno GPU found in Snapdragon X Elite by 68% and is 16% faster than AMD's latest HX 370 GPU.

Additionally, the display engine in Lunar Lake processors can support three 4K HDR displays at 60Hz or a single 8K HDR display at the same refresh rate. They also come with support for AV1 encoding and decoding, making them well-equipped for modern multimedia tasks.

Enhanced Neural Processing Unit (NPU)

Intel's previous Meteor Lake faced criticism for its underwhelming NPU, which could only execute up to 10 TOPS. However, with Lunar Lake, Intel is introducing a more robust NPU capable of performing up to 48 TOPS, exceeding Microsoft’s 40 TOPS requirement for Copilot+ PCs, set to launch in November.

When combined with the overall processing capabilities, the Lunar Lake architecture can achieve a staggering total of 120 TOPS, which surpasses Qualcomm’s Snapdragon X Elite, which has a total of 75 TOPS capability. These figures are calculated based on the INT8 data type, reflecting the processing power available for complex tasks.

Exploring the Range of Lunar Lake SKUs

The Core Ultra (Series 2) processors based on Lunar Lake offer an array of SKUs, each equipped with eight CPU cores. The key differentiators among these SKUs lie in memory, CPU/GPU clock speed, and NPU performance.