AI-powered next-gen processor ‘Lunar Lake’ is cost and power-efficient, says Intel’s Dan Rogers

Intel’s latest Lunar Lake chips are expected to be significantly more power efficient than the company’s first generation of AI PC chips. This is according to Dan Rogers, Vice President and General Manager of Intel’s Client Performance Marketing Lab, Client Computing Group.

During an exclusive interview at the Intel Tech Tour in Taipei, Taiwan, Rogers confirmed that Lunar Lake has a 40% lower power package reduction compared to Meteor Lake, inclusive of memory. This reduction in power consumption is expected to lead to smaller boards, which will help reduce weight and make space for a bigger battery in similar chassis. Additionally, the reduced board size can also lead to a slight reduction in board cost.

Lunar Lake is a next-generation AI processor that features a dedicated neural processing unit (NPU) capable of up to 48 trillions of operations per second (TOPs) of AI performance. This NPU offers both cost and power reduction benefits, making it a more efficient option compared to other AI accelerators.

The Lunar Lake system on chip (SoC) will offer up to 32 GB of RAM, with even the entry-level stock keeping unit (SKU) having at least 16 GB of RAM. This is twice the 8 GB RAM offered on the Apple Silicon that powers the latest MacBook Air and MacBook Pro models.

In terms of AI performance, Lunar Lake offers a total AI performance of up to 120 TOPs, with the central processing unit (CPU) offering up to 5 TOPs, the NPU offering up to 48 TOPs, and the Xe2 graphics processing unit (GPU) delivering up to 67 TOPs. The NPU on Lunar Lake is slightly more powerful than the one on Qualcomm’s Snapdragon X Elite and X Plus, which offers up to 45 TOPs of AI performance.

The upcoming Lunar Lake chips are expected to power Intel-based Copilot+ PCs, launching in the second half of the year, and are likely to compete with ARM-based Copilot+ PCs.

When asked about the presence of a dedicated AI accelerator despite the GPU capable of offering more AI inferencing capabilities, Rogers explained that the GPU will have a higher power budget across the client architecture, including Lunar Lake, while the NPU will be a power-optimized architecture. He emphasized that while a GPU offers peak AI performance, suitable for tasks like on-device image generation, an NPU is an excellent choice for AI programs that need to run continuously or in the background.

Rogers also confirmed that Lunar Lake will not be limited to laptops but will also come in other form factors such as small form factor desktops, commonly referred to as mini PCs. Intel has also announced a new developer kit featuring a Lunar Lake processor, which can be upgraded to a next-generation processor in the future.

Intel is heavily investing in Lunar Lake processors, which can be considered their flagship product, with several first-on-an-Intel-chip features. The first set of Lunar Lake-powered laptops is expected to roll out globally by the end of 2024, and some of these machines are expected to offer on-device generative AI capabilities.