At the Computex 2024 event, Intel revealed some details about their new Lunar Lake processors for AI PCs, which are the next step after the Meteor Lake processors. These new processors promise up to 40% more power and improved energy efficiency. Intel aims to accelerate the AI ecosystem from data centers to PCs with their new technologies and architectures.
The Lunar Lake processors are designed for thin and light computers, offering significant improvements in performance and energy efficiency. They feature a 4th generation Intel NPU with up to 48 TOPS of AI performance, providing four times more AI computing power than previous generations. Additionally, the GPUs have been redesigned to include Xe2 cores for graphics and Xe Matrix Extension for AI, offering up to 67 TOPS performance.
One of the key features of the Lunar Lake processors is an advanced low power island that manages background tasks to extend battery life. Intel has already started mass production of these processors, which will be featured in over 80 different AI PC designs from 20 OEMs.
In addition to the Lunar Lake processors, Intel also launched the new Xeon 6 processors, which incorporate efficient cores for improved performance and energy efficiency in high-density workloads. These processors are designed to handle various usage scenarios and workloads, from AI to cloud-native applications.
The Xeon 6 processors offer rack consolidation, increased performance at the rack level, and improved performance per watt compared to previous generations. They are designed to be the ideal CPU head node for AI workloads and can be paired with Intel Gaudi AI accelerators for optimized performance.
Intel also announced the availability of Intel Gaudi AI accelerators for customers seeking GenAI performance. These accelerators offer a price-performance advantage and can be combined with Xeon processors for efficient AI workloads. Intel is offering standard AI kits that include multiple Gaudi accelerators at different price points to cater to various customer needs.