In a significant reveal at Computex 2024, Intel introduced its cutting-edge consumer and professional CPU architectures and shared advancements in AI technology. The event was a spectacle of innovation with Intel’s Lunar Lake CPU and Xe2 GPU architectures taking center stage, alongside the announcement of the Xeon 6 “Granite Rapids” and “Sierra Forest” CPUs.
Introduction to Lunar Lake: Lion Cove and Skymont Cores
The much-anticipated Lunar Lake architecture was unveiled, promising to shift dynamics in the computing world significantly. Intel is focusing not just on performance but also on AI capabilities, with these efficient System on Chips (SoCs) joining the Microsoft Copilot+ initiative. Lunar Lake is distinguished by a tile configuration that includes seven elements, such as the package, memory, Compute and Base tiles, Foveros interconnect with a stiffener, and Platform Controller Tile. Notably, Intel opted for TSMC’s N3B and N6 node technologies for various components.
Lunar Lake features a hybrid design, using a P-Core and E-Core configuration. The P-Core is based on the new Lion Cove architecture, while the E-Cores implement Skymont architecture. Intel harnesses Thread Director technology, contributing to notable performance improvements across the board.
Lion Cove and Skymont: Performance Gains
The Lion Cove core architecture has achieved a 14% improvement in Instructions Per Cycle (IPC) over its predecessor, with impressive scalability across different power consumptions. The Skymont cores boast a 38% improvement in single-threaded (ST) integer performance and a massive 68% in Floating Point operations when compared to Crestmont E-Cores from the Meteor Lake range. The incorporation of Thread Director ensures optimized power distribution, enhancing the lineup’s power efficiency.
Intel NPU 4 and Connectivity Advances
With Lunar Lake, Intel introduces the “NPU 4” for high-end AI processing, attaining a groundbreaking 120 Tera Operations Per Second (TOPS). This accounts for 40% of the total CPU processing power, marking a 4.36x increase from the NPU in Meteor Lake SoCs. Lunar Lake also brings additional connectivity, featuring support for WiFi 7, Thunderbolt 4, Wi-Fi proximity sensing, and significant improvements in Bluetooth functionality, including faster boot and wake-up times.
Emerging Lunar Lake Products
Though Intel has not yet disclosed specifics about Lunar Lake SKUs, live demos gave insights into potential configurations. One particular SKU, the Intel Core Ultra 7 268V, seems to be an 8-core model, including both P-Cores and E-Cores, coupled with a base clock of 2.20 GHz. It sported 14 MB of L3 and 12 MB of L2 cache while running on the Xe2 graphics architecture alongside the NPU 4 AI engine.
Furthermore, MSI’s Claw 8 AI+ “A2VM” gaming handheld device was spotted, leveraging a Lunar Lake SoC with the latest CPU cores and Xe2 “Battlemage” graphics, underlining the industry’s enthusiasm and readiness for Lunar Lake’s integration.
Intel’s Xe2 GPU Architecture: A Leap in Graphics Performance
Amidst speculation, Intel introduced the Xe2 GPU architecture, streamlining previous naming conventions and doubling down on performance efficiency. The architecture sees a reconfiguration of compute resources, improved vector engines, deeper caches, and a revamped XMX unit. A significant enhancement is the improved RTU (Ray Tracing Unit), designed to boost ray tracing performance at all levels.
Intel promises a 50% performance boost over the preceding Meteor Lake’s GPU, all while achieving reduced power consumption. The integration of the Xe2 GPU will first be seen with Lunar Lake processors, the architecture boasting 8 Xe2 cores for superior execution for both graphics and computing tasks.
Xeon Updates and Xe2 Integration
Intel also previewed developments in its Xeon lineup with “Granite Rapids” and “Sierra Forest,” though details remain sparse. The integration of Xe2 with the “Battlemage” suite was not announced, hinting at potential future disclosures.
In a showcase of next-gen performance, Intel demonstrated a gaming setup featuring an undisclosed Lunar Lake CPU paired with the Xe2 GPU, running within a 17W power envelope. The test was conducted on a high-definition F1 2024 game, indicating that enthusiasts can anticipate substantial generational advancements with the Xe2 architecture.
Overall, Intel’s announcements signify a robust effort to redefine computing norms, emphasizing AI integration, power efficiency, and high-level graphics performance. With these revelations, the tech community eagerly awaits the roll-out of these futuristic computing solutions.Intel is set to strengthen its presence in the data center market with the release of its new server CPUs, designed to deliver enhanced performance in a range of compute-intensive and AI applications, including High-Performance Computing (HPC), databases and analytics, AI, networking, edge computing, as well as infrastructure and storage solutions. The new offerings are part of Intel’s Xeon 6 family and are divided into two main categories: the Granite Rapids CPUs and the Sierra Forest CPUs, each catering to different performance demands.
The Granite Rapids group, also known as the Xeon 6900P lineup, is poised to set high standards for Intel’s server CPU offerings, with the top-tier models expected to debut in the third quarter of 2024. Subsequent members of the lineup, including the Xeon 6700P and Xeon 6300P, are slated for release in the first quarter of 2025.
Focusing on the Granite Rapids CPUs, these P-Core based processors will potentially incorporate up to five chiplets, with the highest-end model featuring triple compute tile configurations that accommodate up to 128 cores. The configurations for the Granite Rapids family vary based on the number of compute and I/O tiles:
1. Xeon 6900P (XCC SKU) – 3 Compute Tiles + 2 I/O Tiles = Up to 128 Cores
2. Xeon 6700P (XCC SKU) – 2 Compute Tiles + 2 I/O Tiles = Up to 86 Cores
3. Xeon 6500P (HCC SKU) – 1 Compute Tile + 2 I/O Tiles = Up to 48 Cores
4. Xeon 6300P (LCC SKU) – 1 Compute Tile + 2 I/O Tiles = Up to 16 Cores
The architectural features of the P-Core series include Redwood Cove cores with hyper-threading capabilities, a generous 2 MB of L2 cache per core, AVX-512, Intel’s AMX technology for vector operations, and an advanced design capable of delivering impressive compute performance.
In contrast, the 6th-Gen Sierra Forest CPUs showcase an assortment of both P-Core and E-Core based SKUs, tailoring to the high-density compute needs and scale-out workloads while aiming to optimize performance per watt. The Xeon 670qE E-Core lineup is designed to significantly upgrade AI data centers currently running on older platforms, offering improved performance with minimal increase in power consumption.
The Sierra Forest series is built on Intel’s advanced 4 process node and uses the Crestmont E-Core architecture, delivering up to 144 cores, substantial CPU TDPs up to 350W, DDR5-6400 and 8000 MT/s MCR DIMM support, as well as numerous PCIe Gen 5.0 lanes and efficient UPI 2.0 links. These processors come in a single compute die package with two I/O dies. Intel has detailed seven SKUs within the Sierra Forest CPU family, featuring core counts ranging up to 144, clock speeds hitting 3.2 GHz, L3 cache sizes between 108-96 MB, and TDPs up to 330W. Performance comparisons highlighted the competitive edge of the Sierra Forest models over AMD’s EPYC Bergamo CPUs, particularly in terms of energy efficiency.
Both the Granite Rapids and Sierra Forest CPUs will be compatible with the LGA 7529 socket platform, which accommodates configurations with substantial TDPs per CPU and supports a 12-channel memory interface.
In addition to data center CPUs, Intel’s announcement included a mention of the upcoming Lunar Lake CPUs, which are anticipated to bring significant advancements to client computing. More news on Intel’s developments is expected to be revealed at their Intel ON event scheduled for September, where launches such as the Arrow Lake-S Desktop CPUs are anticipated.
Overall, Intel’s advances in server processors signal an emerging era of data center technology, promising to satisfy the growing demands for powerful and efficient computing resources across various industries.
