Intel’s upcoming Lunar Lake processors will no longer feature hyper-threading, a move the company disclosed at Computex 2024. This marks a significant strategic change aimed at improving single-threaded efficiency and overall processor performance. Lunar Lake is the codename for Intel’s next-generation mobile processors.
Boosted Single-Thread Performance
Intel´s Lunar Lake enhances single-thread performance by refining the P-core architecture without using hyper-threading. This new P-core design, dubbed Lion Cove, features 2.5MB of L2 CPU cache per core and up to 12MB of shared L3 cache. E-cores under the Skymont architecture come with 4MB of shared L2 cache. Furthermore, the new L0 data cache reduces data access latencies and speeds up performance.
First seen in Intel’s 12th Gen Alder Lake, the hybrid architecture uses a mix of high-performance cores (P-cores) and efficiency cores (E-cores) to distribute tasks based on workload. The initial setup had some inefficiencies, as operating system schedulers frequently favored physical cores over hyper-threads, despite using Intel Thread Director (ITD).
Big news for the AI PC: Up to 40% improvement in power efficiency for battery life, with a massive leap in graphics. 🔋💻
Meet Lunar Lake, the flagship processor for Intel’s next generation of AI PCs. #Computex2024 https://t.co/kVztM7QIiN pic.twitter.com/H94FZSDVtM
— Intel News (@intelnews) June 4, 2024
Smarter Task Scheduling, Lower Power Consumption
Improvements in the Intel Thread Director allow it to favor E-cores, called Skymont, for task scheduling, resulting in reduced power usage. This adjustment has led to a 35% decrease in power consumption for applications like Microsoft Teams. The updated ITD can single out the most power-efficient CPUs on Lunar Lake platforms, letting Windows OS use these CPUs efficiently while others remain idle.
Memory in Lunar Lake includes LPDDR5X-8500, available in 16GB or 32GB configurations, embedded directly on the chip package. This setup reduces latency and board space, decreasing power consumption of the memory PHY by 40%.
Unveiled previously at Intel Tech Tour 2023 in Taipei, Taiwan, the Lunar Lake architecture primarily targets mobile platforms and is set to launch in Q3. Skymont E-cores see IPC gains of 38% and 68%, while Lion Cove P-cores offer a 14% improvement of Instructions per cycle (IPC). The integrated Xe2 GPU promises a 50% performance boost.
AI Integration and Fabrication Choices
The Lunar Lake lineup includes a Neural Processing Unit (NPU) capable of 48 TOPS in AI performance, surpassing Microsoft’s next-gen AI PC standards. Combined with the CPU and iGPU, the system offers 120 TOPS. Intel has partnered with TSMC, employing its 3nm N3B process for the compute tile and the N6 node for the platform controller tile, while Intel handles the 22FFL Foveros base tile.
Beyond mobile applications, these design improvements will be integrated into desktop Arrow Lake processors and Xeon 6 series. The Lunar Lake’s top model features four P-cores and four E-cores, arranged on a dual-logic-tile architecture connected to a Foveros base tile using a 25-micron bump pitch for better interconnect density and power efficiency.
Last Updated on November 7, 2024 7:54 pm CET
