Lunar Lake: New AI oriented Intel Architecture | Hardware and Software of Computer | Cpu Hardware List | Types of cpu | Turtles AI
Intel unveils Lunar Lake: a revolution for mobile and gaming sectors with enhanced performance and energy efficiency.
Intel revealed its new Lunar Lake architecture during the Intel Tech Tour 2024 in Taipei, Taiwan. This launch precedes the Computex 2024 keynote and anticipates a commercial debut in Q3. Lunar Lake introduces significant improvements in all aspects of chip design, primarily targeting mobile devices, with potential impacts on Arrow Lake processors and gaming CPUs. Every component of the Lunar Lake architecture has been optimized for a refined balance of power and performance, revolutionizing expectations for x86 PCs. Among the most notable improvements are 38% and 68% IPC gains for the new Skymont cores, and a 14% increase for the Lion Cove cores, though with some reservations. Graphics performance sees a 50% improvement thanks to the new integrated Xe2 GPU. Lunar Lake includes a new neural processing unit (NPU) for AI workloads, delivering 48 TOPS of performance, well above Microsoft’s requirements for next-gen AI PCs. Overall, the Lunar Lake platform achieves 120 TOPS considering CPU and GPU contributions. The new design methodology, focusing on maximum energy efficiency, will form the basis for Intel’s future products like Arrow Lake and Panther Lake. Surprisingly, Intel chose TSMC’s 3nm N3B process for the compute tile, which houses the CPU, GPU, and NPU, using TSMC’s N6 node for the platform controller tile. The only Intel-fabbed silicon in the chip is the passive 22FFL Foveros base tile, which facilitates communication between components. This choice was made due to TSMC’s superior nodes at the time of chip design, but Intel plans to return to its own nodes in future products. Lunar Lake processors, with four P-cores and four E-cores, include two logic tiles (N3B for computing and N6 for the controller) and a 22FFL Foveros base tile, with improved interconnections to reduce power consumption. The LPDDR5X-8500 memory is integrated into the chip package, available in 16GB or 32GB configurations, communicating through four 16-bit channels, offering up to 8.5 GT/s throughput per chip.
Component | Specifications |
Core | 4 P-core, 4 E-core |
Memory | LPDDR5X-8500, 16GB or 32GB configurations, 4 16-bit channels, 8.5 GT/s per chip |
Cache | L0 cache 192KB, L2 cache 2.5MB, shared side cache 8MB |
Process Nodes | Compute tile N3B (TSMC), Controller tile N6 (TSMC), Base tile 22FFL (Intel) |
Interconnection | 25-micron bump pitch, PMIC on motherboard |
Energy Efficiency | 40% reduction in consumption compared to Meteor Lake |
Graphics | Xe2 GPU, 50% performance improvement, 8 Xe cores, 64 vector engines, 8 ray tracing units |
The Lunar Lake architecture introduces numerous enhancements, including an optimized power delivery subsystem, now shifted to the motherboard, with four PMICs providing multiple power rails and greater control, reducing SoC power consumption by 40% compared to Meteor Lake. Lunar Lake’s P-cores offer an average 14% IPC gain, although Intel removed hyperthreading, deeming it irrelevant in a hybrid design with more power- and area-efficient E-cores. This decision led to a 15% increase in efficiency, a 10% increase in performance per area, and a 30% improvement in performance per power per area. The new microarchitecture features numerous front-end, branch prediction, cache, and back-end improvements, with a new 192KB L0 cache and an increased 2.5MB L2 cache, enabling IPC enhancements. The more modular and synthesized design allowed for additional configuration options and greater transferability across different process nodes.
The Skymont cores, the third generation of E-cores for x86 hybrid processors, introduce significant improvements, with a 38% IPC gain for integer workloads and a 68% gain for floating-point tasks. Intel has increased the out-of-order allocation, retire, out-of-order window size, and enhanced parallelism, doubling vector and SIMD workload performance. The shared L2 cache has been increased to 4MB, improving bandwidth.
The integrated Xe2 GPU offers up to 1.5 times the graphics performance of Meteor Lake, supporting INT2, INT4, INT8, FP16, and BF16 for AI operations. The graphics pipeline has been improved with higher throughput, new vertex management, and increased ray tracing efficiency. The GPU is modular, scalable, and includes 8 Xe cores, 64 vector engines, 8 ray tracing units, and 8MB of L2 cache. The media engine supports up to 8K60 HDR decoding and encoding, with support for all major standards.
The platform controller tile manages external I/O interfaces, including Wi-Fi, Bluetooth, USB, and Thunderbolt. Intel guarantees at least two Thunderbolt 4 ports for every Lunar Lake laptop, with some variants offering up to three, also supporting the new Thunderbolt Share feature for file and peripheral sharing.
In conclusion, Intel’s Lunar Lake architecture represents a significant advance in processor design, with a strong focus on performance and energy efficiency, supported by a wide range of technical improvements. These advancements not only strengthen Intel’s position in the laptop market but also pave the way for future generations of desktop and server processors.
Highlights
- Energy Efficiency: 40% reduction in power consumption compared to Meteor Lake.
- AI Performance: Total of 120 TOPS, with NPU providing 48 TOPS and Xe2 GPU offering 67 TOPS.
- Integrated GPU: Up to 1.5 times the graphics performance of Meteor Lake.
- Optimized Memory: Integrated LPDDR5X-8500, reducing latency and power consumption.