Intel Changes Mobile PC Architecture With Lunar Lake AI SoC
How does the design of Lunar Lake differ from its predecessor, Meteor Lake, in terms of die configuration and manufacturing process?
The design of Lunar Lake differs from its predecessor, Meteor Lake, in several ways. Firstly, Lunar Lake has a two-die/tile configuration, whereas Meteor Lake had four tiles. The Lunar Lake SoCs feature a compute tile and an I/O tile, both manufactured by TSMC. The compute tile is built on the N3 process node, and the I/O tile is built on the N6 process. This shift to TSMC's process technology puts Intel on par with AMD in terms of manufacturing process.
In contrast, Meteor Lake had a more complex design, with separate tiles for low-power, compute, GPU, and I/O functions. The move to a two-tile configuration in Lunar Lake simplifies the design and enables more efficient power management, as all processing functions are consolidated onto the compute tile. This includes the low-power island for the E-cores and NPU.
Furthermore, Lunar Lake marks a shift in how future Intel PC processors will be designed, with significant enhancements in performance, efficiency, and architecture over Meteor Lake. Lunar Lake SoCs will feature eight CPU cores, including four performance (P-cores) and four efficiency (E-cores) cores, compared to the 12 cores used in Meteor Lake. The new cores in Lunar Lake are more optimized and efficient, with the E-cores seeing a significant performance boost while remaining small and efficient.
In summary, Lunar Lake differs from Meteor Lake in its die configuration, manufacturing process, and overall design approach, with a focus on improved performance, efficiency, and power management.
Can you describe the changes in the GPU architecture from Meteor Lake to Lunar Lake and the expected performance improvements?
The transition from Meteor Lake to Lunar Lake brings significant changes in the GPU architecture. Lunar Lake introduces the new Xe2 graphics architecture, which delivers a substantial 50% increase in overall GPU performance compared to Meteor Lake4. The Lunar Lake GPU features eight Xe2 cores, each with 64 vector engines, support for Intel's 16-bit SIMD instructions, XMX vector instructions, and data types ranging from INT2 to FP32.
In addition to the core architecture improvements, the Lunar Lake GPU includes an enhanced media processing core and a new display processing core. These updates contribute to a more efficient and powerful graphics processing unit, capable of handling a wider range of workloads.
One notable aspect of the Xe2 architecture is its focus on AI performance efficiency. The Lunar Lake GPU supports 67 TOPS of AI performance, a significant increase from its predecessor4. This improvement in AI performance is crucial as more applications begin to leverage AI for various tasks, including generative AI large language models (LLMs).
In summary, the GPU architecture changes from Meteor Lake to Lunar Lake involve the introduction of the Xe2 graphics architecture, enhancements to the media and display processing cores, and a focus on AI performance efficiency4. These updates result in a 50% increase in overall GPU performance and a substantial boost in AI capabilities.
What improvements have been made in the CPU cores of Lunar Lake compared to Meteor Lake?
The CPU cores of Lunar Lake have seen significant improvements compared to Meteor Lake. Both the performance (P-cores) and efficiency (E-cores) cores have been enhanced.
For the P-cores, Intel has removed hyperthreading, which was limiting single-thread performance and required more die area. This change improves performance and efficiency, with the new smaller, more optimized, and higher frequency P-cores providing 15% more performance/watt, 10% more performance/die area, and 30% more performance/power/area. Additionally, the P-core was rearchitected from the ground up to be more efficient and performant, resulting in a 14% IPC (instructions per cycle) improvement from the previous generation P-core used in Meteor Lake.
As for the E-cores, Intel has increased everything to make it perform better while keeping it small and efficient. The E-core has been scaled up from a limited function CPU core to a full CPU core with a broader frequency range and paired with L2 cache that doubled in bandwidth and size to 4MB. This results in a 38% improvement in integer processing and a 68% improvement in floating-point processing over the Meteor Lake E-core.
Overall, the CPU core improvements in Lunar Lake showcase Intel's commitment to enhancing performance and efficiency in their processors.