Lunar Lake: Intel relies on TSMC for new SoCs based on hybrid architecture

Lunar Lake: Intel relies on TSMC for new SoCs based on hybrid architecture

Intel has officially taken the wraps off its new architecture Lunar Lake, in the starting blocks to support the operation of next-generation processors. The chips intended for the “mobile” market which will be launched on the market around next September-October will thus be able to benefit from significant improvements. Intel’s stated goal is to redefine expectations for i PC based on x86-64ensuring the perfect combination of power and performance.

Lunar Lake: how Intel’s new SoCs are structured

Nowadays we are currently talking about SoC (System-on-a-Chip) also at Intel. Lunar Lake chips consist of two tile: one deals with actual computing, hosting CPU, GPU and NPU; the second manages the external I/O interfaces. For the production of the two elements, Intel relies on TSMC and, respectively, its 3 nm N3B and N6 construction processes.

And tile basic passive type Foverosfacilitates communications between the other components and represents the only “piece of silicon” physically produced by Intel in the case of the new Lunar Lake SoCs.

Intel Lunar Lake architecture

The Lunar Lake CPU integrates four core P high performance and four core E highly energy efficient in the top of the range configuration. The cores P Lion Cove promise a gain in terms of IPC (instructions per clock cycle) of 14%, the E Skymont offer quantifiable improvements of up to 68%.

The graphics section integrated benefits from the new GPU Xe2, which – according to Intel technicians – should lead to a 50% increase in performance compared to the previous generation. To complete the configuration, an Intel NPU (Neural Processing Unit) 4.0, called upon to optimize workloads linked to artificial intelligence applications and capable of expressing up to 48 TOPS.

However, considering the overall power of the chips, the technicians of the Santa Clara company speak of 120 TOPS (calculating the combined contribution of CPU, GPU and NPU): a value well above the 40 TOPS requirement requested by Microsoft for AI PC.

Goodbye to traditional Hyperthreading

In a move that caused astonishment, Intel engineers opted to remove the historic functionality of Hyperthreading from the new Lunar Lake. According to the company led by Pat Gelsinger, in fact, Hyperthreading – which normally improves the IPC value by 30% in workload in multi-threaded – not so relevant when aiming for a design ibrido like the one implemented in modern Lunar Lake.

The highly energy-efficient E cores, in fact, have already proven themselves to be optimal, leading to a 40% reduction in consumption compared to Meteor Lake.

Lunar Lake processor performance

Field tests suggest that the removal of Hyperthreading has significantly improved both performance and efficiency.

Lunar Lake introduce numerose architectural optimizations: the mechanism of branch prediction has been significantly improved and extended, Lunar Lake has more bandwidth available for the fetching instructions and are able to decode them more quickly. The execution engine out-of-order integrates a series of measures to increase its flexibility. The memory sees the addition of a new one L0 cache 192KB between L1 and L2, with the latter growing to 2.5MB.

Intel prepares to repel the attack of competitors, called Apple M3, Qualcomm Snapdragon X Elite and AMD Ryzen AI

We have already highlighted how Intel has chosen to adopt i construction processes by TSMC because, evidently, recognized as the best available during the design phase of the new Lunar Lake. This is an important innovation: if on the one hand Gelsinger has demonstrated his desire to strengthen Intel’s production plants and to make the factories available for the production of chips on behalf of third parties (program Intel Device Manufacturing 2.0), on the other hand, the Santa Clara company has encountered considerable difficulties precisely on this front. So much so that we have to rely, this time, on the know-how of the Taiwanese TSMC.

Intel’s “smart move” is that the architectural work conducted by its engineers to create the Lunar Lake chips is easily transferable to other production nodes. So, the innovations will certainly be moved into future Intel products made at its own factories.

Intel’s new offering is designed to address the Apple M3the new ones Snapdragon X Elite of Qualcomm and processors AMD Ryzen AI 300. The Santa Clara company will present Lunar Lake in two models, at least initially, but has not yet shared the final specifications for those models.

Intel plans to offer availability of approximately 40 million processors enabled for AI operations by the end of the year.

The Lunar Lake architecture and all of its associated intellectual property represents a clear shift by Intel towards power-focused design to maximize duration not performance from the drums.

The improved design methodology and CPU and GPU microarchitectures will soon also be used in Intel’s other mobile products: think about the offering Panther Lakehave chip Arrow Lake for desktop PCs and processors Xeon 6 per i data center.

Leave a Reply

Your email address will not be published. Required fields are marked *