Intel announced its next-gen highly efficient mobile architecture Lunar Lake with the ultimate goal of scaling up to the AI optimised PCs. Intel’s announcement of Lunar Lake at the Computex comes at a very proactive time. Sitting in front of the new AI generation processor was indeed futuristic and exciting. Image credit: Intel.The announcement at the Computex marked the next big step forward in Intel’s PC strategy shift to AI-centric computing aimed at championing efficiency, computation, graphics and, above all, AI performance. Intel’s Lunar Lake is expected to go into the market in Q3 2024. Intel is aiming Lunar Lake to compete with AMD’s upcoming Raphael and Qualcomm’s future Snapdragon chip technologies for the AI PC market.
It’s a brand-new microarchitecture, very much from the ground up: Lunar Lake is not the successor to Meteor Lake, but to the one after that. That means new P-cores and E-cores, a brand-new Battlemage GPU, an enhanced NPU, and the most radical innovation Intel has ever baked into an architecture – a new generation of embedded LPDDR5X memory, available with up to 32GB of capacity. Intel will be using TSMC for the first time: the 3nm process will be used to manufacture the CPU’s compute tile.
Whereas TSMC’s N3B node is used for the compute tile, just like the A17 Pro SoC in the iPhone 15 Pro, Lunar Lake has a simplified two-tile design (compute and ‘platform controller’) compared with the four-tile approach of Sapphire Rapids. Intel wants to go as fast as it can, and save as much power as possible.
You can see this in Lunar Lake’s 4+4 core configuration: two lions and two skymonts. Intel has gone all-in here on achieving unmatched efficiency – skymonts have become so efficient that they’ve thrown hyper-threading out the window. Intel slew the thread. Two reasons for this: one, they’re betting on E-cores being so efficient that they don’t need hyper-threading. The second is that they’re retiring hyper-threading alongside the legacy code it encouraged. Intel has long been a champion of pro efficiency, but Lunar Lake’s four Lion Cove P-cores are designed for 14 per cent average IPC improvements over the last iteration, a jump that will rival AMD’s Zen 5 architecture.
Intel’s AI-powered Xe2 (ie, its leap in graphics) is known as Battlemage 67 TOPS AI performance, 50 per cent increase in raster performance over its predecessor, Alchemist. Battlemage scores ‘a 67 TOPS of AI performance’ Lunar Lake is poised to be a superb platform for both visual and complex AI tasks. Intel’s rivals, meanwhile, are also going all-in on AI PCs. On 7 December, Nvidia’s CEO Jensen Huang showed off a concept PC with its next-gen GeForce RTX4070 graphics chip, available in 2023.
Intel makes sure Lunar Lake has lots of enterprise connectivity up its sleeves, meaning each system has at least two Thunderbolt 4 ports and Wi-Fi 7 and Thunderbolt Share as standard, as well as full PCIe Gen 5 and Gen 4 lane support. The ‘Platform Controller Tile’, which is really the chipset, is Intel’s nod to the connected future.
But it’s also evident that Lunar Lake is tuned for performance in a specific form factor, overall suited to low-power laptops. Intel hasn’t leaked any TDP numbers. But the key enhancements in Lunar Lake contest directly with the upcoming M3-powered MacBook Air, with Qualcomm’s newest CPUs, and with AMD’s forthcoming Ryzen AI 9 CPUs, all in the same performance envelope of 30 W TDP. Lunar Lake marks a major inflection point in Intel’s mobile strategy, and a monumental moment in the fledgling category of the AI PC.
It’s in the name ‘pro’, in that professional-grade level of innovation, efficiency and performance with which Intel is delivering its architecture, as well as the core, and the pro graphics with Battlemage and more to come to Lunar Lake. And then also the pro connectivity so that it creates a super-fast, super-high-speed communication fabric to enable the fast, modern, AI-centric computing environments. So this is ‘pro’, Intel on the AI PC frontier. This is Intel and Lunar Lake.
© 2024 UC Technology Inc . All Rights Reserved.