Intel presented its new mobile architecture, Lunar Lake, at Computex. The company is targeting AI-enabled PCs with Lunar Lake, which is packed with new technology aimed at increasing AI efficiency, computing power, graphics, and performance. Intel says Lunar Lake will arrive in the third quarter, but is not talking about specific CPU SKUs at this time. Instead, it is revealing its architectural details as it prepares to battle AMD and Qualcomm for bragging rights in the AI PC market.
Lunar Lake has a completely new microarchitecture Compared to its predecessor Meteor Lake, with new P cores, E cores, Battlemage graphics, a new NPU and a big first for Intel.integrated LPDDR5X memory in 16 GB or 32 GB configurations. Another big news is that Intel is using TSMC's 3nm process for the CPU compute tile, marking the first time it has used its biggest foundry rival to make a chip instead of its own factories. Still, Intel handled the packaging and architecture design.
The compute tile is manufactured using TSMC's N3B node, the company's first-generation 3nm process and the same node used by Apple for the A17 Pro SoC in the iPhone 15 Pro. There is also a “chipset” tile. Made of 6nm TSMC. Yes, there is only two tiles in Lunar Lake, one compute tile and one “platform controller” tile, instead of the four tiles it used previously (compute, SoC, I/O, and graphics).
Lunar Lake is practically new from the ground up and offers a considerable jump in performance and efficiency over Meteor Lake.
Credit: Intel
Lunar Lake will have a 4+4 core configuration, which confirms previous rumors. There will be four Lion Cove performance cores and four Skymont efficiency cores, and hyper-threading has been officially removed, indicating that it will likely never appear in any Intel CPUs again. The gist is that Intel believes it's unnecessary given how good its E-cores have become, and running two threads per core requires additional power, hardware, space, and logic, so it's now considered an unnecessary waste of resources. The “low power island”, composed of two low power cores in Meteor Lake, also disappeared. Its functions have now been incorporated into the Skymont E-cores.
There are only two tiles here, along with a filler tile in the corner. The GPU, NPU and CPU are now part of the main tile.
Credit: Intel
The Lion Cove's P-cores have been redesigned to be wider than their predecessors, allowing them to process more data despite slightly lower clock speeds for this mobile platform. Intel says that on average the instructions per clock advantage here over Meteor Lake P cores is 14%, although it can be as high as 20% in some applications. This is in the same ballpark as AMD's Zen 5 announcement from yesterday, which indicated that the IPC increase from Zen 4 to Zen 5 is 16% on average.
The Skymont E-cores have also been radically improved, with a 68% improvement in single-threaded floating-point performance compared to those in Meteor Lake. Intel has also expanded these E-cores, added more cache, and made the queues deeper. Multi-threaded performance is also improved over the Crestmont over a wider range of frequencies and TDPs.
Both P-Cores and E-Cores are completely new to Lunar Lake and offer (supposedly) revolutionary efficiency without hyper-threading.
Credit: Intel
Equally important news is that Lunar Lake features Battlemage graphics, also known as Xe2 graphics. Overall, Intel claims that it is 50% faster than the Arc Alchemist GPU on Meteor Lake while also functioning as an AI GPU that delivers up to 67 TOPS of AI performance. This Battlemage chip also supports ray tracing and DisplayPort and HDMI 2.1 connections.
Intel claims that Xe2 graphics on Lunar Lake offer a 50% increase in raster performance over its Alchemist predecessor.
Credit: Intel
There's also a revamped neural processing unit, or NPU, which Intel says is now its fourth-generation NPU. Intel also claims to have seen a big jump in performance, from ~11 TOPS for the NPU on Meteor Lake to 48 TOPS for Lunar Lake. That's not as high as AMD claims with its 50 TOPS All-new Ryzen AI 9 mobile CPUs, but it is higher than Qualcomm's 45 TOPS with its X Elite SOC. It will pave the way for Intel to finally be included in the new wave of AI PCsas the initial release only included Qualcomm-powered machines.
Intel's claim of 48 TOPS for its NPU will qualify it for AI PC tasks, as we believe the threshold is 45 TOPS.
Credit: Intel
Finally, connectivity has been improved by adding a minimum of two Thunderbolt 4 ports to each Lunar Lake system, although the specific OEM can add a third if desired. It also supports Wi-Fi 7 and Share lightning and will offer four PCIe Gen 5 lanes and four PCIe Gen 4 lanes. All of this hardware is contained within the platform controller tile, which is basically the chipset.
This block diagram shows the location of the various components in the package, and it's notable how small the E-cores have become.
Credit: Intel
Overall, Lunar Lake sounds like a big step forward for Intel's mobile strategy, but remember, this is only for low-power laptops. We didn't see any TDP numbers in the countless presentations on this topic. Still, previous leaks mentioned that it maxed out at 30W, so it will compete directly with the M3-powered MacBook Air, Qualcomm's X Elite and X Plus CPUs, and AMD's new Ryzen AI 9 CPUs.
Leave feedback about this