Foveros packaging at its greatest with new P- and E-cores, Intel Thread Director, new NPU, and Arc Xe2 GPU
Intel has shed extra particulars about its Lunar Lake structure in the course of the Tech Tour occasion, together with particulars about all of the enhancements it is going to convey to the brand new Core Extremely cell processors. That is an in depth listing that contains new Lion Cove P-cores, Skymont E-cores, new NPU, the Arc Xe2 Battlemage GPU, and enhancements to the Intel Thread Director, all neatly packaged right into a single-tiled SoC.
What makes the Lunar Lake particular is the truth that it’s manufactured by TSMC, with the compute tile made on TSMC’s N3B node, whereas the SoC tile, or what Intel now calls the “Platform Controller”, is made on TSMC’s N6 node. Intel is utilizing its Foveros packaging expertise to convey all of it collectively together with as much as 32GB of LPDDR5X reminiscence sitting on the identical bundle. This additionally signifies that reminiscence isn’t person upgradable, which we wrote about right here. Intel claims that placing the reminiscence on the identical bundle reduces the PHY energy by 40 %, in addition to will increase space financial savings by 250 mm2.
New Lion Cove P-cores and Skymont E-cores
The Lunar Lake CPU half has a complete of 8 cores, 4 of that are Lion Cove efficiency cores (P-cores) and 4 of that are Skymont effectivity cores (E-cores). Not like earlier Intel processors, this time these cores don’t share an L3 cache or the identical ringbus if you’ll, however slightly simply share the identical die.
The P-cores, though meant for efficiency, are additionally effectivity optimized and don’t function Hyper-Threading expertise (HTT), as Intel claims that this elevated efficiency per energy by 5 %, decreased the efficiency per space by 15 %, and elevated efficiency per energy per space by 15 %. The final improve in IPC is 14 % in comparison with the Redwood P-cores in Meteor Lake, with vital energy effectivity financial savings. The P-cores additionally received a brand new energy administration system with an AI-driven self-tuning controller that reduces energy and clocks through the use of 16.67 MHz intervals, which is way more exact in comparison with 100MHz used earlier than.
On the structure aspect, Intel has fully redesigned the entrance finish with as much as 8x bigger department prediction block, wider fetch, and different enhancements; improved integer and vector execution; improved reminiscence subsystem, and extra. When it comes to numbers, every Lion Cove has 192KB of L1 cache, 2.5MB of L2 cache and people 4 cores share 12MB of L3 cache.
Whereas the P-cores received a major overhaul and first rate IPC enchancment, the star of the present are the Skymont E-cores, which convey a powerful 68 % IPC improve in comparison with the Meteor Lake E-cores, no less than in floating level efficiency. The integer IPC acquire is up by 38 % as nicely. This interprets to a 300 % efficiency to energy improve on some degree. The Skymont E-cores additionally received overhauled with wider machine predict and decode, a brand new out-of-order engine, deeper queueing extra dispatch ports, increased vector efficiency, and reminiscence subsystem enhancements. In numbers, these 4 E-cores share 4MB of L2 cache and Intel managed to double the L2 bandwidth and improve L1 to L1 transfers amongst cores.
Intel Thread Director is the important thing to getting all of it proper
Intel has additionally made enhancements to its Thread Director, one thing that may be a should so as to use the proper CPU cores for every processing load and migrate it appropriately. Intel has improved OS and OEM integration, gave the algorithm finer granularity, launched OS containment zones, and a brand new dynamic scheduling coverage.
Arc Xe2 Battlemage GPU will increase efficiency by 50 %
Intel’s Lunar Lake additionally brings the brand new Xe2 Battlemage structure, promising a 50 % enchancment in gaming efficiency in comparison with the Xe-LPG GPU within the Meteor Lake. The GPU core has eight 512-bit vector engines, eight 2048-bit XMX matrix engines, and has an elevated 192KB L1 cache per SLM. Those self same Matrix Extension Engines (XMXs) the iGPU with 8 GPU cores get 1,024 unified shaders and put up 67 TOPS of AI efficiency. It additionally means it packs 8 next-generation ray tracing items, which places it consistent with DirectX 12 Final necessities.
The GPU additionally will get a brand new media engine with {hardware} acceleration for AV1 encoding and decoding, assist for VVC {hardware} decoding, and extra, whereas the show engine enhancements convey assist for as much as 3 shows, assist for the brand new eDisplayPort 1.5 commonplace, along with the DisplayPort 2.1 and HDMI 2.1, and extra.
The Intel NPU 4 will get Intel onto the Microsoft Copilot+ practice
Though that is the second-generation NPU, Intel calls it the NPU 4, however that doesn’t imply that it doesn’t get numerous enhancements, together with as much as 4 instances increased AI inferencing efficiency in comparison with the Meteor Lake NPU, pushing it up from 12 TOPS to 48 TOPS. Which means Intel is safely above the 40 TOPS requirement for Microsoft Copilot+.
Connectivity aspect will get Thunderbolt 4 and extra
The Platform Controller or the I/O, which is constructed on TSMC’s 6nm N6 course of, contains an built-in Thunderbolt 4 controller with assist fo as much as three 40Gbps ports. It additionally offers Lunar Lake 8 PCIe lanes, largely used for Gen 5 SSDs and different platform connectivity. The I/O additionally will get WiFi 7, Bluetooth 5.4, and different connectivity choices.
Intel managed to just about shock everybody with its Lunar Lake structure and it’ll definitely put numerous stress on the likes of Apple and Qualcomm with M-series and Snapdragon X Elite chips. It targets the identical ultra-portable and thin-and-light units, and whereas each of those have an honest headstart, Intel might have an edge with its new E- and P-cores, new NPU, and particularly the brand new Xe2 iGPU.
Sadly for Intel, Lunar Lake is at the moment on observe to launch in Q3 2024. After all, that timetable nonetheless places it in time for the vacation shopping for season, and Intel claims it is going to energy greater than 80 completely different AI PC designs from 20 unique tools producers (OEMs), with Intel anticipating to deploy greater than 40 million Core Extremely processors out there this 12 months.
