The Taiwan event allowed us to delve a little deeper into what the company’s new chips will look like. They focus above all on artificial intelligence, not only in Windows the power of games is also highlighted, although we do not know much more in this regard.
The most notable of all is undoubtedly the disappearance of HyperThreading. New Lunar Lakes more or less lose processing threads. Efficient Cores (E-Cores) will begin to perform this task, which should improve overall CPU performance.
Design of the new Intel Lunar Lake
These new processors pack up to seven different elements into one package. We have no less than three types of cores, as well as integrated graphics. In addition, we have a Foveros interconnection base with a platform controller. Finally, in the same package we have the RAM memory.
The bone P cores they become architecture Lion Covewhile the Electronic cores they become architectsa Skymont. Intel implements Thread Directory technology in these new chips, which provides significant performance gains.
Intel says the Lion Cove Cores (P-Cores) offer a 14% IPC improvement over their predecessor. Where this improves the most is in the Skymont Cores (E-Cores) which deliver a 38% integer and 68% floating point performance improvement. Thread Director allows for better power distribution, which also makes this design more efficient.
These processors have what is called NPU4focused on providing energy to Microsoft Copilot+. The NPU 4 offers a total power of 48 TOPS, which is a sharp increase compared to Meteor Lake, which only offered 11 TOPS. Additionally, Intel claims that the entire Lunar Lake SoC (adding CPU+GPU+NPU) can deliver up to 120 TOPS.
Despite everything, Intel has not given the names and specifications of the Lunar Lake processors. Thanks to our colleagues at Wccftech, we have data on the Ultra 7 268V Core, which the company let them see. This processor has 4 P-Cores and 4 E-Cores operating at a base frequency of 2.2 GHz and is complemented by 14 MB of L3 cache and 12 MB of L2 cache. It is complemented by an unspecified Xe2 graphics card and 4 NPU cores.
Intel Xe2 architecture for next-generation graphics
Perhaps the most important aspect is that the naming of these components is simplified for the user. They end with the LP, LPG, HP and HPG ranges; a bunch of acronyms that will be left for internal use only. The highlight is that they achieved an efficiency gain of up to 12.5 times. To achieve this, the Alchemist Xe2 They have been completely redesigned.
What you’re probably most interested in is the inclusion of new ray tracing units (RTU) which offer a significant increase in performance. It also features new vector engines, deeper caches, and an additional set of new features to deliver a huge performance boost.
There iGPU Xe2 It will be integrated for the first time in the new Lunar Lake chips, which will have up to 8 Xe2 cores. Each core will have 8 XMX units and 8 vector units, a load/store unit, a thread marshaling unit and a dedicated L1/L2 cache. Note that each of these four cores constitutes a single render slice.
During an internal presentation, an Xe2 could be seen running the game F1 24 In 1080p resolutionin high graphic quality and with a rate of 60 frames per second defraud XeSS enabled. This is great information, because this game is quite demanding in terms of graphics.
Lunar Lake Xe2 running F1 24 at FHD at 60 FPS using XeSS https://t.co/dM39X9qWN7
June 10, 2024 • 11:02 a.m.
