CPU, GPU & AI Leadership Over Qualcomm & AMD, Much Better Battery Life

A new beast is born at Intel in the form of Lunar Lake CPUs, powering the Core Ultra 200V lineup on thin & light platforms and bringing CPU, GPU, and AI leadership with leading performance, and efficiency versus Qualcomm and AMD offerings.

Intel Core Ultra 200V “Lunar Lake” CPUs Bring A Historical Leap To Thin & Light Platforms, Brand New CPU, GPU, AI Cores With Leadership Performance & Efficiency

In our deep dive back at Computex 2024, we detailed the chip itself and its newest Xe2 GPU, both of which are major upgrades over Meteor Lake. Today, Intel is officially lifting the curtains off of the Lunar Lake lineup which is called Core Ultra Processors Series 2 or Core Ultra 200V. Lunar Lake is designed first and primarily for thin and light platforms so let’s start with what the chips have on offer and how they compare against the competition as per the official results.

Thin & Light is a tricky segment, you need to have a chip designed to meet the low-power demand while keeping strong performance and efficiency. You can scale certain chips down to lower TDPs but Intel went a different route. Instead of taking this approach, Intel made a dedicated chip called Lunar Lake which embeds several new technologies that target several segments including gaming, creation, and AI PCs.

For efficiency, Intel made several key energy innovations for Lunar Lake which will play a crucial role for next-gen thin and light platforms. These include:

• Memory on package – Up To 40% reduced PHY power
• Power Management enhancement and integration through Thread Director
• Improved E-Core clusters
• New power delivery architecture for better control, enhanced power utilization
• New 8 MB Memory side cache – Efficiently feeding memory-hungry IPs

Tuning The Cores For Maximum Efficiency & Power Savings

Intel’s Core Ultra 200V “Lunar Lake” CPUs leverage two types of core architectures. There are the Lion Cove P-Cores which are tuned towards performance and then there are the Skymont E-Cores which are tuned towards efficiency. For Lunar Lake, Intel is using Low-Power variants of these E-Cores called LP-E which sit on a dedicated low-power island.

This is an evolution over the Meteor Lake CPUs which came with 2 Crestmont LP-E cores with 2 MB of shared L2 cache. For Lunar Lake, Intel has doubled the LP-E core count to 4 while using the next-gen architecture which brings major IPC uplifts, and also increased the cache to 4 MB (shared L2).

Intel then uses its various power management techniques including Thread Director to deliver big efficiency gains at low-power using this low-power island. In comparison to Qualcomm’s X1E80100 and its own Meteor Lake Core Ultra 7 165H, the Lunar Lake CPUs deliver a 7% gain in performance while offering up to 2.29x improvement in perf/watt. Lunar Lake showcases almost half the power (package) versus Meteor Lake.

These efficiency gains aren’t limited to just the CPU cores as Core Ultra 200V “Lunar Lake” CPUs with Xe2 graphics architecture bring up to 2x the performance per watt improvement in a range of titles. In games like Cyberpunk 2077, the GPU onboard the Lunar Lake SoC runs 44% faster & sips at 22% lower power.

Overall, the Lunar Lake CPUs use up to 50% lower package power than Meteor Lake and that’s while featuring 32 GB of LPDDR5X-8533 memory onboard. The package power includes the LP5X memory so that’s an astonishing leap within one generation.

Blue Team Tackles Qualcomm’s & AMD’s Battery Life Performance

All of these efficiency tuning should make battery times great on Core Ultra 200V “Lunar Lake” PCs and that’s what Intel showcases in its next slides which compare three systems from the same OEM, using 14-16″ chassis, 1080p displays, and a ~75Whr battery against each other in various workloads such as UL Procyon Office and Microsoft Teams 3×3.

Here, the Intel Lunar Lake CPUs sit ahead with up to 14 hours of battery life in UL Pryocon and up to 9.9 hours of battery life in Teams 3×3. That’s a 39% improvement over a system with the AMD Ryzen AI 9 HX 370 and a 47% improvement over a system with Qualcomm’s X1E-78-100 chip.

Intel also showed another battery life comparison using the same OEM and the same chassis. One is running its Core Ultra 7 268V SKU and the other is running the Qualcomm X1E-80-100. The Lunar Lake chips offer up to 20.1 hours of battery life in UL Procyon which is an improvement of up to 9.2%.

Intel Lion Cove P-Cores “The Fastest Cores. Period.”

While we have discussed the architecture of the new Lion Cove P-Cores and Skymont E-Cores in the past, Intel dives deeper into the performance aspects of each respective architecture.

Lunar Lake is adopting an all-new low-latency fabric which provides the following inter-core and DRAM latency figures:

• E-Core to E-Core = ~23ns
• P-Core To P-Core = ~26ns
• E-Core To P-Core = ~55ns
• DRAM Latency = ~90ns

We also detailed the inter-core latency numbers from Hot Chips a few days ago but the most interesting highlight is that Lunar Lake SoCs offer 40% reduced DRAM latency versus Meteor Lake and 30% reduced DRAM latency versus AMD’s Strix Point offerings. This is thanks to the on-package memory rather than relying on off-package DRAM.

The inter-core latency being robust is essential for Core Ultra 200V CPUs as the way scheduling works is through a dynamic method that prioritizes the first single E-Core as long as it fits the work and then expands out to other E-Cores when multi-threaded performance is needed. The scheduler will then move to P-Cores based on higher performance demands.

So in terms of performance, Intel showcases the performance of its top Core Ultra 9 288V against AMD’s Ryzen AI 9 HX 370 and Qualcomm’s X1E-80-100/X1E-78-100 offerings. Here, the Lion Cove P-Core stands out with up to 61% increase over the Qualcomm offering in Specrate 2017 (INT), up to 21% faster in Geekbench 6.3 1T, and up to 20% faster in Cinebench 2024 1T.

The company also claims a 3x performance per thread improvement for Lunar Lake versus Meteor Lake CPUs and you can see the benefits at lower TDPs such as 17W which will be the baseline for most Lunar Lake platforms.

The 8-Core design offers leadership perf/watt at each respective TDP limit, matching Qualcomm in performance with 40% lower power. Across various productivity applications, Lunar Lake either sits on par or offers almost double the performance improvement.

Xe2 Puts Adreno In The Rearview, Faster Than RDNA 3.5 In Double-Digits Range

Xe2 is a big component of the Core Ultra 200V “Lunar Lake” CPUs. This new Intel GPU architecture is shaping up to be a major deal, not just for the integrated graphics segment but also for discrete graphics which will come later in the form of “Battlemage” in the Arc lineup. The company has so far stated a 50% performance increase, much better ray tracing units, and better utilization of the GPU cores versus Alchemist Xe-LPG offerings.

Intel kicks off the comparisons against its own Xe1 iGPUs on the Core Ultra 7 155H and compares it to the Core Ultra 9 288V across a wide variety of games. Do keep in mind that both architectures share the same drivers and here, Xe2 offers a 31% uplift which is over 60-70% in many instances.

The next comparison is a brutal one. Against Qualcomm’s Adreno GPU featured on the X1E-84-100, the Lunar Lake Arc 140V iGPU offers a 68% performance increase which is over 2x in several games but what’s more interesting is that the Adreno GPU failed to even run 23 of the games in the full test suite which shows one of the key areas that Qualcomm’s AI PC platforms lacks within. Just a few days ago, we posted the first benchmarks of an upcoming 8-core Qualcomm SOC which offers miserably low gaming and graphics performance and things aren’t going to look great once Xe2 starts rolling out.

The main comparison is against today’s fastest iGPU solution which is undoubtedly the Radeon 800M series from AMD based on its RDNA 3.5 graphics architecture. This architecture has shown great gaming performance and efficiency across a range of tests and Intel claims that its Xe2 GPU sits 16% faster than the competing solution. In many games, the Arc 140V and Radeon 890M iGPUs were on par with the RDNA 3.5 chip even exceeding it in some instances but RDNA 3.5 has 12 CUs versus 8 Xe2 cores on Lunar Lake chips at max.

Although both of these architectures are very different from one another, the Lunar Lake CPUs could’ve integrated a few more Xe2 cores like 10 or 12 to really solidify its graphics leadership but still, this is very good for a thin and light chip.

On the software side, Intel’s graphics division has done a phenomenal job and at launch, Xe2 will take advantage of over 120 games that are fused with XeSS support and this time, the iGPU will enable faster performance and upscaling quality thanks to the XMX hardware featured on the SoC versus DP4a units on previous architectures such as Meteor Lake.

With upscaling applied (Performance Mode), you can get up to a 62% performance uplift and if you already have over 80-100 FPS, then you can go for the “Quality Mode” preset and use that extra FPS headroom for better higher fidelity.

Ray Tracing is another big thing with Xe2 as we mentioned at the start and here, Xe2 cruises ahead of the competition, delivering up to 30% faster RT performance and 99th percentile over 30 FPS. It is mentioned that the Xe2 RT engine is a fully functionalRT implementation that runs via DirectX12 Ultimate API.

Upgraded Media Engines With AV1, H.265 & VVC Support

The Core Ultra 200V SoCs include a new Media Engine which is part of the Xe2 iGPU. This new engine supports up to 8k60 10-bit HDR decide and encode, AVC, VP9, H.265 HEVC, AV1 (Encode / Decode) and VVC Decode.

This solution provides strong transcode performance against the rivaling chips with better encode and decode engines, leading to faster media and higher-quality media playback.

AI AI AI, The Most AI TOPS On A Thin & Light SoC

Intel is offering the most AI TOPS with its Lunar Lake “Core Ultra 200V” CPUs rated at 120 TOPS versus 80 TOPS on Strix Point and Qualcomm offering up to 45 TOPs with its NPU. The Lunar Lake SOC comes with three AI engines with the GPU offering the highest amount of TOPS rated at 67, the NPU offering 48 TOPS and the CPU offering an additional 5 TOPs. These three AI engines are essential for diverse workloads.

The GPU is useful for gaming and creator AI workloads, the NPU is useful for AI assistants and creation workloads and the CPU is good enough for Light AI workloads. You can also use the combination of all three engines through Lunar Lake-optimized software platforms. So how is the AI performance looking on Lunar Lake?

In Stable Diffusion, the Core Ultra 200V SoCs complete a 20 iteration run in 3.89 seconds on the GPU whereas the Qualcomm SoCs feature no AI acceleration via the GPU. Comparing the NPU to NPU, the Lunar Lake SoC completes the same run in 5.28 seconds versus 5.89 seconds on the Qualcomm SOC.

In UL Procyon AI compute vision, the Lunar Lake SoC once again leads with Qualcomm and AMD failing to run the test due to insufficient support for data types or software. In Image Generation, the Lunar Lake SoC offers over 2x the FP16 performance against AMD’s Strix Point GPU.

Putting everything into perspective within the Geekbench AI suite, the Intel Core Ultra 200V drives the fastest AI performance, delivering substantial gains with multi-engine and framework leader-ship. Across a range of AI applications, the Lunar Lake SoC stands out with 58% better AI compute capabilities.

Intel is all in on the “AI PC” ecosystem and has 300+ ISV features along with solid compatibility for various software applications. This software enablement scales across multiple operating systems, and 100s of applications and is the hard work of 1000s of developers who will be fine-tuning Lunar Lake and taking these features to the next generation. Some noticeable improvements for apps include:

• +86% uplift versus Qualcomm X1E-78-100 in Adobe Photoroom
• +76% uplift versus Intel Core Ultra 7 155H in Adobe Photoroom
• +145% uplift versus Qualcomm X1E-78-100 in Adobe Lightroom
• +79% uplift versus Intel Core Ultra 7 155H in Adobe Lightroom

Intel will also be utilizing its OpenVINO framework for its Lunar Lake NPU & GPU Engines in CANVI which will be available via an update in November while MAGIX will be utilizing the ONNX-RT with OpenVINO EP framework for AI-based video edition (GPU) with the update rolling out in October. TREND will also leverage the OpenVINO framework (NPU & GPU) for data protection services via AI and the update is planned for October. These are some of the few examples of how Intel is working with ISVs and software developers to utilize its AI engines featured on modern “AI PC” SoCs such as Lunar Lake.

Core Ultra 200V SKUs & Platform Features

In terms of SKUs, the Intel Lunar Lake “Core Ultra 200V” lineup will feature a total of nine unique chips. All chips come with the same 4+4 (P-Core + LPE Core) configuration but use different clock speeds, different GPU configs/clocks, and TDPs.

The flagship is the Core Ultra 9 288V which comes with 4 P-Cores, and 4 LP-E cores, in an 8-core and 8-thread configuration. It rocks a base clock of 3.7 GHz and a boost clock of 5.1 GHz with 12 MB of L3 cache. The chip features the Intel Arc 140V iGPU with 8 Xe2 cores clocked at 2.05 GHz and there are 6 Gen4 Neural Compute Engines as part of the NPU. The chip packs 32 GB of LPDDR5X-8533 MT/s on-package memory and has a base power of 30W (min:17W) that goes up to 37W at MTP.

The rest of the SKUs come with a mix of 16 GB and 32 GB (Dual-Channel) LPDDR5X-8533 on-package memory and the entry-level SKUs feature 7 Xe2 cores within the Arc 130V iGPU. The entry-level SKUs feature a base TDP of 17W that can be configured down to 8W but retain the 37W MTP rating. The NPU is also scaled down on the entry-level SKUs to 5x Gen4 NCEs.

All CPUs come with 8 MB of Memory side cache, feature eDP1.5, DP2.1, and HDMI 2.1 outputs, offer integrated WiFi7 (5Gig), 3x integrated Thunderbolt 4 ports, Bluetooth 5.4 support, LE Audio and GbE LAN (1219-LM). The I/O side includes 5 PCIe 4.0, 4 PCIe 5.0 lanes, 6 USB 3.2 outputs, 2 USB 3 outputs, eSPI, SPI, MIPI, and CSI.

The Intel Core Ultra 200V “Lunar Lake” CPUs will be available globally on the 24th of September in more than 80 consumer designs from more than 20 different manufacturers including Acer, ASUS, Dell, HP, Lenovo, MSI, LG, and Samsung. Pre-orders on these laptops begin today and all platforms will be eligible to receive the Copilot+ PC features through a free update starting in November.

The main highlights of the EVO platform include:• Cooler, quieter performance, and responsiveness in ultra-thin designs.

• Long real-world battery life.
• Built-in security that helps prevent malware attacks and minimizes vulnerabilities.
• Built-in Intel Arc graphics for accelerated creation and smoother game play, even while on the go.
• Lightning-fast connectivity with Intel® Wi-Fi 7 (5 Gig).
• Ability to connect PC to multiple monitors, transfer files, and charge PC with Thunderbolt Share.
• Instant wake and fast charging.
• The highest sustainability standards with EPEAT Gold certification.

Intel Lunar Lake “Core Ultra 200V” CPU Lineup:

SKU Name	Cores / Threads	Cache (LLC)	P-Core / E-Core Boost	GPU (Max Clock)	PL1/Min/MTP	Memory Configuration	NPU / XMX (GPU) TOPs
Core Ultra 9 288V	8/8	12 MB	5.1 / 3.7 GHz	Arc 140V @ 2.05 GHz	30/17/37W	32 GB (2R) LPDDR5X	48/67
Core Ultra 7 268V	8/8	12 MB	5.0 / 3.7 GHz	Arc 140V @ 2.00 GHz	17/8/37W	32 GB (2R) LPDDR5X	48/66
Core Ultra 7 266V	8/8	12 MB	5.0 / 3.7 GHz	Arc 140V @ 2.00 GHz	17/8/37W	16 GB (1R) LPDDR5X	48/66
Core Ultra 7 258V	8/8	12 MB	4.8 / 3.7 GHz	Arc 140V @ 1.95 GHz	17/8/37W	32 GB (2R) LPDDR5X	47/64
Core Ultra 7 256V	8/8	12 MB	4.8 / 3.7 GHz	Arc 140V @ 1.95 GHz	17/8/37W	16 GB (1R) LPDDR5X	47/64
Core Ultra 5 238V	8/8	8 MB	4.7 / 3.5 GHz	Arc 130V @ 1.85 GHz	17/8/37W	32 GB (2R) LPDDR5X	40/53
Core Ultra 5 236V	8/8	8 MB	4.7 / 3.5 GHz	Arc 130V @ 1.85 GHz	17/8/37W	16 GB (1R) LPDDR5X	40/53
Core Ultra 5 228V	8/8	8 MB	4.5 / 3.5 GHz	Arc 130V @ 1.85 GHz	17/8/37W	32 GB (2R) LPDDR5X	40/53
Core Ultra 5 226V	8/8	8 MB	4.5 / 3.5 GHz	Arc 130V @ 1.85 GHz	17/8/37W	16 GB (1R) LPDDR5X	40/53

Intel Lunar Lake “Core Ultra 200V” SoC Platform: