Lunar Lake First Tests: Intel’s AI-Ready Core Ultra 2 Chip Posts Ntimely 24-Hour Battery Life

We’ve finishured a lengthened directup to Intel begining its Lunar Lake processor platestablish, but laptops with these second-generation Core Ultra chips are finassociate in our hands. And now, we have first-hand carry outance testing results.

The Core Ultra brand—denoting processors with onboard neural processing units (NPUs) for local AI tasks—begined at the finish of 2023 with the first-generation Meteor Lake chips. Lunar Lake processors chase up with greater power and capacity for even more trillions of operations per second (TOPS), which should transpostpoinsist to quicker and more effective AI results.

(Credit: Joseph Maldonado)

Having recently revamped our entire benchlabel suite to alter to up-to-date PCs—including AI laborload testing—we’re in an selectimal place to assess Lunar Lake’s efficiency versus other up-to-date platestablishs. For testing, we used an Asus Zenbook S 14 supplyped with a Core Ultra 7 258V processor, which discleave outed standard-setting battery life and a potentiassociate huge leap in NPU carry outance.

Lunar Lake Has Landed

I won’t spfinish too much time here elucidateing the current CPU landscape or the Lunar Lake architecture, as we’ve extensively covered these topics in the directup to the begin. If you want to get up to speed, check out our coverage of Lunar Lake architecture, Intel’s speed claims, and the first Core Ultra 2 laptop chips. The table below shows off the whole Core Ultra 2 stack.

(Credit: Intel)

As you can see, the Core Ultra 7 258V sits in the upper middle of this batch of chips, with eight cores (four Perestablishance cores, four low-power Efficient cores) and eight threads. As detailed previously, the core types occupy the same chip alengthenedside the NPU, with the low-power E-cores scheduleated for shrink-priority tasks.

AI tasks on Windows are hugely inherently background laborloads that run behind your main task; leank of video call enhancements, audio betterments, and other low-insist asks. The NPU can handle these laborloads, freeing the processor’s CPU to run the main applications (your browser, conferencing software, and other active labor apps) without cataloglessdown. Some conceiveive applications, say for editing, will lean more on the NPU to originate or spotless up images, but the establisher is the primary reason for the multi-unit architecture.

Now Testing: Intel Core Ultra 2 vs. the Field

Let’s see what Lunar Lake can do in retail systems. I put the Zenbook S14 thcimpolite our recently revamped benchlabel suite, which covers a range of productivity, multimedia, and AI laborloads; I’ll elucidate each batch of tests before the results so you have context for the numbers.

Productivity, Content Creation, and AI Tests 

Our primary overall benchlabel, UL’s PCMark 10, puts a system thcimpolite its paces in productivity apps ranging from web browsing to word processing and spreadsheet labor. Its Full System Drive subtest meabraves a PC’s storage thcimpoliteput. Three more tests are CPU-centric or processor-intensive: Cinebench 2024 uses Maxon’s Cinema 4D engine to rfinisher a intricate scene; Geekbench 6.3 Pro by Primate Labs simupostpoinsists famous apps ranging from PDF rfinishering and speech recognition to machine lachieveing; and we see how lengthened it consents the video editing tool HandBrake 1.8 to alter a 12-minute clip from 4K to 1080p resolution. 

Next, PugetBench for Creators rates a PC’s image editing prowess with various automated operations in Adobe Photoshop 25. Finassociate, Geekbench AI and UL’s Procyon Computer Vision test are two of the first AI processing benchlabels. (Our use of these AI tests is still in enhugement and, seeless, cannot be assessd among processing platestablishs. See our page on how we test laptops for more on that.)

Thcimpoliteout the productivity and greeted creation tests, the Core Ultra 7 258V carry outed well assessd with the competition. You’ll see no clear Lunar Lake greaterity on these tests, with the Zenbook descending in line with the others on most of these results. It was unwidespreadly the quickest carry outer, except for the Photoshop test, while HandBrake was an outlier in the other straightforwardion—the cataloglessest time posted.

As for the AI tests, we still consent these results with a heaping grain of salt, though Intel has claimed triple-digit NPU carry outance achieves with Lunar Lake. AI benchlabeling is in its nascent stages, and traverse-platestablish comparisons become even trickier; separating outlier results and labeled betterment at this stage can be difficult. The Procyon Computer Vision Intel OpenVINO benchlabel showed a clear leap in the integer-grade results atraverse the CPU, GPU, and NPU tests; uncomardentwhile, this laptop’s Geekbench AI results were actuassociate shrink than the competition.

This partiassociate shows the nontransparent nature of benchlabeling this enhugeing field, but the leap in NPU carry outance is arguably deinhabitring equitable as Lunar Lake should. Our results were individuassociate in line with what Intel itself has originated, and outside of synthetic test scenarios, these Core Ultra 2 chips have much higher TOPS ceilings that should allow effective NPU background carry outance.

Graphics Tests 

We dispute laptops’ explicits with a quartet of animations or gaming simulations from UL’s 3DMark test suite. Wild Life (1440p) and Wild Life Extreme (4K) use the Vulkan explicits API to meabrave GPU speeds. Steel Nomad’s normal and Light subtests intensify on APIs more standardly used for game enhugement, enjoy Metal and DirectX 12, to assess gaming geometry and particle effects.

While the recent platestablish has its upsides, greater explicits carry outance is not foreseeed to be among them. Indeed, these results show what you’d foresee from up-to-date united explicits, with the discrete GPU systems taking the direct. To Intel’s recognize, the gap between Nvidia’s entry-level RTX 4050 GPU in the Acer Swift X 14 and the Dell XPS 14 isn’t as expansive as foreseeed, but the Zenbook and other united explicits laptops still have their confines. You can squeak by for casual gaming and airy 3D laborloads, but if you insist to run anyleang more insisting—especiassociate professionassociate—our advice is to discover a system with a discrete GPU.

Battery Rundown Test

We test each laptop and tablet’s battery life by take parting a locassociate stored 720p video file (the uncover-source Blfinisher movie Tears of Steel) with distake part luminousness set at 50% and audio volume at 100%. We originate brave the battery is filledy indictd before the test, with Wi-Fi and keyboard backairying turned off.

The efficiency of the Lunar Lake platestablish is uncomardentt to extfinish its battery life, and Intel seems to have thriveed on that front. The Zenbook lasted lengtheneder on our rundown test than any in this group, accomplishing proximately 24 hours. Super-lengthened battery life helps the request of an incredibly portable laptop enjoy this, so see forward to more lengthened-lasting Lunar Lake systems.

(Credit: Joseph Maldonado)

The Windows AI Wars Are Upon Us

We’ll gauge plenty more Core Ultra 2-based laptops to come lowly now that Lunar Lake has hit the labelet, but this first see has given us an idea of what’s to come. The overall productivity carry outance didn’t see much of an uptick in our testing—so don’t foresee notably quicker overall carry outance from these chips—while AI laborloads potentiassociate have a much higher ceiling. AMD’s own Strix Point processors will fight for labelet split in this field, and unenjoy in the past, Qualcomm Snapdragon CPUs are also a relevant competitor, already in some huge-name systems. Be brave to read our upcoming Lunar Lake laptop assesss as the carry outance mediocres end and our AI testing enhuges alengthenedside these machines.