Intel Teases Lunar Lake CPU Ahead of Computex: Most Power Efficient x86 Chip Yet
by Ryan Smith on May 20, 2024 2:00 PM ESTThe next few weeks in the PC industry are going to come fast and furious. Between today and mid-June are multiple conferences and trade-shows, including Microsoft Build and the king of PC trade shows: Computex Taiwan. With all three PC CPU vendors set to present, there’s a lot going on, and a lot of product announcements to be had. But even before those trade shows start, Intel is looking to make the first move this afternoon with an early preview on its next-gen mobile processor, Lunar Lake.
While Intel hasn’t said too much about what to expect from their Computex 2024 keynote thus far, it’s clear that Intel’s next-gen CPUs – Lunar Lake for mobile, and Arrow Lake for Mobile/Desktop – are going to be two of the major stars of the show. At this point Intel has previously teased and/or demoed both chips (Lunar more so than Arrow), and this afternoon the company is releasing a bit more information on Lunar Lake even before Computex kicks off.
Officially, today’s reveal is a preview of Intel’s next Tech Tour event, which is taking place at the end of May. Unofficially, this is the exact same date and time as the embargo on Qualcomm Snapdragon X laptop announcements, which are slated to hit retail shelves next month. Lunar Lake laptops, by contrast, will not hit retail shelves until Q4 of this year. So although the additional technical details from today’s disclosure are great to have, looking at the bigger picture it’s difficult to interpret this reveal as anything less than a bald-faced effort to interdict the Snapdragon X launch (not that Qualcomm hasn’t also been crowing about SDX for the last 7 months). Which, if nothing else, goes to show the current tumultuous state of the laptop CPU market, and that Intel isn’t nearly as secure in their position as they have traditionally been.
And with that out of the way, let’s dive into the meat of Intel’s announcement.
Lunar Lake: New Architecture Focused on Power Efficiency & AI
Ever since it was initially demoed by Intel last fall, the company has been promoting Lunar Lake as a highly efficient mobile processor. And that remains the story today. The low-power successor to Meteor Lake, Lunar Lake is Intel’s second mainstream tiled CPU, and will see Intel bringing new CPU and GPU architectures to bat in an iterative improvement over Meteor Lake.
And while Intel isn’t throwing out Meteor Lake quite yet, by Intel standards it’s going to have a relatively short shelf-life. From a high-level roadmap perspective, the groundbreaking tiled chip was late, and consequently as Intel Foundry works to catch up via its 5 Nodes in 4 Years (5N4Y) manufacturing strategy, Intel isn’t going to wait to bring out their next chips. They can’t; not with AMD and now Qualcomm both making serious efforts to cut into Intel’s mobile market share.
Development on Lunar Lake is already complete, with Intel claiming that they meet or beat all of their production milestones for the chip. This also means that Lunar Lake is already in production, using a B-stepping design.
Intel expects to begin shipping the SoC to OEMs in Q3 of this year, which means finished laptops will reach retail shelves in Q4 – though at this point it’s unclear if that will be early or late into the quarter. Meanwhile, Arrow Lake, which is going to be both for desktop and mobile usage (as a higher performance chip), is set to ship from Intel in Q4. We’re told we’ll find out more about Intel’s plans there at Computex.
Intel CPU Architecture Generations | |||||
Alder/Raptor Lake | Meteor Lake |
Lunar Lake |
Arrow Lake |
Panther Lake |
|
P-Core Architecture | Golden Cove/ Raptor Cove |
Redwood Cove | Lion Cove | Lion Cove | Cougar Cove? |
E-Core Architecture | Gracemont | Crestmont | Skymont | Crestmont? | Darkmont? |
GPU Architecture | Xe-LP | Xe-LPG | Xe2 | Xe2? | ? |
NPU Architecture | N/A | NPU 3720 | ? | ? | ? |
Active Tiles | 1 (Monolithic) | 4 | 2 | 4? | ? |
Manufacturing Processes | Intel 7 | Intel 4 + TSMC N6 + TSMC N5 | Intel + TSMC? | Intel 20A + More | Intel 18A |
Segment | Mobile + Desktop | Mobile | LP Mobile | HP Mobile + Desktop | Mobile? |
Release Date (OEM) | Q4'2021 | Q4'2023 | Q3'2024 | Q4'2024 | 2025 |
Despite all of this, Intel has still not confirmed what manufacturing processes are being used for Lunar and Arrow Lake. As part of Intel’s efforts to right themselves and separate their Foundry and Products businesses, Intel’s chip design groups are free these days to pick whatever fab they’d like – which in practice means using TSMC to produce some of their dies, as what happened with Meteor Lake. So we remain in the dark on which parts of the chips are being fabbed where.
For today’s preview, Intel is using the same Lunar Lake die diagram as they used at Intel Vision last month. This shows a more compact design with 3 tiles – SoC (top), compute (bottom), and then a third empty tile that is presumably being used as structural silicon. This is down from 4 active tiles on Meteor Lake.
Given Intel’s energy efficiency focus, fewer tiles is not surprising to see. Even with the ultra-short distanced enabled by Foveros/EMIB 2.5D packaging, moving data off of a die remains a very power-expensive process, so there are some significant energy benefits to keeping things within a die as much as possible. This is also part of the reason we’re seeing on-package LPDDR5X memory for Lunar Lake, as this cuts down on the memory bus length while making the overall SoC + RAM package more compact.
As expected, Lunar Lake is getting new P-core and E-core architectures, which are further evolutions of Intel’s Cove and Mont CPU core lineups. The P-cores will be the new Lion Cove architecture, while the E-cores get Crestmont’s successor, Skymont.
While Intel isn’t going into deep detail on these architectures today, what they are promising is being positioned as a significantly bigger step up than what Meteor Lake (Redwood Cove/Crestmont) delivered, with Intel being extremely optimistic about CPU core performance. Both the P and E-cores will deliver significant IPC and performance-per-watt improvements – and even if the latter comes entirely from process node improvements, IPC improvements hint at greater internal changes.
The footnotes of Intel’s presentation also confirm that at least one SKU configuration for Lunar Lake will be a 4P + 4E design, with an undisclosed number of low-power E-cores. Meteor Lake-U, by comparison, is 2P + 8E + 2LP. Ignoring the LP cores for the moment, that means Lunar Lake is ending up with two fewer CPU cores overall, but it’s getting twice as many P cores. Overall, it’s a much more balanced design than the E-core-heavy designs of Meteor Lake and Alder/Raptor Lake.
The mobile chip will also come with a new GPU architecture, Xe2. This is the same GPU architecture as will be featured on Intel’s forthcoming Battlemage discrete GPUs.
Details on Xe2/BM are slim, but most notably, the implementation going into Lunar Lake is getting Intel’s Xe Matrix eXtention (XMX) cores, which have been a feature of the architecture since Alchemist, but not included with the Xe-LPG design used for Meteor Lake. This change should bring Intel’s iGPU closer to feature parity with its desktop counterparts, as well as giving it the means to run the high-quality version of Intel’s XeSS temporal image upscaling technology. Meanwhile the new iGPU should also deliver a 50% increase in raster graphics performance, for a sizable boost in the pixel pushing power of the pint-sized chip.
And in this era of AI everywhere, we’d be remiss not to mention the AI potential for the XMX cores. These give Lunar Lake yet another high-performance processing block to run neural networking computations, with an expected performance of over 60 TOPS (INT8).
Lunar Lake is also getting a new NPU, though Intel hasn’t gone into any architectural details here (or even named it). As previously disclosed by Intel, the Lunar Lake NPU is rated for a minimum of 45 TOPS at INT8 precision, over 3x the throughput of Meteor Lake’s 11 TOPS NPU. The overall performance of the entire chip, meanwhile, is expected to reach over 100 TOPS.
At this point, Intel, Qualcomm, and AMD are all simultaneously promoting AI workloads as the next big thing and squabbling over who has (or will have) the best NPU. Intel expects Lunar Lake to lead in raw performance, but they also are pushing the software angle pretty hard, outlining their toolset and developer support. No one in the industry seems particularly happy with TOPS comparisons right now (not the least of which is because all the same players regularly fail to disclose at what precision), but until the first killer app comes along that they can use as a meaningful benchmark, we’re going to continue to see raw throughput figures thrown around.
Otherwise, the final and biggest push on the architectural side of Lunar Lake is power efficiency and battery life. While Meteor Lake and Alder/Raptor lake have been okay in these departments, Intel hasn’t had a product that’s truly knocked things out of the park since Tiger Lake in 2020. And everyone is still chasing the Apple M-series, which is now on its fourth iteration. Lunar Lake is dedicated mobile silicon that’s been designed in particular for low-power devices, so Intel is expecting to deliver significant gains here (and fend off their competition), with Intel product managers calling it a “transformative moment for x86 power consumption.”
More details on how Intel is accomplishing this are to come, but a big part of that appears to be the new SoC die and its low power island. According to Intel the new chip will improve on every aspect of power efficiency, including what’s arguably the most critical aspect for them at this point, idle power consumption.
Performance Expectations
Finally, Intel is publishing brief set of performance slides for Lunar Lake, outlining their performance and battery life expectations versus themselves, as well as both AMD and Qualcomm. With the latter’s laptops not yet shipping, all of this should be taken with multiple grains of salt as things can and do change over time. But for now, we have performance slides to fight other performance slides.
It should be noted that Intel’s footnotes indicate they’ve tested Lunar Lake systems at both 17W and 30W (complete chip power, including DRAM). Though other than their graphics performance claims, Intel hasn’t made it clear which TDP configuration was used for which test. So for the time being, we’re assuming everything is with Lunar Lake at 30W unless otherwise noted.
Underscoring today’s themes, Intel goes right for a Qualcomm AI performance comparison, claiming that they can beat their southern neighbors by 40% in Stable Diffusion 1.5 running on the GIMP photo editor. One of Qualcomm’s favorite benchmarks, Intel reports that this is in comparison to Qualcomm’s published figures. Similarly, Intel claims they have the lead over both the Snapdragon X Elite and AMD’s Ryzen 7 8840U based on undisclosed performance estimates.
Finally, in GPU performance comparing 17W Lunar Lake versus 15W Meteor Lake Core 7 165U, Intel is seeing 50% better performance in 3DMark Time Spy, a raster-only workload.
And for power consumption, Intel put together a workload in Microsoft Teams that involves 9 video windows with AI effects (Windows Studio Effects) layered on top. This was a metered test using hardware in Intel’s labs, so the comparisons here are the Ryzen 7 7840U and Qualcomm’s antiquated Snapdragon 8cx Gem 3. Unsurprisingly, Intel has the lead here, and this is despite the handicap of package power measurements for Lunar Lake including the memory while the Ryzen and Snapdragon do not. Unfortunately, Intel is not releasing any data comparing Lunar Lake’s power consumption to Meteor Lake.
Lunar Lake: Retail Devices Coming in Q4
Wrapping things up, today’s disclosures on Lunar Lake are the tip of the iceberg – designed to whet the appetite – ahead of Computex and beyond. With Intel holding their Tech Tour event at the end of this month, and then a Computex keynote a few days later, Intel will be providing significantly more details in short order – as will the rest of their competitors, for that matter.
Beyond those events, Intel intends to start shipping Lunar Lake processors to OEMs in Q3 of this year. This will get laptops out in Q4, just in time for the all-important holiday period.
62 Comments
View All Comments
meacupla - Monday, May 20, 2024 - link
I am so confused with what you are trying to say here.Try running your native language through an AI translator.
nandnandnand - Tuesday, May 21, 2024 - link
His problems only start with the language barrier.Iketh - Friday, May 24, 2024 - link
He did. It didn't understand either.hecksagon - Friday, May 24, 2024 - link
Initial reviews show it trading blows with the M3 and Meteor Lake, even when running x86 native code. It would have no problem running most if not all business software.Dante Verizon - Tuesday, May 21, 2024 - link
I don't see any reason for AMD to move a single penny in prices. X86 has the compatibility crown, and several years of work on the software/drivers side ensures relative stability. Qualcomm is launching a product that should be seen as in beta, you're buying a problem.meacupla - Tuesday, May 21, 2024 - link
Laptop OEMs have been disappointed with Intel's recent offerings. The time is ripe to go aggressive and gain market share.Wintel's last bastion of laptops, Dell XPS will have AMD as an option for the first time in 2026.
Bruzzone - Wednesday, May 22, 2024 - link
I have Snapdragon X at $68 to $80 from TSMC to QCOM + 55% gross = $105 to $124 in volume.AT $145 / 2 = $72.5 from TSMC to QCOM so in the ballpark.
Raptor mobile average gross on q1 2024 = $186 but if you pull the R&D charge at 'run end' = $122. Meteor Lake however is costly around $227 with marginal cost just now dipping below $105. Intel can sell Mobile in q1 on average all the way down to $98 and desktop to $76 and still eek out a bit of margin.
There is a price war between AMD and Intel currently into OEM and contract manufacturers that extends into the supply chain all the way to end sale where the PC market is facing a deflationary price decline for current generations of product through q4 into q1 2025. It's a really good time to buy a laptop and gets better into the second half on the channel clearing overage. A lot of overage, AMD and Intel have been dumping on each escalating q3 2023 continuing.
Snapdragon aims for a production trough between Raptor and Lunar and Rembrandt / Dragon and Strix. Everything in between represent intermediate bridge technologies, first attempts similar SDX; Phoenix/Hawk and Meteor Lake, where OEMs are waiting for next generation.
I'll also raise the question of Intel dealing group interest in Snapdragon X barking for Intel next generation mobile access and allocation pulling the old 'negotiating chip' routine.
Here's current ARM PC share on channel data;
All categories M_ = 68.7%
Pro X SQ 1/2 = 23.1%
Snap on Windows 10 laptop = 8%.
My impression is Snapdragon X is meant to offer an upgrade path to current Snap on 10 and SQ 1/2 for business user base. The user base of these two platforms was basically abandoned or at least not attended to.
Snapdragon X TAM is 45.4% of current M_ installed base just touching 60 M units so said by Apple Mac trackers.
QCOM is aiming for some percentage of 30 M units offering Snapdragon and ARM Surface users the X / Elite upgrade path. There is a market off the blocks representing the installed base of ARM on Windows and Elite / X is that option today on preorder currently.
Specific timing Snappy X may present a better tool for 'AI' PC applications development also offering a performance per watt advantage for business use Still, Qualcomm is launching into a PC market deflationary price spiral and AMD and Intel are prepared to dump current channel holdings all over ARM-based Windows attempts through end of year and into q1 2025.
mb
Kangal - Tuesday, May 21, 2024 - link
Seriously this.Intel is such a corrupt, monopolistic, capitalist company and have such a shady track record. It is more intelligent to assume they have lied, are lying, and will lie again soon.
I am no fanboy of AMD, but what happened with the Windows Kernel and the Bulldozer architecture was not very fair. And then what happened with AMD and the OEM companies during that 2003 - 2013 period was also horrible. It took until the AMD r7-3700x in Q2-2019 to catch up to Intel, and they managed to surpass them with the r9-3950X in Q4-2019. It basically took them a decade to recover from what was basically bankruptcy.
I mention the above as context, because we've been devoid of competition in this sphere for a while. Look at how quickly things progressed from the Q1-2017 to the Q4-2020 (r7-1700 to r7-5800x) all on the same motherboard. In less than 3 years, we saw an uplift of x2.5 times, or x3 if you stretch it out to include the 5800x3D.
In terms of companies; I put as Intel worst. Then Microsoft a close second, Nvidia as a distant third, Apple as a close fourth, and lastly Qualcomm as the least worst consumer tech corporation. There's more out there, but I just wanted to focus on the big ones here.
GeoffreyA - Tuesday, May 21, 2024 - link
I'm not sure how I'd rank these four bad apples, but I'd let them fight it out for first place on the throne of rottenness.Oxford Guy - Thursday, May 23, 2024 - link
'what happened with the Windows Kernel and the Bulldozer architecture was not very fair'The Bulldozer architecture was an unforced error on AMD's part. How much it can be blamed upon the OEM scam, though, I can't say.