News Intel launches Lunar Lake: claims Arm-beating battery life, world’s fastest mobile CPU cores

[dada_dave]

Intel finally revealed its performance benchmarks for its Lunar Lake Core Ultra 200V-series processors here in Berlin, Germany, touting ‘historic x86 power efficiency’ coupled with the world’s fastest mobile CPU cores and 30% faster gaming performance than competing processors.

Intel launches Lunar Lake: claims Arm-beating battery life, world’s fastest mobile CPU cores : Read more

I believe there is a small mistake in the article:

These advances culminate in an Intel claim of 20.1 hours of battery life with the Core Ultra 7 268V in the UL Procyon Office Productivity benchmark and 10.7 hours of battery life in a Microsoft Teams call. This beats Qualcomm’s 18.4 and 12.7 hours of battery life with the X1E-80-100 chip.

12.7 > 10.7 so ... shouldn't it read something like "This trades blows with Qualcomm’s X1E-80-100 chip which lasted for 18.4 and 12.7 hours respectively"? - after all the Qualcomm chip lost by 2 hours in one test and won by two hours in the other?

Lunar Lake's advantage lies in its superior efficiency at lower power levels (sub-25W), rivaling the efficiency of the M3, which is commendable.

While I'm sure Intel has improved significantly, I would be very cautious about accepting that result at face value as it was in SPEC - while Intel has switched compiler backends, historically Intel's first party SPEC(Int) results have been very inflated relative to 3rd party reviewers (and with Anandtech gone it is harder to find reviewers who reliably do SPEC, I can only think of Geekerwan? Does anyone know another reviewer that does regularly does the SPEC benchmark?). While the text in the test description is very fuzzy and hard to read, I couldn't see where Intel said how they compiled SPEC for each machine. My guess is that they are still using their libraries/compiler front end and therefore getting a lot of extra free compiler optimizations. Still even if you knock off 20% (making the number up to be clear) that is still a massive improvement compared to where they were with Meteor Lake.

EDIT: In fact on the slide itself it might even say Intel Compiler? I can't really read it.

 

[kjfatl]

People who game do not care a lot about power efficiency and typically purchase a gaming laptop with dedicated GPU.
People who occasionally game on their laptop will not care much about a 30% performance increase over competing processors.
That leaves power efficiency as the main attractant. However, I bet $5 OEM's will sacrifice battery size to provide a slimmer, lighter laptop experience and the actual battery will be roughly the same as current generation, leaving nothing noteworthy about this processor release on the table.

I thought that people who care about gaming bought a desktop with a high end GPU. This produce isn't aimed at serious gamers. It's aimed at the other 95% of the laptop market.

 

[Alvar "Miles" Udell]

All this talk of battery life makes me think back to my college days of having both a 6 cell AND a 9 cell battery for my MSI laptop and still not making it though the day without seeking an outlet...

 

[baboma]

>I thought that people who care about gaming bought a desktop with a high end GPU. This produce isn't aimed at serious gamers. It's aimed at the other 95% of the laptop market.

I'd re-categorize the above to "high-fidelity" or "AAA" gamers, as I play strategy games, which could be pretty doggone serious.

Semantics aside, it's an oversimplification to say "gamers need high-powered (discrete) GPUs"--which points to just how small a niche that HW sites cater to, when they emphasize gaming performance as the end-all (to wit: Ryzen 9K's reception). I dare say that much of the "other 95%" also play PC games at least occasionally. We just don't need "highest FPS" graphics.

So, for the "rest of us," iGPU is important, more important than would be for "AAA" gamers. On this, I'm happy to see iGPU improvement on the mobile side (starting with Phoenix Point to Strix Halo), but also on the desktop side (ARL with 4 Alchemist cores.) Desktop iGPU has been stagnant for the longest time, so Alchemist's inclusion is a significant milestone.

ARL-Refresh next year will reportedly get an actual NPU to satisfy Copilot+ requirement. My hope is that it will also get 4 Battlemage cores for desktop. If Windows handhelds--read: gfx/watt--will become a thing. then it may be the tide that lifts all boats.

Speaking of high-powered GPUs, latest rumors sez 5090 will need 600W, 5080 for 400W. 5080 purportedly will have 1.1x perf over 4090. Love to see pricing on either. SWAG: 5090 will MSRP at $2K, 5080 at $1.5K.

https://videocardz.com/newz/nvidia-...0w-with-10-performance-increase-over-rtx-4090

 

[Giroro]

This is a multi-thread compute chart comparing power efficiency: The 165H, with its 14 cores (6 Performance and 8 Efficiency), outperforms Lunar Lake 8 cores (4P/4E) at power levels above 25W, not too surprising with the 6 missing cores.

Lunar Lake's advantage lies in its superior efficiency at lower power levels (sub-25W), rivaling the efficiency of the M3, which is commendable. It's unfortunate, however, that it's still dependent on running Windows.

But "more cores/threads = more efficient" really undercuts their repeated marketing message of "We went with 8 cores and no hyperthreading because it's the most efficient, so deal with it".

I get the impression that the engineers sat down with the goal of trying to make the most efficient <25W processor that they possibly could, but then later some exec stepped in and said that this one part had to be an entire product line that had to win in some kind of absolute performance metrics, so they scrambled to try and figure out how to segment their 1 good processor into 9 different processors. Then the marketing department got desperate and lost their minds, apparently.

Which is a shame, because I bet that this CPU is going to do a good job at the one thing it's designed to do: being the most efficient, <20W, 8 Thread processor.
But there's a good reason why Apple isn't out there pretending like their 8 Core M3 is 5 different processors covering the entire gamut of performance (9 when you factor in memory configurations).

If efficiency is the whole point of these CPUs, and these CPUs are only at their most efficient under 25W (probably 17W) then they should not be boosting to 37W, and the 30W Ultra 9 should not exist.
I get the impression the claims of great performance only apply to the Ultra 9 when running at an inefficient power level, and the claims of great efficiency will apply to the lower power parts, which will not have groundbreaking performance.

Which is par for the course, really, it just rubs me the wrong way that they want us to believe that the "Ultra Performance" and the "Ultra Efficiency" will be happening at the same time across the entire product line, when really they're just min-maxing two different CPU configurations for two unrelated tests.

 

[Gururu]

First reviews are up, looks like it can dance.

 

[Pierce2623]

Wow their benchmarks are blatant lies. The Ryzen 9 beats the Snapdragon X Elite in any kind of Cinebench run.

 

[Pierce2623]

I thought that people who care about gaming bought a desktop with a high end GPU. This produce isn't aimed at serious gamers. It's aimed at the other 95% of the laptop market.

Plenty of gamers like the benefits of the PC ecosystem without the drawbacks of a massive tower. I have a massive 4k gaming rig with a 4080 in my office but 90% of the gaming I do is on 1080p medium settings on a mini-pc with a discrete 6650m attached to my big TV in the den.

 

[Giroro]

I'm not one of Intel's genious marketing executives - who apparently are still around to guide the company through their declining sales, loss of profits, and resulting layoffs of 10% of the company. I don't even know which part of the company that marketing execs are even directly responsible for, but they must have been REALLY good at it to keep their jobs.

But here is how I would have approached the branding for this product:

1. Instead of 5 nearly-identical CPUs with various memory capacities, it would be one CPU, with at most 3 memory configurations. - Which is what Apple did with the M3 this is directly supposed to compete with.
2. Drop the stupid V, because when you're talking about electronics and a V is located after a number, it's pronounced "Volts". They only used a V because 'V is one more letter than U' - but they used U for so long for a reason: It was short for "Ultrabook®".
3. Simplify the branding in general- Because Intel Core Ultra 5/7/9 is less than 2 generations old, and it's already become unnecessarily confusing.
4. Don't jump through so many hoops to pretend like the CPU is the absolute performance leader, when clearly that was never going to happen for an Ultra-mobile chip. It just comes across as contradictory, which is undercuts the chips real advantages, like efficiency.

So with that in mind, Instead of calling this 8 core CPU that whole table of Intel Core™ Ultra 5/7/9 2xxV, It would instead be called The Intel 8 Core Ultra - abbreviated Intel 8U.

2 SKUs 832U for 32GB, 816U for 16GB. (The U stands for "Ultra", and retains brand recognition between products, now that everything else is different).
8 Core CPU, 8 Core GPU (assuming there's good yield on the 8 core GPU). Simple, and right there in the name.
17 Watt Part, assuming that is it's peak of efficiency.

It's more catchy, easier to remember, and has meaning. 8U might flow as good as M3, but it's still way better than what they are going with.
If OEMs want a 3rd SKU, then we would be cheeky and give them an Ultra low power, entry level Intel 808U - 8GB memory configured to 8 Watts power draw (Maybe call it an 888U). Although I would prefer the memory system be reworked a 3 SKU lineup could be 12/24/48GB.

This branding helps differentiate the low power chip the Ultra 5/7/9 mainstream and high performance mobile chips, and prevents it from diluting the Ultra 9 brand - because there's no way this chip was ever going to meet peoples performance expectations from a "9" series chip - so don't even try.

 

[watzupken]

If I give Intel the benefit of the doubt that what they claim is true, the problem for Intel will be pricing. On their own foundry, an equivalent Intel based laptop usually cost more than competition. On TSMC 3nm, I think they have little choice but to pass the cost onto us. AMD and Qualcomm are both on TSMC N4/5, which likely will give them a cost advantage without significant efficiency and performance disadvantage.

 

[thestryker]

Is that chart saying that Ultra 9 288V is less efficient than Ultra 7 165H? Because to me it looks like they've had a significant regression in Perf/W at high(er) package power, yet the words they chose are "Astonishing perf/W improvements across the entire operating range".

Keep in mind package power for LNL includes the DRAM which while isn't a high number (and doesn't change the absolute performance since LNL seems to plateau) isn't the case for Zen 5/MTL (add probably 3-5W). The only reason I can think of for the higher TDP is the integrated graphics as it would allow both to be fed more consistently as it definitely doesn't appear to be CPU related.

 

[thestryker]

Wow their benchmarks are blatant lies. The Ryzen 9 beats the Snapdragon X Elite in any kind of Cinebench run.

The tests on that slide are only 1T and X1E-80 and above are faster than most x86 laptop CPUs there.

 

[TheSecondPower]

Intel's old Golden Cove core keeps showing up near the top of benchmarks (as Raptor Cove), even in Ryzen 9000 reviews. The problem is it thirsts for power. Redwood Cove on Intel 4 does a lot better, especially with help from updated Crestmont e-cores. But Lunar Lake has Intel's first new p-core in years. It had a huuuuge new e-core. It has Intel's very latest and greatest graphics architecture. And a new Apple-like chip layout and a simplified tile architecture. And the compute tile is built on TSMC's N3B node.

If it doesn't outperform Snapdragon X in power efficiency and Ryzen 300 in low-power performance, I will be surprised.

Intel has 7 models because Intel bins their CPUs more than AMD. Fewer models means more variation within a model. More models means the top model is only golden bins; it'll clock higher or use less power than the others.

Intel doesn't need this for every market. Arrow Lake is coming soon for the higher-power markets. But based on rumors it'll be held back a little by its tile architecture and its older tiles carried-over from Meteor Lake and possibly from less bleeding-edge CPU tile node(s). So this is Intel's halo chip until next generation.

 

[baboma]

On pricing:

I perused the web looking for Ryzen 365/370 laptops to compare against LNL. Despite launching over a month ago, I was hard put to find anything other than Asus models, all of which are available on Bestbuy. (Amazon also has some weird brand called HIDevolution for ~$2K.) Ryzen 300's availability on any other brand is pretty much zilch.

From Bestbuy (excl models w/ dGPU):

$1200 Asus Vivobook S 14 (Ryzen 365, 24GB RAM, 512GB SSD)
$1400 Asus Zenbook S 16 (Ryzen 365, 24GB RAM, 1TB SSD)
$1700 Asus Zenbook S 16 (Ryzen 370, 32GB RAM, 1TB SSD)

Per leaked LNL pricing (from EU shops), Asus LNL models will likely start at $1K or under. (The models w/ leaked pricing are Asus Expertbook, which is normally slotted between Vivobook and Zenbook pricewise.) Per above, Ryzen 300 laptops will come at a significant premium over LNL, with perhaps worse availability. This is surprising, as AMD prior to Ryzen 300--ie 7040 and 8040 series--has had a price advantage over Intel MTL. Apparently, the opposite will be true for this gen.

https://videocardz.com/newz/core-ul...acer-and-asus-start-at-e996-according-to-leak

The three Asus LNL models Intel showed for its briefing are the Zenbook S 14, Zenbook S 16, and ExpertBook P5. Hopefully they will be released soon, and we can get like-for-like compare on both pricing and performance.

https://www.techopedia.com/news/int...ds-on-with-laptops-that-take-on-apple-and-amd


BTW, Win11 24H2's implied release date is "beginning of November."

https://www.engadget.com/ai/copilot...by-intel-and-amds-latest-chips-190707475.html

"Qualcomm’s exclusivity period on Copilot+ PCs is winding down. Microsoft confirmed on Tuesday that Intel’s new 200V processors and AMD’s Ryzen AI 300 series chips will add Copilot+ AI capabilities beginning in November."

 

[dada_dave]

I believe there is a small mistake in the article:



12.7 > 10.7 so ... shouldn't it read something like "This trades blows with Qualcomm’s X1E-80-100 chip which lasted for 18.4 and 12.7 hours respectively"? - after all the Qualcomm chip lost by 2 hours in one test and won by two hours in the other?



While I'm sure Intel has improved significantly, I would be very cautious about accepting that result at face value as it was in SPEC - while Intel has switched compiler backends, historically Intel's first party SPEC(Int) results have been very inflated relative to 3rd party reviewers (and with Anandtech gone it is harder to find reviewers who reliably do SPEC, I can only think of Geekerwan? Does anyone know another reviewer that does regularly does the SPEC benchmark?). While the text in the test description is very fuzzy and hard to read, I couldn't see where Intel said how they compiled SPEC for each machine. My guess is that they are still using their libraries/compiler front end and therefore getting a lot of extra free compiler optimizations. Still even if you knock off 20% (making the number up to be clear) that is still a massive improvement compared to where they were with Meteor Lake.

EDIT: In fact on the slide itself it might even say Intel Compiler? I can't really read it.

Better slide deck here:

https://download.intel.com/newsroom/2024/client-computing/Intel-Core-Ultra-Series-2-Media-Deck.pdf

Definitely Intel Compiler, so the Intel SPEC results for ST and MT are likely inflated. I mean that's already fairly obvious as the single thread performance uplift over the Snapdragon 78 is 61% in SpecInt but only 20% in Geekbench and Cinebench. While of course different tests can have different results, I strongly suspect that if/when 3rd parties test Lunar Lake, its ST SPECInt score will be closer to 20% than 61% higher. Also it claims that the Snapdragon 80 using 50W is the same performance as the M3 in multithreaded SpecInt. That would be very odd if true. While I found the Qualcomm multithreaded performance scores a tad disappointing in general, they weren't that bad. It also says something about Intel Instrumented for the Qualcomm chip which isn't explained even in the notes as far as I can tell.

So yeah as the article states, as always, wait for 3rd party reviews and benchmarks, hopefully someone will do SPEC.

 

[strobolt]

As others have said, I am tired of hearing about salt in every article.

Personally, using phrases like this is like a trademark currently and I don't mind. Also, I like to think that it throws of chat bots trying to scrape the text from here and end up using them incorrectly. I'm not saying it's likely to happen, just that I find the thought itself amusing.

 

[TheHerald]

At that point I fear that Intel has become Topper from Dilbert :

AMD : My CPU has 16 cores
Intel : That's nothing! My CPU has 8+16=24 cores (even though you'll use only 8 in games)

Apple : My battery life is 20 hours
Intel: That's nothing! My battery life is 20,1 hours ! (and it could be even greater if I just had only 2 cores without SMT)

I think they just have a narcisstic image problem that can't allow them to admit they are behind.

Also at that point, if microchips are a US national security issue, I'm confident that the health of Intel is also a US national security issue, with all that entails (ahem).

Games definitely (some of them, obviously) use more than 8 cores. And yes, they use ecores extensively too.

 

[bit_user]

Still even if you knock off 20% (making the number up to be clear) that is still a massive improvement compared to where they were with Meteor Lake.

One point these numbers really drive home is what an absolute dog Meteor Lake was! I think that was pretty clear, but it's still somewhat shocking to see it next to Lunar Lake. I was getting major Ice Lake deja vu, with that one!

 

[newtechldtech]

on package DDR5 ??? no HBM technology >>> big question mark.

 

[newtechldtech]

They only used a V because 'V is one more letter than U

knowing intel , They used V for "Victory"

 

[Marlin1975]

If I give Intel the benefit of the doubt that what they claim is true, the problem for Intel will be pricing. On their own foundry, an equivalent Intel based laptop usually cost more than competition. On TSMC 3nm, I think they have little choice but to pass the cost onto us. AMD and Qualcomm are both on TSMC N4/5, which likely will give them a cost advantage without significant efficiency and performance disadvantage.

Yea the TSMC 3nm plus onboard high speed memory will give intel more performance and power advantage. But as said will not be cheap.

Either they will have to sell at cost to keep market share or charge a premium to keep profits up.

 

[jp7189]

No.. Intel's fabs just weren't ready for THIS product announcement. The contract to fab Lunar Lake on TSMC was likely completed years ago, before Intel had embarked on it's fabled "four nodes in 5 years" death march. Intel appears to be on track with at least 3 of those four nodes and if they do complete it, you bet that Panther Lake will be manufactured in Intel factories.

That's what I thought before reading this. They jumped 2x perf/watt mostly from switching to a different fab. When was the last time you saw 30% better perf and 50% better power in a single generation?

I'm now thinking 18A is going to have a hard time being better. Sure on paper it has every advantage, GAA and BSPND, but if end products aren't good, none of that stuff matters.

 

[Pierce2623]

The tests on that slide are only 1T and X1E-80 and above are faster than most x86 laptop CPUs there.

Yeah and if you look up REAL scores the new Ryzen smokes EVERY snapdragon x elite in single core Cinebench runs

 

[Pierce2623]

Games definitely (some of them, obviously) use more than 8 cores. And yes, they use ecores extensively too.

Other than MSFS 2020, Cyberpunk is the most multithreaded game I know of and it can only spawn 16 threads.

 

[vink]

This FANTASTIC processor by Patrick Gelsinger is MAGIC. The incredible Lake with TDP 7W I hope to see it quickly for myself and my clients in FANLESS LAPTOPS (no dust-no service) for work, gaming, audiophile, cinephile ... for anything with the extremely versatile Windows Ms.