Intel Lunar Lake debuts as "most efficient" x86 design with combined 120 TOPS AI performance

[Redaktion]

Intel is offering deeper insights into the Lunar Lake architecture at Computex 2024. Lunar Lake will be a tiled 3D packaged design for laptops with a focus on per-per-Watt and AI performance. Lunar Lake marks the arrival of Lion Cove and Skymont cores, Xe2-LPG Battlemage iGPU, and a 47 TOPS NPU along with a host of other platform enhancements.

https://www.notebookcheck.net/Intel-Lunar-Lake-debuts-as-most-efficient-x86-design-with-combined-120-TOPS-AI-performance.843906.0.html

[RobertJasiek]

"The battle for AI supremacy [...] Together, the CPU, Xe2-LPG GPU, and the NPU can afford a total of 120 TOPS platform AI compute performance."

Do you believe every PR-nonsense of Intel? There cannot be AI supremacy by iGPUs while prosumer dGPUs have up to >1300 TOPS, i.e. more than 10x the speed (due to much higher TDPs etc.)!

"Lunar Lake Efficient Cores [...] >2X Over last gen low power island [...] Overall, Intel claims a 40% lower power requirement for the Lunar Lake RVP compared to Meteor Lake. This is during playback of YouTube 4K 30 AV1 video."

And elsewhere, Intel claims >20% better performance per watt of lower power cores.

Now, Intel has thrown out three numbers: >2X, -40%, >20%. Which of them is the closest to the truth?

What are "last gen low power island" and "RVP", please?

"XMX is Intel's equivalent of Nvidia Tensor cores in GeForce RTX cards"

"equivalent" is the strongest possible word, but do you actually mean it?! Does Intel license the Nvidia Tensor cores? If so, are Intel's libraries for those cores as good as Nvidia's TensorRT libraries, which presume usage of CUDA and CuDNN libraries? Or does Intel just offer some matrix cores of its own?

"Lunar Lake is the first design that is outsourced to an external fab instead of being forged in Intel's own foundries. [...] TSMC N3B."

Now, this is worth more than one news article! It is the most important information about Intel chips for years...!

It might also explain the sudden jump in power efficiency.

[NikoB]

Now, this is worth more than one news article! It is the most important information about Intel chips for years...!

It might also explain the sudden jump in power efficiency.

As I wrote earlier, you again did not see the key news - Intel is able to get a large share of the pipeline at TSMC, and why did AMD not want to expand its market share when literally everything was in its favor.

Conclusion - AMD is a 100% fake company designed to divert the attention of antitrust authorities in the US and other countries from Intel and nothing more. And soon everyone will see it.

If the company is market-oriented - it will strive to expand its market share at the slightest opportunity - what prevented it from increasing the production of AMD processors on TSMC assembly lines all these years? Why is the ratio of AMD SoC to the share of Intel SoC still 1:5, and not 50/50, at a minimum?

And now AMD is again moving into the category of a catching-up company, as soon as Intel, to its shame, nevertheless decided to use the same technical processes at TSMC as AMD. What chances does AMD have now? None. Everything is over. And soon its shares will begin to plummet downward.

[Toortle]

NikoB went from "Intel is bankrupt" to "AMD is a fictional company and Intel controls the world" 🥴

[Hotz]

Why is the ratio of AMD SoC to the share of Intel SoC still 1:5, and not 50/50, at a minimum?

Some say that Intel has contracts with all computer brands to favor Intel chips instead of AMD.

Others say that Intel has engineers actively helping those manufacturers to integrate Intel chips, while AMD has not.

Again others say, that insiders told them that lots of manufacturers - while still being bound to Intel by contracts - are actually fed up with all the nonsense from Intel the last couple of years (because Intel is constantly delaying their chips, hardly any substantial progress or innovations, seems objectively worse than AMD), and that they will soon integrate more AMD chips in their hardware.

I don't know if the last paragraph is true or not, but there have been in fact much more AMD devices announced compared to 1-2 years ago.

[NikoB]

A simple example in the motherboard market is the ratio of Intel motherboards with a contact group for SPDIF output or a rear SPDIF output (optical) compared to AMD motherboards of the same class (15-20): 1. This alone stops me from buying an AMD desktop system.

AMD doesn't seem to care at all about the desktop and laptop market. If earlier some of the opponents objected (and lied in the press) that supposedly TSMC had everything booked on its assembly lines for years in advance and there was no room for expanding orders, now these false arguments are not worth a damn - when Intel needed a large-scale order of chips from TSMC, there was room on the assembly lines found here. What prevented AMD from taking this place for itself, besides the fact that it is Intel's anti-monopoly gasket? Loans were cheap. You could dramatically expand your market share, and then start making profits many times greater.

And now everything is back to normal. Intel, in addition to creating new factories (taking into account government subsidies) and thanks to orders from TSMC, no longer lags behind AMD in terms of technical processes. What are the chances now for AMD, which still has much less money and a market share that is 5 times lower? None. This is oblivion again, unless a miracle happens. A chance in the market is given once - AMD has not taken advantage of its chance since 2015. This means that this is not a market company.

[Hotz]

AMD doesn't seem to care at all about the desktop and laptop market.

... What prevented AMD from taking this place for itself

 Intel, in addition to creating new factories no longer lags behind AMD in terms of technical processes.

This means that this [AMD] is not a market company.
 

I came to similar conclusions. I asked myself the same questions for the last 2 years, when their 680m iGPU came out and outperformed everything from Intel by 100% - and yet only was integrated in very few devices in nonsensical combinations with dGPUs, or in unknown experimental Mini PC brands, or in handhelds. It's weird... I mean they had 2 years to dominate absolutely everything, but stayed in the niche they always were. Either this or they *wanted* Intel to catch up first. Strange...

Back then I had the assumption they only care about *gaming devices* (gaming consoles, gaming handhelds), and that's the reason behind it (matching your quote about AMD not caring about desktop and laptop market). Maybe it's still the main reason behind it.

Another possible example for not caring about other markets could be the popular AsRock Deskmini Mini PC: the AMD version has very few connectivity options, requiring you to buy an additional USB port replicator, while the Intel version has much more connectivity out of the box. It's like AMD often cheaps on important things, because they still only think in terms of "gaming" and thus not integrate much connectivity (only allowing you to plug in a gamepad and it's done...), and not in terms of universally usable PCs with plenty of connectivity options.

Well... it's a lot of guessing and speculating, but something definitely doesn't feel right. Not that Intel feels great either... but ugh... it has become tedious... and annoying... everything...

What shouldn't be forgotten though, is that AMD had a platform change from AM4 to AM5 in that time frame where we expected it to grow and dominate the market. And it had to do that, because AM4 did only support up to DDR4 RAM. So maybe that took a lot of time and was holding AMD back...

[Ted Talk]

Some have also suggested it's a mindset issue. AMD is still stuck in the mentality that they're almost bankrupt (which was true not too long ago for the longest time) and are a smaller company in general.

As such their decision making largely stems from that era. They don't try to take as much risks or be leaders / pioneers in a market but rather go the well trodden path. By this I mean, there's no real guarantee that making a chip for x laptop oem will sell in y in volume / sales and meet z quota. The contracts are just not as strong. Its not like in data centre once you win a contract, it's usually a multi year one and guaranteed to print money as margins are huge. Or consoles, where the margins maybe razor thin, but this is offset by the fact you know Sony is guaranteed to sell hundreds of millions of APUs.

It's a culture issue which was developed out of necessity. Not having enough money. The single biggest thing which effects culture within an organisation is the leadership. They help create the vision and the values necessary for an environment to foster such a culture. One could argue maybe Lisa Su currently isn't the best CEO to help guide AMD into this next transition but she definitely did help get AMD get out of bankruptcy from a decade ago.

[A]

Do you believe every PR-nonsense of Intel? There cannot be AI supremacy by iGPUs while prosumer dGPUs have up to >1300 TOPS, i.e. more than 10x the speed (due to much higher TDPs etc.)!

Yes, there is iGPU AI supremacy. Most people are not training models, or trying to get the AI to generate for them some fancy image. For that people may need a dgpu. But for basic tasks like text to voice, add alt tags for screen readers, OCR, finding files on your computer via context, fix your grammar, those low hanging fruit can be done near realtime by those igpus on small less precise models

[RobertJasiek]

While I get your point that basic AI can run on every chip, presumably most people do not use AI at all yet. Besides, Intel and the article have tried to argue about supremacy in terms of TOPS, and this is plainly false as dGPUs reign there.

[George]

Ok, for FAR to long:

• 'Team Blue' has released "up to %10-15 improvement" (much to the HATE of just about EVERYONE)
• The 'OEM Designs' their partners produce are mostly variations of 'reference designs' Intel produces.
• Many OEM's are "locked in" with purchasing contracts (must buy xx#)
• AMD MAY be or MIGHT OF BEEN ahead in many of their products HOWEVER it takes TIME for IT/Product managers to 'shift gears' and change/demand products with other companies parts (IE: AMD)
• Individual companies (with IT departments) have their OWN PC "refresh cycles" that are often contracted out 6mo-1yr in advance with specific OEM's and specific models available.
• While Intel has publically mentioned its attempts to get others to use their (outdated) FAB's they have also mentioned their own use of TSMC while Intels 'newer FABs' (that they of course claim that will 'leap frog' TSMC's) are being built in Germany & the US.

[NikoB]

But for basic tasks like text to voice

Everything is exactly the opposite - the key basic task for the average consumer from "AI" is voice to text. And even in a difficult noise environment.

Or on-the-fly translation of a conversation or "dubbing" into a movie with perfect preservation of the intonation (including the sound atmosphere with reverberations) of the original actors and the exact semantic correspondence of the text to the original. And this does not take into account jokes, which are often understandable only to speakers of the particular language in which the film was created, which requires additional intellectual effort from the "AI" to understand what analogue of a joke or sarcasm in one language can correspond in another language. These are real "super" tasks, although they are banal. And the text in voice - who needs that anyway? Besides the blind?

Well, what SoC is capable of doing this today with a minimum level of errors in any language on the planet?
The correct answer is that even the super-computers in the Top 10 are not capable of this. Not to mention the ridiculous models that fit into children's NPUs of consumer SoCs.

I repeat once again - the modern fashion for "AI" is nothing more than a convenient reason for increasing hardware prices, because even large data centers with giant networks that are several orders of magnitude superior in performance to an ordinary SoC are not able to solve this seemingly "banal" problem. household tasks.

It will take at least another 50-70 years before the average consumer actually gets in their pocket something similar to something that is capable of solving the above 2 banal everyday problems from the point of view of the average person.

The capabilities of ordinary hardware must increase by hundreds of millions of times + thousands of new technologies at all levels in order for such solutions to become possible on a mass scale. And this will really change civilization - when language barriers are almost completely erased, and communication with the machine will be in an ordinary voice (and then at the level of transmitting mental commands).

It's like trying to build a household refrigerator 1000 years ago. Even understanding the goal - without thousands of related technologies and industrial chains, this was impossible.

Even rifled firearms became possible only after the advent of hundreds of other technologies.

"Any sufficiently advanced technology is indistinguishable from magic" (c) A. C. Clarke

[George]

*sigh*

Surely it is all about the Western & modern marketing economics that drive EVERYTHING.

Oh? You have a perfectly functional non-electronic device? Better get a electronic one as they are BETTER!

Oh? Your device is 'dumb'? Better get a 'smart' device as they are BETTER!

Oh? Your smart device does not have AI? You better get one as they are BETTER!

As the sales/marketing engines around the world shift into "high gear" that "AI" is the "must have" feature EVERYTHING has to have included!

So for the most part WIN10 is still more popular than WIN11.

Sadly for MiniSquish and countless OEM's the sheep & lemmings are NOT replacing their otherwise perfectly functional devices with models that will run MS's latest OS.

Forgetting for the moment that the OS is CONSTANTLY changing is one of the reasons MANY refuse to use it the idea that "AI" will save the day (industry) is now fully baked into the sales & marketing and how WELL (or not!) devices make use of the feature NOBODY wanted or asked for will be flooding consumers with information.