Massive leak details Intel Lunar Lake MX architecture releasing late 2024

[Redaktion]

Intel's upcoming Lunar Lake MX processors will come in Core 5 and Core 7 flavors with 4 P-cores and 4 E-cores and up to 8 Xe2-LPG Battlemage graphics cores that can almost match Apple's M2 iGPU performance. The entry-level 8 W SKU will not require active cooling.

https://www.notebookcheck.net/Massive-leak-details-Intel-Lunar-Lake-MX-architecture-releasing-late-2024.770735.0.html

[NikoB]

Intel has completely disgraced itself.

Again, shameful TB4/USB40, and not TB5/USB40 V2, again shameful eDP1.5/DP1.4, and not eDP2.0+/DP2.0+ with UHBR20 mode.

Again, the shameful 2-channel 128-bit RAM, instead of HBM3 at least 512 bits, since everything is soldered "on the chip".

And again, a shameful 16GB (yes, 32GB is also shameful for 2025) of RAM, which cannot even be expanded.

How does this fit with their assurances about progress in AI, if any locally executed advanced expert system based on neural networks requires as much RAM as possible? There should already be terabytes of RAM, and not measly tens of gigabytes. This is all reminiscent of the early days of the PC in the 90s. Only now the requirements for RAM (and its bandwidth) will grow exponentially if all this cries about "AI" actually lead to something useful in local chips.

It's funny to see how in data centers they are already looking at 2048-bit HBM4 (with hundreds of gigabytes on each chip) and all this in order to at least get closer to the mind of a 5-year-old child, but here they are still delving into the shameful and outdated ddr5 , which has long been in the trash heap.

It is the lack of RAM bandwidth that is forcing Intel (and AMD) to hesitate with the introduction of TB5/USB40 V2.0 and full versions of DP2.0+ and other high-power interfaces - what is the point in them if one port eats up literally the entire RAM bandwidth , do you need several of them?

Until the RAM bandwidth on regular PCs and laptops increases to at least 512GB/s, there will be no real progress. Forget about it. It won't be the same with local neural networks running on your equipment until the amount of RAM is measured in tens of terabytes. Not to mention long-term memory of tens of petabytes.

Did you think that neural networks are a cheap toy? No, this is an extremely expensive super task in terms of resource requirements, and there is no talk of any kind of "AI" here at all, just so that you have in your pocket a fairly advanced expert system, tailored to solve a specific range of practical problems.

[NikoB]

When will we finally get native support for 8k@120Hz on monitors and PCs/laptops, which I'm tired of waiting for more than 10 years to get banal 300ppi+ on a 27-31" monitor?

It seems that video cards with their fast GDDR6 (and in some, HBM) have long been able to support 8k@60-120Hz for 2D, but the problem is that neither Intel nor AMD (whose built-in memory controller is traditionally even weaker) can offer adequate RAM controllers have been around for over 5 years!

Let me remind you that in 2025 the pci-e 7.0 bus will be launched, which will allow you to pump data to video cards at speeds of up to 512GB/s.

And pci-e 5.0 x16 ALREADY exceeds in its requirements (128GB/s) the capabilities of controllers even in top consumer Intel/AMD for desktop processors!

Remember, I wrote many times - why did the ridiculous AMD team build a pci-e 5.0 bus into Zen4, as well as a monstrous 28(!) lines, when the 7945HX RAM cannot even provide 80GB/S on two DDR5 5200 channels!

Why do they sell this useless and meaningless garbage? Fools who don't understand the bottlenecks in architecture? So these are the majority of the world's population...

It's funny, but from China they are now selling motherboards along with Xeon processors from 2013-2014 (once costing $2000-4500 apiece) and 64GB DDR4 ECC (16x4), which, in terms of the speed of a 4-channel memory controller, surpass EVERYTHING laptop motherboards with top 7945HX! Just think about it! Processors for servers in 2013, which now cost pennies, are faster than the latest top-end AMD processors for "high-performance" HEDT laptops in terms of memory controller speed!

A whole 10 years have passed!!! And desktop and laptop processors still have shameful 128-bit memory controllers, which are not even faster than 10 (!) year-old Xeons, which now cost mere pennies! This monstrous backwardness of the Intel/AMD consumer segment is simply amazing, as is the lack of a mandatory ECC control option, which should now be mandatory on all motherboards WITHOUT exception!

[Enma45]

We have been refried at 10nm+++++ for 4 years now and this is why people are now looking for AMD Zen 4 Phoenix in all laptops but these are scarce so if you find one, don't hesitate to buy it.
With Intel we continue for 4 years with the same IRIS graphics with which it is impossible to play AAA Games, render 3D.
People are looking for AMD Zen 4 manufactured at 4nm + RDNA 3 + Artificial intelligence

[MyGoodnessMe]

Enma45 + NikoB back to back, what more could you ask for.

[Tim Cook]

Good morning NikoB and Emma45,

We are thrilled to announce the next generation of Mac computers. These are the best mac we have ever created. Available at just $7698.98

It's time for you to make the switch to mac and leave the boring pc world.

[Sinocelt]

> Lunar Lake MX chips are said to be co-developed with Microsoft

It feels a little unfair. AMD has produced better chips than Intel for many years now, yet still Microsoft priviledges Intel.

[Ordinary Man]

Is there any company that hasn't completely disgraced itself in NikoB's blessed eyes?

Intel has completely disgraced itself.

[TXW]

NikoB is a "bit" hyperbolic on the description on Lunar Lake. I think it represents Intel getting their head out of a** (maybe investors putting pressure on leadership). Lunar Lake, while not any sort technology lead forward, represents a major step forward in recognizing the marketplace needs. This is a 7000 AND 8000 U and H series  killer (and I am on Team Red).

While I agree that the next major advancement in PC cpu architecture is HBM, to really take advantage of it will require a wholesale redesign on the internal cpu architecture.
In the processor marketplace, the are 2 major segments, mainstream business laptops and data center server processors. Yes, there are a bunch of niche markets, but, in general, they don't move the financial needle much. Intel used to own both of those market places. EPYC gave XEON a run the money. But Intel still owned the business laptop processor market. With the Ryzen 3 and 4 nm processors and the integrated RDNA 3 based APU in the 7000 and 8000 series, it presented a compelling sku to the big laptop  manufacturers (of which there are only 3; Dell, HP, and Lenovo).
The AMD 7000/8000 series gave them a viable alternative. Lunar Lake gives them a better alternative.

And I guessing Intel's deal with TMSC will tie up their 3nm fab capacity for the year, maybe two. The 3nm fab machines are not easy to build and scale up. That will crush AMD near term options.

Is lunar lake any sort of technology leap? Nope. Is it a market leap? You bet. That business laptop marketplace is the bread and butter in the the processor and packaging world. And while HBM may be the future for CPU's, high density chiplet integration (and low TDP) is the here and now of the laptop marketplace.
Team Blue can still secure defeat from the jaws  of victory if they can't deliver in volume in CY24. And I have my doubts. Their management seems to lack a sense of urgency,  like they still own the current AND future markets. We will will see.

[tipoo]

This says Lunar Lake at peak > M2's IGP, contrary to this notebookcheck saying it lags behind it, in the anandtech forum image. Did you mind the greater than and less than signs lol?