News Intel unwraps Lunar Lake architecture: Up to 68% IPC gain for E-cores, 16% IPC gain for P-Cores

[usertests]

We'll see. One handheld design win for MTL was MSI Claw, and it was savaged in reviews--to wit, The Verge's "The MSI Claw is an embarrassment." Maybe the Claw successor based on LNL will fare better. Then again, maybe not.

MSI Claw is doing better now, with double-digit gains after driver updates. A jump over that level of performance while cutting power consumption will make Lunar Lake good for 1080p. We'll see if AMD can do better. RDNA3.5 seems good but they haven't given enough information yet. Kraken Point (4+4) would be an obvious chip to compare to Lunar Lake but it's probably launching in 2025.

I'm not sure you realize how petty that sounded.

I will take any opportunity to bash them, especially if they get the basic facts horribly wrong.

 

[Evildead_666]

Wiktionary disagrees with you, citing quotations from as far back as 1857.

​

https://en.wiktionary.org/wiki/architect#Verb​

​

Indeed.

The further back we go, the less knowledge people had of the English language.
I would have thought the Electrical Review back then would have had better vocabulary
An American author could be forgiven, as they do it just to annoy the British, see zucchini and Eggplant and pants for examples (the scallywags!).

The Oxford English dictionary requires a login, so they won't get a mention unfortunately.

Architected/ing, doesn't sound right, and there are other publications that refute its verb status.

I dont want to get all in a bother about it, i've already polluted this thread enough.
We just shouldn't be encouraging its usage

(And no i don't give a crap about missing punctuation on "teh internet". Lol)

On topic :
Looking forward to the hands on, real world tests

Thanks muchly.

 

[peachpuff]

Really 5 (five) pages !!!

Just buy it!

 

[TerryLaze]

Cross-CCD communication isn't very common, though. Especially for NUMA-aware schedulers, which operating systems are becoming more adept at doing.

That's neither here or there though.
If foveros doesn't have the lag, again I don't know if it does or not, then the intel quote was and still is justified.

 

[t3t4]

I'd like to see extensive power consumption/performance benchmarks for the next gen. Intel, AMD, ARM CPUs and how slower and faster SSDs affect the battery life.

Far too long I've been asking for power vs performance vs thermal results.... All anyone keeps measuring though is this chip greater than/less than the last chip in some meaningless synthetic benchmark. So I wouldn't hold your breath waiting on this data if I were you. Likely the only way to get the data you seek is to formulate, test and measure everything yourself.

 

[redgarl]

If you remember how much a power hog was the last generation, then you understand how Intel can spin the numbers concerning E-Cores.

The fact they are comparing LP E-Cores is saying everything. Be ready for another 300+ Watt CPU for about 16% IPC uplift, same as AMD, no surprise in the industry.

 

[abufrejoval]

Am I the only one to think that leaving out external DRAM should at least receive a mention?

It may well be the new normal for ultrabooks, but does it mean that Intel is yielding the complete H segment to Strix Point?

Or are we likely to see some CAMM2 equipped variants soon?

 

[Eximo]

Am I the only one to think that leaving out external DRAM should at least receive a mention?

It may well be the new normal for ultrabooks, but does it mean that Intel is yielding the complete H segment to Strix Point?

Or are we likely to see some CAMM2 equipped variants soon?

One chip at a time. They are announcing what they have nearly ready.

It has been somewhat typical for Intel to start with ultrabooks "U", then move on to laptop "H", and finally do desktop "S"

Though I want to say Alder lake desktop came first which was atypical. Q4 2021 Alderlake S vs Q1 2022 for Alderlake U/H

 

[Phil995511]

We don't want machines with RAM soldered to the motherboard !!!

 

[Notton]

We don't want machines with RAM soldered to the motherboard !!!

Well this one is soldered to the substrate, which means the RAM is part of the CPU, not the mobo.

 

[Phil995511]

Well this one is soldered to the substrate, which means the RAM is part of the CPU, not the mobo.

It amounts to exactly the same...

 

[abufrejoval]

Well this one is soldered to the substrate, which means the RAM is part of the CPU, not the mobo.

Since you're nitpicking: and the substrate is soldered to the mobo...

Point being: they are inseparable and a key element of personal computers, configurability and the ability to change/upgrade, is lost.

 

[abufrejoval]

One chip at a time. They are announcing what they have nearly ready.

The part that I am missing a lot is the future projection.

Supporting both on-substrate RAM and something external, even if it's as similar as CAMM2 with LPDDR5, will be almost punitively expensive in terms of amplifiers, beachfront and perhaps even require extra memory channels, although there the 32-bit width of LPDDR may help.

So something like Apple's tiling approach used at least for one generation of Ultra almost seems likely for anything up to say 65 Watts.

And with that the original PC desktop sweet spot around 130 Watts and discrete GPUs and RAM becomes awfully small, especially with gamers just abandoning PCs and consoles for Strix Point like mobile consoles (or gamer laptops), doing double duty as desktops via TB.

 

[Notton]

Since you're nitpicking: and the substrate is soldered to the mobo...

Point being: they are inseparable and a key element of personal computers, configurability and the ability to change/upgrade, is lost.

Yeah, but it's not like it's locked down with a useless amount, like 8GB.
It has 16GB minimum on a low power platform.
I kind of don't see it benefiting from more than 32GB either.

 

[Notton]

It amounts to exactly the same...

DDR5 SODIMM maxes out at 5600
CAMM2 LPDDR5 maxes out at 7500
Lunar Lake gets 8500

So you are saying you want to use slower RAM when the iGPU is going to be starved for memory bandwidth?

 

[abufrejoval]

Yeah, but it's not like it's locked down with a useless amount, like 8GB.
It has 16GB minimum on a low power platform.
I kind of don't see it benefiting from more than 32GB either.

That exactly why the PC was so successful for decades: it supported usage scenarios its designers didn't anticipate ("640k should be enough for everyone").

Part of my work is simulating and testing infrastructures. I do that with VMs, sometimes quite a few. So I use relatively low-cost hardware (NUCs + PC based workstations) with lots of RAM. And some of my laptops allowed for 32/40/64GB expansion, which helps with local demos and tests.

RAM in DIMMs is cheap, so cheap it's easier to just buy more than thinking about how to shoehorn a workload into less.

But that's changing when vendors can charge punitive prices on RAM capacity.

 

[Notton]

That exactly why the PC was so successful for decades: it supported usage scenarios its designers didn't anticipate ("640k should be enough for everyone").

Part of my work is simulating and testing infrastructures. I do that with VMs, sometimes quite a few. So I use relatively low-cost hardware (NUCs + PC based workstations) with lots of RAM. And some of my laptops allowed for 32/40/64GB expansion, which helps with local demos and tests.

RAM in DIMMs is cheap, so cheap it's easier to just buy more than thinking about how to shoehorn a workload into less.

But that's changing when vendors can charge punitive prices on RAM capacity.

I disagree. The PC was only successful because of software.

And your work requiring 64GB is definitely in the minority.
Not everyone needs 64GB, and this chip isn't for you, so go find some other chip to use.

 

[abufrejoval]

DDR5 SODIMM maxes out at 5600
CAMM2 LPDDR5 maxes out at 7500
Lunar Lake gets 8500

So you are saying you want to use slower RAM when the iGPU is going to be starved for memory bandwidth?

iGPU very rarely uses all RAM, unless it's actually/finally GPGPU or HSA.

That's why I mix a 32GB SO-DIMM with 8GB of onboard RAM with my Alder-Lake notebook to run VMs: the iGPU gets dual channel bandwidth, the CPU may have to contend with somewhat less DRAM bandwidth for half of the RAM.

And in that scenario the hypothetical question is: do I want to starve RAM and start paging or do I starve game performance on a laptop?

 

[abufrejoval]

I disagree. The PC was only successful because of software.

And your work requiring 64GB is definitely in the minority.
Not everyone needs 64GB, and this chip isn't for you, so go find some other chip to use.

Thank your for your clear advice as to exactly what I should do!

 

[NeoMorpheus]

As usual, media runs wild the whatever intel marketing team says and as usual, they conveniently forget that they have always lied.

When real test are done by unbiased reviewers, then we can talk, until then, i don’t believe one word from intel.

 

[Roland Of Gilead]

My question is, what does this mean for Intel's 5N4Y plan?

Sitting in mid 2024, and the expected 5th node to come before the end of 2025, when in actuality they've only achieved one of those nodes. Being on TSMC for this round, means 3 other nodes in 18 months. I just don't see it.

 

[bit_user]

An American author could be forgiven, as they do it just to annoy the British,

I've got no time for this. Please take your nationalism elsewhere.

Architected/ing, doesn't sound right, and there are other publications that refute its verb status.

I don't care what sounds right to your ear. The core activities of an architect deserve a more descriptive verb than "designing", since an architect can do design work that wouldn't be classified as the creation of architecture.

I'm sure we could agree on some abuses of the English language (don't get me started on "learnings"), but this isn't one of them.

I dont want to get all in a bother about it,

It's a bit late for that!
: D

 

[bit_user]

If you remember how much a power hog was the last generation, then you understand how Intel can spin the numbers concerning E-Cores.

In this case, the "last generation" would be Lunar Lake, which instead of being a power hog was an underperformer (see also: Ice Lake).

The fact they are comparing LP E-Cores is saying everything.

I think they have an extremely narrow justification for comparing to the LP E-cores, which is that they wanted to show why they dropped the Low Power island and are just relying on the CPU tile's Skymont cores. However, the fact that all of their comparisons use the LP E-cores is indeed a bit much.

 

[bit_user]

It amounts to exactly the same...

No, putting the RAM on-package saves space and power. It might also let them run higher speeds.

If there weren't substantial benefits, they wouldn't do it because it wouldn't be popular with OEMs and adds cost & complexity.

Since you're nitpicking: and the substrate is soldered to the mobo...

No, it's not. Also, it's not a nit-pick because off-package memory could also use LPCAMM2, which is not an option if it's on-package.

That exactly why the PC was so successful for decades: it supported usage scenarios its designers didn't anticipate ("640k should be enough for everyone").

That's a non-sequitur. He was talking about how to partition up an address space of approximately 1 megabyte. The 640k part was in reference to how much of it to set aside for OS + programs. It was never meant to apply to CPUs that supported > 20-bit addressing, at which point he probably figured you'd have a different operating system with its own memory map.

Part of my work is simulating and testing infrastructures. I do that with VMs, sometimes quite a few. So I use relatively low-cost hardware (NUCs + PC based workstations) with lots of RAM. And some of my laptops allowed for 32/40/64GB expansion, which helps with local demos and tests.

RAM in DIMMs is cheap, so cheap it's easier to just buy more than thinking about how to shoehorn a workload into less.

Then I predict you'll like CXL.mem expansion modules.

 

[ingtar33]

intel tends to conflate IPC with general performance gain from clock speed increases in most of their marketing. so you'll often see things like 68% ipc gain, but what they're comparing it to is a 2.4ghz part vs a 3.2ghz part, meaning almost all the 68% "ipc gain" is from a clock speed improvement. meaning their marketing department very rarely uses IPC correctly confusing the very meaning of the phrase.

if they do get 68% IPC gain that's great, but i doubt it. most intel marketing not only stretches the definition of IPC but usually over estimates the gains by 15-20%