Intel Core Ultra 200V Lunar Lake Family Leaks: Nine Models with One Core 9 Ultra SKU

[ColonelPhantom]

If that's true, these are going to be such a mixed bag. There's supposed to be a 15% IPC P-core gain for Lunar Lake over Meteor Lake, but Meteor Lake topped out a 6+8 with HTT. So you'll get better single core, but it's probably not going to match on multi-core. Yeah, there's a more powerful GPU and NPU, but what will that actually amount to in real life use?

Lunar Lake is the successor to MTL-U, not MTL-H. MTL-U tops out at 2+8+2 cores. The LP-E cores should also probably be disregarded, giving 2+8. As such, going from MTL-U to LNL we can see 4 fewer E-cores (= 1 cluster, similar to 1 P-core in die size), and together with hyperthreading, it was traded for 2 more P-cores.

 

[Vayra86]

If that's true, these are going to be such a mixed bag. There's supposed to be a 15% IPC P-core gain for Lunar Lake over Meteor Lake, but Meteor Lake topped out a 6+8 with HTT. So you'll get better single core, but it's probably not going to match on multi-core. Yeah, there's a more powerful GPU and NPU, but what will that actually amount to in real life use?

Single thread perf is still king in everything that's not parralelized... you need the extra cores just for background tasks, but 95% of users have just one primary task they work on.

 

[john_]

EVERY single CPU in this list is 4P + 4E cores, which is hilarious.

Really? I was looking at that list and was thinking "TSMC's 3nm is proof that Intel engineers can create good stuff when (Intel's) manufacturing isn't an obstacle".
Now you say that these are 4+4 designs and I am like "LOL....NO...my mistake".

 

[Seeprime]

Putting 200V in the name made me initially think it requires 200 volts.

 

[john_]

Given that Meteor Lake's biggest SKU was a 16-core part (6 + 8 + 2), this does not bode at all well for Arrow Lake - especially given that Hyper-Threading is now gone! Unless Intel has managed to increase IPC by 100%, which they obviously have not, these CPUs are going to be far worse than what we've seen before.

Hyper Threading gives about a +30% extra performance to one P core. So, if the P core can get smaller by removing HT as much as needed to leave enough space for, let's say, half E core, Intel can build and advertise 8 core CPUs, instead of 6 core CPUs (4+2) with 10 threads with the same die space. Considering that the majority of consumers only read core count and not thread count, from a marketing perspective removing HT is the better option for Intel.

These CPUs will probably end up being great for handhelds and cheap/mid range laptops , if of course they don't end up too expensive.

 

[P4-630]

200 x as powerful than Volta....

 

[Vayra86]

Ah... Raja was just a decade too early with his Poor Volta comment then. It all clicks now

 

[john_]

I care about being able to upgrade and expand.

The biggest problem here I think is being able to replace.
Bad RAM?.... Change one dimm.
In this case?
Bad RAM?..... Change the laptop? What?

 

[Denver]

View attachment 1000017563.webp
It seems that the next generation is not so "revolutionary" after all, which puts Lunar Lake ST 10-11% slower than Zen5 at similar clockrate. But it's worth noting that Geekbench is a terrible benchmark.

 

[WatchThe80s]

Finally, no more laptop with 4-8GB non-upgradable RAM, at least from Intel. Though, i don't know if it's worth the 4P+4E Cores to have more RAM.
This might be handy if they put in handhelds.

 

[bug]

This is an instant deal-breaker. Not acceptable at all.

It's a SoC, the memory is on package...
And it's meant for ultra-portables, those things haven't been upgradable in ages.

 

[Evrsr]

Given that Meteor Lake's biggest SKU was a 16-core part (6 + 8 + 2), this does not bode at all well for Arrow Lake - especially given that Hyper-Threading is now gone! Unless Intel has managed to increase IPC by 100%, which they obviously have not, these CPUs are going to be far worse than what we've seen before.

Do you care about benchmarks or sustained loads? Also, higher end parts will be based on desktop chips, not something with packaged memory.

 

[phanbuey]

The biggest problem here I think is being able to replace.
Bad RAM?.... Change one dimm.
In this case?
Bad RAM?..... Change the laptop? What?

Thin and lights have been like this for years. Snapdragon x and Apple same thing. They're basically glued shut.

 

[Ruru]

EVERY single CPU in this list is 4P + 4E cores, which is hilarious.

Reminds me of the old days when they sold dualcore i7 mobile CPUs.

 

[Bobaganoosh]

You missed my point, the bold and underlined point. I don't give a flying rats bum what people think is enough. I care about being able to upgrade and expand.

You're supposed to throw it away after a year or two and buy a new one. -intel (probably)

 

[Patriot]

Now. Not in 4 years. (or 7, or 11, for those of these that survive).
8GB was fine 6 years ago.

in 4 years, you arent going to want 4 non-ht pcores and 4 e cores.

 

[Fouquin]

You missed my point, the bold and underlined point. I don't give a flying rats bum what people think is enough. I care about being able to upgrade and expand.

This segment hasn't been upgradeable in awhile. The ULV chips that are intended for the consumer market are mostly strapped to soldered LPDDR memory. I do think with LPCAMM which can house an upgradeable 4x32-bit LPDDR5x RAM config on a single tiny PCB we should be seeing expansion options in this market. Yes, the primary RAM is soldered to the SoC's package for the benefit of power and performance, but you could also be allowed to buy an LPCAMM with LPDDR5x-7500 to increase the size at the cost of some performance. Or not, it could treat the SoC DRAM as a defacto write-through L4 with the iGPU reserving its 2-4GB similar to how Intel has previously employed eDRAM configurations as both system memory and high-speed iGPU buffer.

 

[InVasMani]

Intel could eventually in theory could pair like 8 memory chips all the way around CPU die for more capacity and bandwidth. All they need to do arrange it the right way so around a square chip die and they can just connect it along half of each edge length side of the chip die. They need a certain amount of space to do something like that, but still x8 the capacity and bandwidth of a individual chip or x4 the capacity and bandwidth of just two chips like illustrated.

So eventually we might see something like basically a 128GB with 68GT/s chip out of Intel with this kind of technology. Depending on size of chip and socket he memory socket could be repurposed for cutting edge for M.2 storage Gen 5, or Gen 6 down the road. They could also potentially tuck a M.2 slot directly in back of the memory package though keeping it close to memory and CPU which would be about the same either way. I think just replacing the DIMM slots with like 2 to 4 M.2 slots would be more sensible if they've already got plenty of on package memory.

This is clearly a mobile device though, but I'm looking at desktop space implications of it a bit and it's certainly interesting and could indicate a lot of changes coming in the future to desktops.

The core count on these are pretty tame, but for low power mobile I really don't think the majority of consumers need more or are looking for more necessarily. The AI tech is probably not that important to most consumers at this stage however. People are pretty reluctant and dubious about it at this stage. I do think there is a market for it that will embrace the technology and leverage it, but is it a big enough one for the average consumer is tougher to analyze at this stage. I do feel more people will reluctantly at first start to embrace AI over time however once actually start engaging with it more and figuring out more about how, where, and why it's useful to get more efficient usage out of it. It's certainly not going away AI is here to stay.

I see it as more useful than VR ever appeared to me honestly. Like I never really got the big buzz around VR it didn't come across to me as a particularly great technology and full of gimmicks and design flaws and problems to sort out. I wouldn't say VR has no potential, but it doesn't seem as a useful as it was hyped at the same time. I don't want a sweaty face mask that's heavy on my head for hours with dubious resolution, refresh rate, and performance requirement implications. I think the head tracking technology itself is cool, but otherwise I really don't see anything about VR that would make me want to engage with it over a desktop mouse and keyboard or standard game controller. Like what is the point other than filtering out some things in my peripheral vision!!?

I don't think 4P 4C is terrible by any stretch for mobile device on core count. It's sufficient enough especially for every day consumers. It can keep costs lower and at the same time keep heat and power more in check as well so I think that's mostly fine for the target audience in mobile market space. I mean I don't think the average laptop users is looking to cram a RTX 4090 in it with CPU with as many cores as possible. That's what a enthusiast might want, but just someone looking to pick up a cheap laptop not so much. Battery life is important and as long as that's good I think these should be fine if priced appropriately.

 

[bug]

This segment hasn't been upgradeable in awhile. The ULV chips that are intended for the consumer market are mostly strapped to soldered LPDDR memory. I do think with LPCAMM which can house an upgradeable 4x32-bit LPDDR5x RAM config on a single tiny PCB we should be seeing expansion options in this market. Yes, the primary RAM is soldered to the SoC's package for the benefit of power and performance, but you could also be allowed to buy an LPCAMM with LPDDR5x-7500 to increase the size at the cost of some performance. Or not, it could treat the SoC DRAM as a defacto write-through L4 with the iGPU reserving its 2-4GB similar to how Intel has previously employed eDRAM configurations as both system memory and high-speed iGPU buffer.

It may not be that simple. At the operating frequencies of current DRAM the power requirements to send a signal 5mm away are very different from those for sending it 10cm away. Adding enough power for both scenarios is probably not worth it. And it would take a toll on energy efficiency regardless.

 

[Fouquin]

It may not be that simple. At the operating frequencies of current DRAM the power requirements to send a signal 5mm away are very different from those for sending it 10cm away. Adding enough power for both scenarios is probably not worth it. And it would take a toll on energy efficiency regardless.

10cm is a bit of hyperbole. We haven't seen trace runs like this in any mobile form factor since switching to DDR5 which has a maximum CLK and CMD line length of just over 6cm. Not that the DDR5 spec even applies here because we're talking about LPDDR. The advantage of LPCAMM is that the physical interface can be placed within close proximity of the SoC, even butted right up to it, and the package vias are situated directly below the memory modules. The trace length of such an implementation would be well within bounds of 7500MT/s but obviously wouldn't be able to synchronize with on-package 8533MT/s memory, hence why I suggested it would take a performance hit by dropping to 7500MT/s. The power requirements may be higher but still drastically lower than traditional DDR.

 

[R0H1T]

Reminds me of the old days when they sold dualcore i7 mobile CPUs.

No the hilarious part is they sold i3/i5/i7 all as dual core ULV chips right till KBL, when Zen launched

 

[lexluthermiester]

you missed the point that this is a low power solution for thin and light laptops.
where ram is soldered on the motherboard to get them as thin and light as possible already.

it changes nothing

For ultra portables, sure. But not all laptops are ultra portable. Some of us expect upgradeability regardless of form-factor. You discounting that doesn't change the expectation.

Upgrade? The benefit of this is that LPDDR5-8533 memory is likely much faster than regular SODIMMS..

Maybe, maybe not.

When you need 64GB RAM+ on your laptop you're probably doing it wrong and you better use a desktop PC IMO.

There is a number of flaws in that logic. How people use their computing devices is up to them and should not be limited by short-sighted narrow thinking.

The biggest problem here I think is being able to replace.
Bad RAM?.... Change one dimm.
In this case?
Bad RAM?..... Change the laptop? What?

And that is just one possibility. It's the same reason soldered-on storage is a really stupid idea.

It's a SoC, the memory is on package...

What? Unless Intel is trying something new, no one in the world packs system RAM on the SOC package. Fair certain Intel would have shouted that from the rooftops as a feature, which is hasn't. So no, the memory is not on the SOC package. Scratch that. I missed the pictures in the article.. This is something new from Intel.

This segment hasn't been upgradeable in awhile.

That depends on where you look. Not that I'm disagreeing. Just voicing the concerns for the makers to see so that they know there is a section of there buyers that want upgrade capability.

Yes, the primary RAM is soldered to the SoC's package for the benefit of power and performance

I think a closer examination of the specs is in order. This is not correct. No one is doing this...yet. Unless I've missed something? EDIT: Yup, I missed that. Glossed it over. I didn't look at the included images.. (suddenly feels the complete dumbass)

EDIT2: Just had a thought. The on-package RAM could act as a L4 cache for more RAM on the motherboard.

 

[N/A]

laptops memory with 1Rx16 4 chips is slower than 1Rx8 with 8. so 2 chips is probably even worse 1Rx32. Does 4x16 channels somehow mitigate the problem. I guess we'll never know.

 

[Noyand]

View attachment 352889
It seems that the next generation is not so "revolutionary" after all, which puts Lunar Lake ST 10-11% slower than Zen5 at similar clockrate. But it's worth noting that Geekbench is a terrible benchmark.

Geek bench also supports AVX512

 

[Darmok N Jalad]

For graphics performance to improve, they need more memory bandwidth, and that’s just looks to be a lot harder to attain with non-soldered RAM.