News Lunar Lake's integrated memory is an expensive one-off — Intel rejects the approach for future CPUs due to margin impact

[Admin]

Panther Lake, Lunar Lake, and their successors will not have on-package memory.

Lunar Lake's integrated memory is an expensive one-off — Intel rejects the approach for future CPUs due to margin impact : Read more

 

[DS426]

Well at least they didn't get low volume on low margins. Wouldn't one normally price a niche market product for having higher margin on lower volume? It just seems like Pat is contracting himself. I mean if anything, most of the margin drop is due to using, TSMC, yeah?

They must be banking on LPCAMM2 being a big leap forward to say they won't do this again.

 

[Li Ken-un]

most of the margin drop is due to using, TSMC, yeah?

And offending them

That’s some market power right there. For TSMC to take back the 40% discount and Intel to suck it up despite the increased cost.

 

[Notton]

Classic intel design win they ditch way too early.

 

[JRStern]

What a stupid announcement. It's probably a strong marketing insight that integrated memory is a strong plus for client systems.
It's a quick and easy way to get a strong performance boost.
In fact they should probably integrate some gigabytes of flash in there, too - except flash has limited life, but that's *good*, it'll sell more Intel replacement processors!

 

[Alvar "Miles" Udell]

It's a design they can revisit when it makes more financial sense, or even apply their design knowledge for a semi-custom or custom order as there are plenty of applications for a more compact SoC that don't need to be cutting edge quick.

 

[Kamen Rider Blade]

Or they can finally move onto CAMM modules as promised if they wanted "Low Profile" memory while having flexibility.

Doesn't matter if it's LPDDR5 or standard DDR5, CAMM can take care of it.

It's a form factor optimized for NoteBooks.

Start using the bloody thing, what's the hold up?

 

[TheHerald]

They didn't know that before designing lunar lake? What they smoking?

 

[Pierce2623]

Well at least they didn't get low volume on low margins. Wouldn't one normally price a niche market product for having higher margin on lower volume? It just seems like Pat is contracting himself. I mean if anything, most of the margin drop is due to using, TSMC, yeah?

They must be banking on LPCAMM2 being a big leap forward to say they won't do this again.

LPCAMM2 doesn’t offer any of the benefits from the on-package memory, like reduced latency and power usage. It’s literally just a different form factor than SODIMMs, where they put both sticks into one module.

 

[Root Canal]

It's a quick and easy way to get a strong performance boost.

Snapdraggon EliteX would have crushed Intel but for this.

 

[SunMaster]

Perhaps it¨s a way of saving face vs. ending the TSMC cooperation (in the not too distant future). It would perhaps have been solid economically with 40% discount with TSMC.

 

[AkroZ]

They didn't know that before designing lunar lake? What they smoking?

Lunar Lake was a joint effort between Intel and TSMC with CPU, GPU and NPU tiles produced by TSMC.
TSMC was giving a 40% discount to Intel.
Pat Gelsinger disparaged TSMC to acquire clients for Intel Foundry, so TSMC has canceled the discount.
It's why some months ago Intel was labeling Lunar Lake as a flagship and now It's labeled as an experiment that will not be reproduced.

 

[thestryker]

A lot of people don't seem to understand what's gone on or is going on for some reason. I thought the article did a pretty good job of laying it out there.

LNL was being designed well before Microsoft decided to dump a boatload of advertising money into "AI PC". It was never going to be a good margin part because of the twin cost of external foundry and on package memory. This is also why despite tripling their order from TSMC it's still not going to be a big money maker despite selling well.

The article explained very thoroughly the problem with using on package memory for Intel so if someone doesn't understand it... try reading the article.

As for the 40% discount nonsense it's just that. There's no way TSMC is offering Intel a 40% wafer discount on their leading edge node which had the first run entirely bought by Apple. Reuters has been publishing unsubstantiated hit pieces on Intel for a while now and really shouldn't be trusted unless they're putting people on record.

 

[Alvar "Miles" Udell]

Remember the AMD Fury series? That was a one-off too, but no doubt it gave them experience they used for their future 3D Cache chips.

 

[SunMaster]

As for the 40% discount nonsense it's just that. There's no way TSMC is offering Intel a 40% wafer discount on their leading edge node which had the first run entirely bought by Apple. Reuters has been publishing unsubstantiated hit pieces on Intel for a while now and really shouldn't be trusted unless they're putting people on record

I can choose between believing Reuters, and the report they're quoting, or you, then.
Reuters being non-thrustworthy only really flies in conspiracy-land.

 

[Aart de Vries]

I specifically bought a Lunar Lake laptop because of the integrated mem (258V with 32GB) over an AMD Strix Point model, and I'm loving it. It is give or take just as fast as my 5600X desktop, so I dug up an old USB-C docking station, and it now replaced my desktop. As a Java developer I want a lot of memory bandwidth, and this is giving me that. Both Single and Multi-thread performance is higher on mine than what is posted in most initial reviews, certainly after the recent firmware update (it was even noticable as a user in the front-end). The only thing I think Intel got really wrong here is the core configuration. When this chip would give up 2 high performance cores, and got 8 more efficiency cores at almost no cost to the die-area, Lunar Lake would have been a killer in the benchmarks not only for SingleThread, but MultiThread as well, as most of the MultiThread performance comes from the e-cores (as the P-cores clock down too low due to heat/wattage constraints). I think 2 high performance cores would be more than sufficient for most users, as single core tasks still dominate in most software, and a user would be hard pressed to load-up their computer with more than 2 such processes running simultanously under normal useage scenarios for a prolonged period of time.

 

[thestryker]

I can choose between believing Reuters, and the report they're quoting, or you, then.
Reuters being non-thrustworthy only really flies in conspiracy-land.

What report are they quoting? Here's a hint: they aren't because there isn't one. They've been using nothing but anonymous sources for all of their hit pieces. This reporting hasn't been backed up by any other media outlets either. You feel free to believe that in 2021 TSMC was giving Intel a 40% discount on N3 wafers though.

 

[usertests]

Intel is sacrificing on-package memory (which is not magical but improves power efficiency which is great for mobile) because they couldn't pass along the costs to the OEMs or consumers. This news is very bad. Intel can't even innovate properly anymore. Adamantine cache is also considered to be dead.

There were some weak arguments against Lunar Lake's on-package memory. 16 GB and 32 GB serves most. Adding 24 GB or 48 GB options could help (24 GB is a gaming sweet spot). Soldered memory is already ubiquitous in the industry.

It will be interesting to see if AMD's mobile bets pay off. They want to make mobile dGPUs but struggle to get adoption. Strix Halo is a big, innovative chip that will require some hard work and marketing to make a permanent fixture on roadmaps. It could become a big deal or end up being a one-off like Lunar Lake if it doesn't go well.

 

[bit_user]

This allows for the saving of plenty of space inside laptops, as memory modules (and soldered-down memory chips) occupy a lot of space. That area can be used to install bigger batteries, ensuring greater battery life or some additional logic to boost functionality. As a bonus, on-package memory can help to cut latencies and power consumption.

Nope. It's not about latency. The distances are far too small for it to make any difference, there. The benefits are just space and power.

However, on-package memory means that Intel needed to procure these LPDDR5X devices at prices higher than those available to large OEMs. This, for obvious reasons, affects Intel's own profit margin. Handling that memory and installing it on the package also costs money, another factor that affects the profitability of the Lunar Lake product.

This is also some sketchy reasoning. I think the real issue might have to do with the cost of die-stacking required to fit that much capacity on package, as well as the cost of integrating it. If I'm right, then the added DRAM costs can't be passed on to the consumer, because they just see a laptop with 16 GB of RAM (for instance). You can't really argue they should pay more for it than one which uses chips soldered on the motherboard, because it's the same RAM, it performs the same, and neither is upgradable. So, Intel might be having to eat part of that cost delta.

Oh, and another thought I had is maybe OEMs don't like it due to the lack of repairability. Even with soldered-down RAM, it's possible to replace a bad chip. However, when the RAM is integrated with the CPU, you'd probably end up throwing out not only the CPU, but even the entire motherboard. This would increase the cost of warranty coverage.

 

[bit_user]

In fact they should probably integrate some gigabytes of flash in there, too - except flash has limited life, but that's *good*, it'll sell more Intel replacement processors!

SSDs have a best-case read latency on the order of about 10 microseconds. An electrical signal can travel about 1.5 km, in that amount of time. So, please explain to us what you think the benefit would be of having NAND flash chips so closely integrated with the CPU.

 

[bit_user]

Doesn't matter if it's LPDDR5 or standard DDR5, CAMM can take care of it.

What do you even mean by that? First of all, there are different CAMM form factors for LPDDR5 vs. regular DDR5. Secondly, CAMMs don't magically erase the downsides of using regular DDR5 vs. LPDDR5.

LPCAMM2 doesn’t offer any of the benefits from the on-package memory, like reduced latency and power usage.

It improves signal integrity, which enables higher speeds. I'm pretty sure it uses more power than on-package memory, but perhaps a little less than SO DIMMs. However, no benefits you can't get with soldered-down LPDDR5(X), aside from obviously repairs and upgrades.

 

[bit_user]

Remember the AMD Fury series? That was a one-off too,

Fury was followed by Vega. AMD had no flagship GPU, in between.

Then, Vega was followed by Radeon VII. After that, I guess strides in GDDR memory meant that the added cost of HBM was no longer warranted, in a consumer-grade GPU.

but no doubt it gave them experience they used for their future 3D Cache chips.

Huh? One is putting multiple dies on a (probably passive) interposer, while the other is highly-sophisticated silicon-bonding. There's no comparison.

That said, maybe it gave them the experience & confidence to use the chiplet approach they adopted in EPYC. However, even then, it's not like multi-chip modules were a remotely new thing.

 

[bit_user]

Snapdraggon EliteX would have crushed Intel but for this.

LOL, wut?

Source: https://chipsandcheese.com/p/lion-cove-intels-p-core-roars
​

As a Java developer I want a lot of memory bandwidth,

As a Java developer, maybe you should've gone with Snapdragon X?

Both Single and Multi-thread performance is higher on mine than what is posted in most initial reviews, certainly after the recent firmware update (it was even noticable as a user in the front-end).

AFAIK, the updates just tuned the performance/efficiency bias, for different power plan settings. You're probably getting more performance, at the expense of worse battery life or at least more fan noise.

The only thing I think Intel got really wrong here is the core configuration. When this chip would give up 2 high performance cores, and got 8 more efficiency cores at almost no cost to the die-area,

Nah, Skymont is bigger than you think. 2 P-cores is only worth about 5.25 E-cores, but they come in clusters of 4. So, what you'd likely get is a 2P + 8E configuration, which is actually something Intel did with their Alder Lake-U, if I'm not mistaken.

Snapdragon-X could give you up to 12 P-cores.

 

[Giroro]

Memory manufacturers presumably make some amount of profit off of DRAM, even if that margin might be lower than what Intel prefers to make from its CPUs. Intel could take some amount of profit from selling the integrated DRAM, or they could make zero profit by giving the DRAM sales back to the memory OEMs. It's a very Intel-like choice to prefer zero profit over less profit.

I think the real problems with Lunar Lake is that they have completely failed at marketing it on every possible level. It's a shame because it could have been a neat and relatively interesting product. They completely overcomplicated the product line with too many boring SKUs, seem to be failing to win over Laptop manufacturers, and have lost their only chance to excite retail customers.
They didn't even give it a real brand name to replace the development code word, we still have to call it Lunar Lake.
Your best buy sales rep could be hyping lunar lake as the key to unlocking super thin fancy Ultrabooks that last forever (Sure we've had those products for a decade, but it at least worked at some point). But instead, customers are just being told that this year's Ultra i9 200 Whatever V product has less cores and lower performance than previous desktop-replacement i9s, so they should just keep buying AMD. Or, customers hear about this new, easy to understand, and well-branded Apple M4/3/2/1 processor and come in to specifically ask about it - even people who have never used a mac.
Newsflash Intel, nobody cares about your stupid new meaningless Ultra®️ numbering system. You pumped money into pushing out a rebrand without actually bothering to come up with a new brand, and all that has accomplished is to further confuse customers.

It's like they aren't even trying to win back customers or to even actively sell the product. They're still assuming that everybody will always buy intel by default - which hasn't been true for awhile. How's that been working out for them?

 

[bit_user]

Memory manufacturers presumably make some amount of profit off of DRAM, even if that margin might be lower than what Intel prefers to make from its CPUs. Intel could take some amount of profit from selling the integrated DRAM, or they could make zero profit by giving the DRAM sales back to the memory OEMs. It's a very Intel-like choice to prefer zero profit over less profit.

It'd be interesting if Intel worked with multiple DRAM suppliers and gave customers a choice. Then, you could conceivably have the DRAM vendor doing some of the legwork to sell these CPUs and the competition between them might help offset some of the higher costs associated with the on-package solution.

They didn't even give it a real brand name to replace the development code word, we still have to call it Lunar Lake.

Yeah, but they didn't give Raptor Lake a real brand name, either. It was just launched as Gen 13, except of course not all of the Gen 13-branded CPUs even had a Raptor Lake die!

Your best buy sales rep could be hyping lunar lake as the key to unlocking super thin fancy Ultrabooks that last forever (Sure we've had those products for a decade, but it at least worked at some point). But instead, customers are just being told that this year's Ultra i9 200 Whatever V product has less cores and lower performance than previous desktop-replacement i9s, so they should just keep buying AMD.

Yeah, maybe they need to better distinguish their different product segments. Like, they could use Lytrium to brand their "thin & light" CPUs, Econium for lower core-count value CPUs, Rushnium for high-clocking CPUs with mid core counts good at gaming and moderate creative tasks, and Worknium for high core-count workstation-oriented models (okay, dumb names, but you get the point). You could still have 3/5/7/9 tiers within each, but it would be a less confusing situation than having like a i5 HX model that's faster than a i9 U model, for instance.

I think if you put the product line first, instead of as a suffix, then it's less surprising if like a Worknium 5 is faster than a Lytrium 9. People would be like "duh, of course it's faster - it's a Worknium. Lytriums are for thin & light."

Or, customers hear about this new, easy to understand, and well-branded Apple M4/3/2/1 processor and come in to specifically ask about it - even people who have never used a mac.

I do take issue with their use of "Pro" for one of their product tiers. Plus, uh, "Max"? You're going to name a Mac CPU tier something that's a homonym of "Macs"? Call me unimpressed. And then Ultra, when you already have Max? Isn't Max already the maximum? Why is there a tier above that?

But yeah, it's simple and once people know it, they probably don't have much trouble remembering.

Newsflash Intel, nobody cares about your stupid new meaningless Ultra®️ numbering system. You pumped money into pushing out a rebrand without actually bothering to come up with a new brand, and all that has accomplished is to further confuse customers.

I have to agree with this and your point about them creating too many SKUs. I think Apple has only like 2 SKUs based on each die, and most of their range is covered by what are actually different dies!