Intel's Lunar Lake Looks Like a Home Run

[erek]

“A big step forward for Intel …

Intel's turnaround depends on more than Lunar Lake. The company must get its Intel 18A manufacturing process up and running, and successfully sell it to external customers. It also must win back share in the server CPU market after falling behind AMD in performance and efficiency. While this will take time, Lunar Lake looks like a positive development as Intel plots its comeback.

Third-party reviews, which should be available around the Sept. 24 launch, will shed light on whether Intel's performance and battery life claims hold water. If Intel's claims hold up, Lunar Lake will be a potent force in the laptop market this holiday season.

The massive battery life gains provided by Lunar Lake will give those with older laptops a good reason to upgrade, and they won't have to accept the downsides of a Qualcomm-powered laptop. Intel's client computing segment was its strongest in the second quarter, growing by 9% year over year following a post-pandemic slump. If Lunar Lake triggers an upgrade cycle, a big rebound could be in order for 2025.”

Source: https://finance.yahoo.com/news/intels-lunar-lake-looks-home-112000963.html

 

[DukenukemX]

Until there are benchmarks from 3rd parties, these are just ads for Intel.

 

[LukeTbk]

Until there are benchmarks from 3rd parties, these are just ads for Intel.

At least there some I am not supposed to do any benchmark but wink Intel is not lying early talk occuring...

But that "launch"-openning presales 3 full week before an embargo lifting is really wild

 

[Marees]

“A big step forward for Intel …

Intel's turnaround depends on more than Lunar Lake. The company must get its Intel 18A manufacturing process up and running, and successfully sell it to external customers. It also must win back share in the server CPU market after falling behind AMD in performance and efficiency. While this will take time, Lunar Lake looks like a positive development as Intel plots its comeback.

Third-party reviews, which should be available around the Sept. 24 launch, will shed light on whether Intel's performance and battery life claims hold water. If Intel's claims hold up, Lunar Lake will be a potent force in the laptop market this holiday season.

The massive battery life gains provided by Lunar Lake will give those with older laptops a good reason to upgrade, and they won't have to accept the downsides of a Qualcomm-powered laptop. Intel's client computing segment was its strongest in the second quarter, growing by 9% year over year following a post-pandemic slump. If Lunar Lake triggers an upgrade cycle, a big rebound could be in order for 2025.”

Source: https://finance.yahoo.com/news/intels-lunar-lake-looks-home-112000963.html

I believe Intel's claims.

They are on a node that almost matches apple leaving AMD behind in the dust

& they are fabbing on a foundry that has a track record of execution right from the time it was founded.

Let us see. Hope Intel’s claims are right

 

[westrock2000]

Hopefully laptop manufacturers will keep the same size batteries to give longer run times and not decide that customers are used to 10 hours, so that means they can reduce the size of the battery and save a couple dollars.

 

[Sindalis]

I believe Intel's claims.

They are on a node that almost matches apple leaving AMD behind in the dust

& they are fabbing on a foundry that has a track record of execution right from the time it was founded.

Let us see. Hope Intel’s claims are right

Intel and AMD are prioritizing different things right now. We will see which one ends up being the better long term bet.

AMD's margins are going to be far healthier than Intel's given that they are using a slightly older node rather than the absolute best one at TSMC.

So while Intel may have a better overall chip, profit wise it may be a worse situation compared to AMD.

 

[Lakados]

Intel and AMD are prioritizing different things right now. We will see which one ends up being the better long term bet.

AMD's margins are going to be far healthier than Intel's given that they are using a slightly older node rather than the absolute best one at TSMC.

So while Intel may have a better overall chip, profit wise it may be a worse situation compared to AMD.

Intel needs a few wins right now, and if they need the newer better node to pull it off then so be it.

 

[LukeTbk]

absolute best one at TSMC.

Close too or at least just for the poor non-Apple peasant, I think N3E (M4) is better than N3B.

Keeping up with those nodes is not obvious to me, but TSMC N3 mass product launched exaclty a full 2 years ago by now, there probably already a bit of a family of product and price available.

Intel needs a few wins right now, and if they need the newer better node to pull it off then so be it.

To me it feel it would have been worth it for sure, to pay the humiliation to buy TSMC and to go with a really good node as well, another 15600-15900k family of cpu 2% faster using 4% more power than 14xxx family because 20A is not ready this year sound like a disaster.

And Intel were used to Intel foundry really high price, not sure how much we can even call it a choice at this point, and maybe worst/importantly for them the gap on Laptop with AMD was becoming dangerous, the moment they loose momentum-oem relationship-people reflex-abit to buy Intel based on the last 15 years dominance on Laptop, they could have been cook.

 

[DukenukemX]

Intel and AMD are prioritizing different things right now. We will see which one ends up being the better long term bet.

AMD's margins are going to be far healthier than Intel's given that they are using a slightly older node rather than the absolute best one at TSMC.

So while Intel may have a better overall chip, profit wise it may be a worse situation compared to AMD.

Intel has a different goal in mind over AMD. Lunar Lakes goal isn't to have good margins or to sell well, but to change the perception of x86 power efficiency. Apple and Qualcomm's selling point is that they're efficient compared to x86 and Intel knows they're making x86 look bad. Everything about Lunar Lake is designed around efficiency. TSMC's 3nm process which Apple held with an iron grip is now in use with Intel. Intel is using new cores with efficiency in mind. Intel put the ram closer to the SoC, just like Apple has. Intel is putting in more fixed hardware functions to reduce power usage. Once Lunar Lake fixes x86's poor power consumption reputation, we'll see successors using Intel's manufacturing process to cut costs and to further boost performance. AMD on the other hand isn't worried so much about x86's reputation. AMD needs money and that's why they stuck with 4nm. AMD knows they have an edge over Intel in terms of performance. I doubt Lunar Lake is faster than AMD's Strix Point. Lunar Lake's GPU will probably be faster, assuming that Intel got their drivers fully rectified. Which is probably why you'll see more hand held gaming devices use AMD. AMD laptops will also likely be cheaper.

 

[Lakados]

Intel has a different goal in mind over AMD. Lunar Lakes goal isn't to have good margins or to sell well, but to change the perception of x86 power efficiency. Apple and Qualcomm's selling point is that they're efficient compared to x86 and Intel knows they're making x86 look bad. Everything about Lunar Lake is designed around efficiency. TSMC's 3nm process which Apple held with an iron grip is now in use with Intel. Intel is using new cores with efficiency in mind. Intel put the ram closer to the SoC, just like Apple has. Intel is putting in more fixed hardware functions to reduce power usage. Once Lunar Lake fixes x86's poor power consumption reputation, we'll see successors using Intel's manufacturing process to cut costs and to further boost performance. AMD on the other hand isn't worried so much about x86's reputation. AMD needs money and that's why they stuck with 4nm. AMD knows they have an edge over Intel in terms of performance. I doubt Lunar Lake is faster than AMD's Strix Point. Lunar Lake's GPU will probably be faster, assuming that Intel got their drivers fully rectified. Which is probably why you'll see more hand held gaming devices use AMD. AMD laptops will also likely be cheaper.

It also serves to show off Foveros, here Intel is with a mass production consumer part on a process that they have working with the best silicon available.

 

[LukeTbk]

we'll see successors using Intel's manufacturing process to cut costs and to further boost performance.

Not sure if they will be able to be cheaper than buying TSMC (they have been so subsidized that it will maybe work), historically it tend to always be the other way around, they need a big client too and themselve is an easy one to get and if they ever need volume.. instead of fighting with other, they could have it.

 

[DukenukemX]

Not sure if they will be able to be cheaper than buying TSMC (they have been so subsidized that it will maybe work), historically it tend to always be the other way around, they need a big client too and themselve is an easy one to get and if they ever need volume.. instead of fighting with other, they could have it.

There's a reason why nobody else has access to TSMC's 3mn. I imagine that 4nm and 5nm are cheaper, but who knows by how much? Apple most likely paid a fortune for 3nm, and to have it locked out of their competitors. Why you think Apple made the M4 and stuck it into iPads and not a Macbook, besides that Apple just released M3's in November of 2023? Why does Apple have an iPad in 2024 that has an M2 in it, and released the Vision Pro with an M2? Apple must have so many of these chips that they don't know what to do with them, and it's because Apple is paying for the manufacturing time to make sure nobody else has access to it for a while. For Intel to have 3nm TSMC this soon suggests they paid up their nose for it, and they pushed Apple out in the process. Apple probably didn't have access to 3nm for months due to Intel. If Intel doesn't charge a premium for Lunar Lake, then their margins are super low. Intel's own manufacturing might not be cheap to build, but has long lasting cost cutting for them. It also means Intel doesn't have to fight Apple and other CPU manufacturers for the best chip manufacturing process, assuming Intel keeps up with TSMC.

 

[LukeTbk]

There's a reason why nobody else has access to TSMC's 3mn.

Not saying it is cheap, but it is easy to understimate how much Intel spend on foundry R&D and infrastructure versus the cost of buying even the best TSMC node instead.

Intel foundry have lost 4.7 billions in 2024 despite 8.5 billions in revenues, thats not necessarily cheaper (would it have worked). It is very easy for a company like Intel, to make chips at an higher price for themselves than the best of the world at making them (even if they have good margin), specially when you count delay cost and not working nodes cost that you need to amortize.

One reason is also volume (which is maybe not that different than cost), Nvidia not going on N3 for the datacenter line of Blackwell I doubt is for cost reason, they sales so much big chips.

Apple most likely paid a fortune for 3nm, and to have it locked out of their competitors. Why you think Apple made the M4 and stuck it into iPads and not a Macbook, besides that Apple just released M3's in November of 2023?

Considering the 2023 M3 is also on TSMC 3nm, just an older version of it, I am not sure of the implication here.... does Apple release a new M chips every year ? why is this a question why they made an M4. Why they released the M4 inside the iPAD first instead of an MacBook ? Not sure what that change in term of locking competition out of the node, but could it be because the Ipad were still on the m2 and not the macbook ?

For Intel to have 3nm TSMC this soon suggests they paid up their nose for it,

It is the old version of 3nm and we are now a full 2 years since TSMC 3N has been at full volume, it must have been quite expensive but I am not sure it is particularly soon, probably still expensive considering they did not went full N3, just the compute tile.

and released the Vision Pro with an M2?

In june of 2023, m3 was maybe not ready yet (m3 product launch october-november)..

Apple probably didn't have access to 3nm for months due to Intel.

Apple is now on N3E, Intel is on the older N3B.

 

[Lakados]

There's a reason why nobody else has access to TSMC's 3mn. I imagine that 4nm and 5nm are cheaper, but who knows by how much? Apple most likely paid a fortune for 3nm, and to have it locked out of their competitors. Why you think Apple made the M4 and stuck it into iPads and not a Macbook, besides that Apple just released M3's in November of 2023? Why does Apple have an iPad in 204 that has an M2 in it, and released the Vision Pro with an M2? Apple must have so many of these chips that they don't know what to do with them, and it's because Apple is paying for the manufacturing time to make sure nobody else has access to it for a while. For Intel to have 3nm TSMC this soon suggests they paid up their nose for it, and they pushed Apple out in the process. Apple probably didn't have access to 3nm for months due to Intel. If Intel doesn't charge a premium for Lunar Lake, then their margins are super low. Intel's own manufacturing might not be cheap to build, but has long lasting cost cutting for them. It also means Intel doesn't have to fight Apple and other CPU manufacturers for the best chip manufacturing process, assuming Intel keeps up with TSMC.

Intel is using 3B not 3N big difference. When Apple was using 3N they were getting around a 40% yield from it. It was so bad TSMC wasn’t charging per wafer but per viable chip.

3B is something like a 90% yield rate and has far more in common with 4N than 3N.

TSMC has far more facilities capable of 3B than they did 3N.

 

[DukenukemX]

Not saying it is cheap, but it is easy to understimate how much Intel spend on foundry R&D and infrastructure versus the cost of buying even the best TSMC node instead.

Intel foundry have lost 4.7 billions in 2024 despite 8.5 billions in revenues, thats not necessarily cheaper (would it have worked). It is very easy for a company like Intel, to make chips at an higher price for themselves than the best of the world at making them (even if they have good margin), specially when you count delay cost and not working nodes cost that you need to amortize.

It's short term vs long term goals. If you think of it this year with Lunar Lake then it's not saving Intel money. Years from now probably will.

One reason is also volume (which is maybe not that different than cost), Nvidia not going on N3 for the datacenter line of Blackwell I doubt is for cost reason, they sales so much big chips.

Why then is Nvidia not going on N3 if not for cost?

Considering the 2023 M3 is also on TSMC 3nm, just an older version of it, I am not sure of the implication here....

Apple probably lost time on TSMC's 3nm from Intel, so they either made M4's or made more M3's. Apple had the M4 chip back in May, which is crazy to think that Macbooks are still selling with M3's.

does Apple release a new M chips every year ?

Seems like it.

why is this a question why they made an M4. Why they released the M4 inside the iPAD first instead of an MacBook ? Not sure what that change in term of locking competition out of the node, but could it be because the Ipad were still on the m2 and not the macbook ?

Apple's Macbooks aren't selling well and that's a problem for the amount of M-series chips they made. They made the M4's when they could before Intel bought it up, and then stuck them into iPads in hope to gain more sales. During a time when tablets sales aren't hot and the idea of the tablet replacing laptops has never worked. Apple is desperate to sell M-series based products but they also want to keep exclusivity to TSMC's best manufacturing process.

It is the old version of 3nm and we are now a full 2 years since TSMC 3N has been at full volume, it must have been quite expensive but I am not sure it is particularly soon, probably still expensive considering they did not went full N3, just the compute tile.

Lunar Lake is entirely built on TSMC's 3nm, with no real use of their tile technology. The only tile they have that isn't TSMC's 3N, is just a filler tile that does nothing. I'm sure TSMC has improved 3N, but it isn't a huge leap in terms of performance results.

In june of 2023, m3 was maybe not ready yet (m3 product launch october-november)..

This means that Apple's internal communication is terrible.

Apple is now on N3E, Intel is on the older N3B.

What difference this makes for Intel's chips?

 

[LukeTbk]

It's short term vs long term goals. If you think of it this year with Lunar Lake then it's not saving Intel money. Years from now probably will.

It could be the goal, but in the last decade was it ever significantly cheaper for Intel ?

Why then is Nvidia not going on N3 if not for cost?

It could be by how much that Nvidia 4NP node has been optimized for their need, 208 billions transistors on the Blackwell version for the GB200 according to Nvidia that could be a ~30% increase in density from the H100 generation. And just capacity, if the 200 billions in data center revenues happen in 2025, that a lot of 800-850mm die to make.. Maybe the yield is good enough to have volume for 160mm or less A17/M4/Lunar Lake compute tile, but goes down too low to chips millions of 800-850mm die.

What difference this makes for Intel's chips?

Not sure why it make any difference for Intel, I am not sure about your story you are telling about the: They made the M4's when they could before Intel bought it up,, storyline, the idea Apple is not able to outbid intel and get priority, etc... but Intel is not using the same node than the M4 is being used, thus that comment, I am not sure they conflict that much. And one minute it seem, no m4 in the Macbook because they have too many M3 on their hand, the second is because Intel took fab time away from them.

This means that Apple's internal communication is terrible.

It is (for them) a very low seller product, almost devkit level, that it does not get perfect timing is to be expected.

Lunar Lake is entirely built on TSMC's 3nm, with no real use of their tile technology.

Controller-io is on TSMC 6nm (platform controller tile):

I imagine quite similar to AMD using tsmc 6nm for the io die on the Ryzen cpus and tsmc 4-5nm on the logic heavy die.

 

[emphy]

Not sure if they will be able to be cheaper than buying TSMC (they have been so subsidized that it will maybe work), historically it tend to always be the other way around, they need a big client too and themselve is an easy one to get and if they ever need volume.. instead of fighting with other, they could have it.

Not sure whether intel necessarily plans on being cheaper; they purchased the entire 2024 production of asml's most advanced lithography machines. That is more an indication that they primarily plan on grabbing and holding the highest end production than of them trying to save costs.

It'll be interesting to see whether intel can manage to implement this successfully for a change.

 

[Lakados]

It could be the goal, but in the last decade was it ever significantly cheaper for Intel ?


It could be by how much that Nvidia 4NP node has been optimized for their need, 208 billions transistors on the Blackwell version for the GB200 according to Nvidia that could be a ~30% increase in density from the H100 generation. And just capacity, if the 200 billions in data center revenues happen in 2025, that a lot of 800-850mm die to make.. Maybe the yield is good enough to have volume for 160mm or less A17/M4/Lunar Lake compute tile, but goes down too low to chips millions of 800-850mm die.


Not sure why it make any difference for Intel, I am not sure about your story you are telling about the: They made the M4's when they could before Intel bought it up,, storyline, the idea Apple is not able to outbid intel and get priority, etc... but Intel is not using the same node than the M4 is being used, thus that comment, I am not sure they conflict that much. And one minute it seem, no m4 in the Macbook because they have too many M3 on their hand, the second is because Intel took fab time away from them.


It is (for them) a very low seller product, almost devkit level, that it does not get perfect timing is to be expected.


Controller-io is on TSMC 6nm (platform controller tile):
View attachment 678273

I imagine quite similar to AMD using tsmc 6nm for the io die on the Ryzen cpus and tsmc 4-5nm on the logic heavy die.

The biggest issue is SRAM doesn’t scale past 6nm in any manner that you want to pay for.
Relative to the rest of the chip it’s scales so poorly that it looks like an explosive growth.

IO relies on buffers and buffers require SRAM so until someone discovers a better structure, then 6nm will be king for cache, IO, and other such components for a while.

https://semiengineering.com/sram-scaling-issues-and-what-comes-next/