Why Lunar Lake changes (almost) everything
- Thread starter TICin852
- Start date
That looks outstanding. Yay Intel!
MKWVentures
Moderator
There is no real previous gen of this chip. This is a new architecture..... it may end up being lower power, less expensive, higher performance than Intels other chips.
tooLongInEDA
Moderator
If you look closely at the slides in the video, you'll notice that some of the Intel performance slides are marked as "projected or estimated" with "an accuracy of +/- 10%". This seems a little loose given that this is - I assume - working silicon. Curious also that Intel obfuscated the die images they shared even though this will all come out soon enough - why bother ?
Interesting that the only Intel silicon in this device is the Intel 22 interposer base die - i.e. effectively the one without any transistors (or almost none).
Also note that the presenter - quite rightly - states at the start that this was from his all expenses paid trip to Computex paid by Intel.
Interesting that the only Intel silicon in this device is the Intel 22 interposer base die - i.e. effectively the one without any transistors (or almost none).
Also note that the presenter - quite rightly - states at the start that this was from his all expenses paid trip to Computex paid by Intel.
It's pretty much mostly a TSMC-made chip as well: https://www.tomshardware.com/pc-com...-ipc-gain-for-e-cores-16-ipc-gain-for-p-cores
Surprisingly, intel turned to TSMC for its leading-edge 3nm N3B process node for its compute tile, which houses the CPU, GPU and NPU. It also uses the TSMC N6 node for the platform controller tile that houses the external I/O interfaces. In fact, the only Intel-fabbed silicon on the chip is the passive 22FFL Foveros base tile that facilitates communication between the tiles and the host system.
Intel says it chose TSMC’s nodes because they were the best available when the company began designing the chip, a nod to its delays on the manufacturing side of the operation as it looks to regain its lead in foundry technology through its five nodes in four years initiative. However, Intel designed the architectures to be easily portable to other process nodes, so we can expect it to return to using its own nodes with many of these same architectures in its future products.
Surprisingly, intel turned to TSMC for its leading-edge 3nm N3B process node for its compute tile, which houses the CPU, GPU and NPU. It also uses the TSMC N6 node for the platform controller tile that houses the external I/O interfaces. In fact, the only Intel-fabbed silicon on the chip is the passive 22FFL Foveros base tile that facilitates communication between the tiles and the host system.
Intel says it chose TSMC’s nodes because they were the best available when the company began designing the chip, a nod to its delays on the manufacturing side of the operation as it looks to regain its lead in foundry technology through its five nodes in four years initiative. However, Intel designed the architectures to be easily portable to other process nodes, so we can expect it to return to using its own nodes with many of these same architectures in its future products.
MKWVentures
Moderator
In the Q&A Pat Said:
"“Put simply, Lunar Lake picked TSMC as the right technology at that point in time,” Gelsinger said at a post-keynote question-and-answer session at the Computex show in Taipei. “That’s why we ended up using more of it. And obviously, with the results I talked about today, it was a good choice.”
Lunar Lake looks to be positioned as the major notebook architecture.
tooLongInEDA
Moderator
This again raises the question: what if they [Intel Foundry] don't (or can't) ?
We're repeatedly told that Intel will switch back from TSMC silicon at the earliest opportunity (their massive internal fab investments make no sense if this isn't true). But the design teams have only just completed their first major complete CPU designs on TSMC. If these are successful both for the design/product groups (execution was easier) and commercially, surely there's a lot of inertia to continue down that route.
hist78
Well-known member
Without external foundry customers, Intel can't get better scale of economy in terms of volume. It's bad but not a life threatening situation. On the other hand without competitive products, made by internal or external fabs, Intel will be doomed.
Intel makes and will make majority of revenue and profit from its products, not Intel Foundry service. If Intel can't recognize this reality, it will be in big trouble.
Intel products being on TSMC are the result of wafer agreements intel says they signed in 2019. So if intel says that they will go from whatever % internal they get to this year to more historic 20% outsourcing in the 2027 timeframe with the goal of like 90% by 2030, that should be a very high confidence item. The reason I would ve confident saying that is twofold. One intel would need to have had wafer agreements in place for 2027 back in 2022, and 2030 in 2025. Two as you mentioned there is simply no way that intel is dumb enough to build the capacity to make chips they know they already know they aren’t making internally.
Intel is going fab lite (and shell heavy):
Ocotillo sold, Leixlip sold
The characteristics of private equity deals are well known. They provide a lot of fees to the private equity firm, marginally sustainable cash flow burdens to the cash flow source, and sooner or later, bankruptcy for the cash flow source.
I'm just reporting the news here folks. I don't like it that Intel's future is mortgaged to private equity, or that private equity has a history of happily driving firm after firm, most recently Red Lobster, to bankruptcy.
Ocotillo sold, Leixlip sold
The characteristics of private equity deals are well known. They provide a lot of fees to the private equity firm, marginally sustainable cash flow burdens to the cash flow source, and sooner or later, bankruptcy for the cash flow source.
I'm just reporting the news here folks. I don't like it that Intel's future is mortgaged to private equity, or that private equity has a history of happily driving firm after firm, most recently Red Lobster, to bankruptcy.
Very interesting as Fab 34 is an "Intel 4" fab. An advanced process, but also I haven't heard much about Intel 4 being a Foundry offering.
I agree the mortgaging assets is risky; I wonder what the business case was for this other than "raise capital". There has to be something useful for the private equity firm..
intel 3 is just an enhancement of intel 4. Think N7 (Apple only) vs N7P (CIP for Apple+ROW), or 10nm icelake vs 10nm SF or intel 7.
I don't really understand why folks have a hard time wrapping their head around the idea? Is it ideal? Of course not, but intel wasn't forced at gunpoint to do this or AZ. If they are doing it, finance did the math and came to the conclusion that this was better than taking out loans at the interest rates they would have been at or selling a larger stake of the company. I would have to assume they know finance and intel's books/plans better than I do, so I have no reason to doubt this being the correct choice.
I get the reason why - it just makes me a little nervous about the long term health of Intel.
I also rarely see these “asset management firms” actually add any value to businesses.
100% personal biases, I acknowledge
.
Lunar Lake has three tiles: the Compute Tile, Platform Controller Tile and Base Tile. The Base Tile is the only one made on an Intel process. The other two, which are absolutely the most important, are made by TSMC. The Compute Tile, which includes all of the cores, GPU and NPU, is made on TSMC's N3B node.
"Simply put, Lunar Lake picked TSMC as a better process technology at that point in time," Intel CEO Pat Gelsinger says at Computex 2024. "And so that's why we ended up using more of it."
"And next year, when we move to Panther Lake, almost all of the tiles are on Intel."
Almost all? So you can say Lunar Lake has almost none?
"Simply put, Lunar Lake picked TSMC as a better process technology at that point in time," Intel CEO Pat Gelsinger says at Computex 2024. "And so that's why we ended up using more of it."
"And next year, when we move to Panther Lake, almost all of the tiles are on Intel."
Almost all? So you can say Lunar Lake has almost none?
MKWVentures
Moderator
lunar Base tile is passive, correct? It is definitely passive (all wires) on Meteor lake
hist78
Well-known member
I hope Pat Gelsinger is humble enough and smart enough to keep certain amount of Intel chips production on TSMC's leading edge nodes. Although it appears Intel is going all inhouse for new leading edge node products after Lunar Lake. Intel was lucky (really!!) to make a right decision several years ago to use TSMC for Lunar Lake's manufacturing. If not, Intel would have had no meaningful product to compete against AMD or Qualcomm in 2024 on the laptop market.
Pat Gelsinger has repeatedly stated that IDM 2.0 and Intel Foundry model will allow Intel product/design division to choose between Intel internal foundry and external foundries freely. Why would Intel give up such flexibility and competitive advantages to go for all inhouse? It's totally an unnecessary gamble that can kill Intel.
Last edited:
Sierra Forest, a different Intel product line (server-based), uses Intel-only (3,7) chiplets. https://www.tomshardware.com/news/i...-and-granite-rapids-architecture-xeon-roadmap
Maybe that's because the server market is not very price-sensitive but the laptop market is, Intel Foundry wafer/chip costs are higher than TSMC, so they use TSMC where cost matters and Intel where it doesn't?