Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads

[Tigerick]

As Hot Chips 34 starting this week, Intel will unveil technical information of upcoming Meteor Lake (MTL) and Arrow Lake (ARL), new generation platform after Raptor Lake. Both MTL and ARL represent new direction which Intel will move to multiple chiplets and combine as one SoC platform.

MTL also represents new compute tile that based on Intel 4 process which is based on EUV lithography, a first from Intel. Intel expects to ship MTL mobile SoC in 2023.

ARL will come after MTL so Intel should be shipping it in 2024, that is what Intel roadmap is telling us. ARL compute tile will be manufactured by Intel 20A process, a first from Intel to use GAA transistors called RibbonFET.





Intel Core Ultra 100 - Meteor Lake



As mentioned by Tomshardware, TSMC will manufacture the I/O, SoC, and GPU tiles. That means Intel will manufacture only the CPU and Foveros tiles. (Notably, Intel calls the I/O tile an 'I/O Expander,' hence the IOE moniker.)

 

[adroc_thurston]

Nvidia wouldn't use it

Yea they would. For cheaper parts.
There are baby cost down parts in every segment.

I think Nvidia would gladly pay a premium on the node if it meant they hold onto their lead.

Well it's less that and more "not getting rolled over by AMD". They saw what happened to the poor poor Xeon and took notes.

 

[Saylick]

Well it's less that and more "not getting rolled over by AMD". They saw what happened to the poor poor Xeon and took notes.

Again, in a segment where performance drives profit, choosing to skimp on the node is penny wise, pound foolish. If you don't use the best node to give your product the best performance it can get, your competitors will and then your product won't sell which makes the penny pinching moot.

 

[adroc_thurston]

Again, in a segment where performance drives profit, choosing to skimp on the node is penny wise, pound foolish

Costs matter.
Ffs, Blackwell isn't even N3 anything lmao.

 

[Saylick]

Costs matter.
Ffs, Blackwell isn't even N3 anything lmao.

N3E wasn't even ready in time for the volumes Nvidia needs for Blackwell. Nvidia knew it and more importantly they knew that AMD knew it (hence the bulk of Zen 5 is on N4P), so they weren't threatened.

 

[adroc_thurston]

N3E wasn't even ready in time for the volumes Nvidia needs for Blackwell.

But N3b was, lmao.
Again, costs. Matter.

 

[SiliconFly]

.

 

[igor_kavinski]

If 18A doesn't deliver on schedule, Intel is finished.

What? They still have Intel 7 and 4. They can make Skymonts forever!

 

[SiliconFly]

.

 

[adroc_thurston]

Nope. They need a tech wiz at the helm like Lisa, Jensen, Tim, Sathya, Sundar, Mark, etc. Not another paper pusher like Brian Krzanich or Bob Swan who are the real culprits who ran down Intel.

BK was a fab engineer/later manager lol.

 

[SiliconFly]

.

 

[branch_suggestion]

I'm sure 18A is far better than TSMC A16.

Ah yes, matching the SRAM density of the node TSMC put into risk production in... 2019.
Curb your enthusiasm.

 

[adroc_thurston]

Ah yes, matching the SRAM density of the node TSMC put into risk production in... 2019.

well, SRAM is the fundamental physical limit stuff.
now, xtor perf?

 

[cannedlake240]

Ah yes, matching the SRAM density of the node TSMC put into risk production in... 2019.
Curb your enthusiasm.

TSMC this year matched the SRAM density of the 2019 TSMC node too... There's a rumor that actual N2 SRAM cell scaling isn't all that great and TSMC is moving goalposts to claim a larger than practically achievable increase. Just a rumor though we'll see soon enough at IEDM 24

 

[SiliconFly]

.

 

[MoistOintment]

Okay, but is a leading-edge foundry producing consumer or even enterprise products necessary to fulfill that purpose?

Leading edge is one of the primary components of the 3rd Offset strategy. The DoD doesn't need leading edge for missiles. They need leading edge for AI and data fusion. The intent is to saturate the battlespace with sensors and then use "fuse" the various sensor data sources to create a cohesive network battlefield.

That, and also using leading edge for AI to train autonomous weapons systems. They're already using all of that Ukrainian war drone footage for training. Also look at Israel's Lavender system as well.

The 5th Domain is a really good book on this topic. Cyber-War is a key element of future military planning and being able to "out-compute" your opponent will matter.

The only reason military tech (when we think of traditional, kinetic military weapons) isn't using leading edge today is because of the lag time of building out and mass producing these weapons, and then freezing the design for a period of time. They were all leading edge at some point.

 

[branch_suggestion]

Slightly higher density doesn’t help if client/server products on A16 are expected to run hotter. Curb your A16 enthusiasm.

I really doubt there will be much difference in thermals between PowerVia and BSPDN.
Besides, I'm only taking about N2 vs 18A for now.

 

[maddie]

Leading edge is one of the primary components of the 3rd Offset strategy. The DoD doesn't need leading edge for missiles. They need leading edge for AI and data fusion. The intent is to saturate the battlespace with sensors and then use "fuse" the various sensor data sources to create a cohesive network battlefield.

That, and also using leading edge for AI to train autonomous weapons systems. They're already using all of that Ukrainian war drone footage for training. Also look at Israel's Lavender system as well.

The 5th Domain is a really good book on this topic. Cyber-War is a key element of future military planning and being able to "out-compute" your opponent will matter.

The only reason military tech (when we think of traditional, kinetic military weapons) isn't using leading edge today is because of the lag time of building out and mass producing these weapons, and then freezing the design for a period of time. They were all leading edge at some point.

I'm waiting for the hallucinating AI adviser to replicate Bagdad Bob's performance in Gulf war 1.

 

[SiliconFly]

.

 

[DavidC1]

This is still all board's fault. Pat should have got in earlier, instead of Kraznich. Instead, when the Moore's Law gains are really, really slowing, they started investing.

One of Intel’s most successful CEOs was a finance guy. They just need someone who can lead and direct technical progress, regardless of previous education and role.

Who are you talking about? Otellini? He did ok on the small things but he failed catastrophically on the big thing, which is the potential Apple deal.

When it mattered, he failed. The implications of that are still felt today. If it got into at least first few designs, x86 designs would have had a chance, because compatibility would have been better, and Intel would have been forced to focus on low power much faster, much earlier.

Otellini is also the one responsible for firing the then-CTO Gelsinger by the way.

 

[ajsdkflsdjfio]

This is still all board's fault. Pat should have got in earlier, instead of Kraznich. Instead, when the Moore's Law gains are really, really slowing, they started investing.

Who are you talking about? Otellini? He did ok on the small things but he failed catastrophically on the big thing, which is the potential Apple deal.

When it mattered, he failed. The implications of that are still felt today. If it got into at least first few designs, x86 designs would have had a chance, because compatibility would have been better, and Intel would have been forced to focus on low power much faster, much earlier.

Otellini is also the one responsible for firing the then-CTO Gelsinger by the way.

It's really sad since most people here (and imop intel upper management) fail to understand it takes years or even a decade to build a successful foundry from how bad it was pre-Gelsinger. Also they fail to understand how half or more of client/DC products released under Gelsinger were planned out and defined BEFORE Gelsinger's tenure. SRF for example released in 2024 one of the first products planned by Jim Keller... who left in 2020. Keller's eventually cancelled royal core project wasn't going to be seen in consumer products until at least 2026.

The fact that Gelsingers 5N4Y plan even managed to barely reach it's milestones is a testament to how much Gelsinger bootstrapped the foundry side of things. What I don't get is why he was fired just when things were starting to turn around with 18a around the corner widely rumored to be at parity with N3 and intel finally moving to stop outsourcing for their products and return to internal production. Hopefully the new management sees the potential in foundry and doesn't cripple it. TSMC prices are insane, demand is high, and the old 2nd place Samsung foundry is pulling a 2010s Intel right now. If focusing on foundry was ever a good idea, right now would be the time.

 

[desrever]

It's really sad since most people here fail to understand it takes years or even a decade to build a successful foundry from how bad it was pre-Gelsinger. Also they fail to understand how half or more of client/DC products released under Gelsinger were planned out and defined BEFORE Gelsinger's tenure.

People understand that just fine, you just don't understand that Pat sucks even if you exclude all that and all hes been doing is lying to himself and investors for 3 years and it's finally catching up to him.

 

[ajsdkflsdjfio]

People understand that just fine, you just don't understand that Pat sucks even if you exclude all that and all hes been doing is lying to himself and investors for 3 years and it's finally catching up to him.

How does he suck excluding all that? Excluding all that his, 5n4y plan is just coming into fruition despite it's extremely ambitious goals of going from -2 node disadvantage to node parity with 18a vs N3/N2. As for appeasing investors, sorry that investors aren't happy in the short term when significant money is spent to rebuild intel from the ground up and when the actions of previous Ceos are finally hitting during the current year. It isn't magic that Intel missed the EUV train and lost a significant process advantage to TSMC over the span of a decade to the point where they were behind by multiple process nodes.

Also why does everybody blame missing the AI train on Pat. He was too optimistic on reporting Gaudi success but at the same time what do you expect from the CEO? No CEO is completely honest about how successful the company's products are doing. The fact is Intel had missed the AI/GPU train a long time ago. Pat is not responsible for missing out on AI and on the contrary it could be argued that he has build a solid foundation for intel to finally start to grow again. How do you expect intel to compete with no foundry business, no GPU experience, lackluster design. At least pat sought to fix some of those things in the long term even if it doesn't pay off now.

 

[maddie]

People understand that just fine, you just don't understand that Pat sucks even if you exclude all that and all hes been doing is lying to himself and investors for 3 years and it's finally catching up to him.

AND, obviously, the board will know things we don't. Assuming things are good for the future does not reconcile with shaking the company so severely. The safest assumption, is that things are bad with the present path AND is not expected to recover anytime soon. Pie in the sky projections seems to have met their match with reality, which always win.

Anyone has a read on employee reactions. This will be telling.

Quite ironic as I was accused of spreading FUD with Intel 2 days ago, now this.

 

[DavidC1]

He wasn't a perfect CEO. He definitely missed the boat on AI. He could have accelerated it, but didn't see it coming. He also missed forecasting that the revenues will plummet after the lockdowns were over, and interest expenditures skyrocketed. Real estate foreclosures increased a ton.

Pushing to a new process I still think was a right choice. He was too optimistic in the regard, and the road to the destination having numerous troubles, but overall positive.

Every CEO has their own way of doing things. Having to go through a period of changing CEOs again is going to have further negative impact. Intel was criticized for having bad BoD for a while.

This is very negative for Intel long term. They aren't ready for splitting of the company, but that may be the future facing them. Neither will likely survive, or be much of left over when the dust settles.

 

[H433x0n]

If 18A doesn't deliver on schedule, Intel is finished.

It doesn’t matter, it’s a Pyrrhic victory.

Based on Intel 3 performance the odds are good that 18A is at parity with N3P, but if the supporting ecosystem isn’t competitive with what TSMC offers what is the point? Rationally speaking, why take such a gamble? If I’m running QCOM, AVGO or NVDA why would I gamble with 18A? The only solid rationale would be that 18A has a clear competitive advantage over N3P and even in that scenario I would hedge my bets.

It’s past the point where IFS can bail out Intel, money in 2027 and 2028 isn’t going to keep the lights on right now.