This is for Gaudi 3 but yeah not really inspiring numbers during the gold rush.
I think that number was for all AI accelerator sales.
Well yeah FCS is a GPU. An actual SIMD machine.he distinguished between Gaudi and Falcon Shores that Falcon Shores will be fully programmable, addressing Intel's "deficit" in that area...
It does seem like for the last few years Intel has been hamstrung by poor decisions from former management, decisions in my opinion that were an overreaction from what the competition was doing or where the industry was heading. For example, this whole issue about MTL being volume limited by wafer assembly strikes me as being the same mistake as Ponte Vecchio: in a time where Intel was getting screwed because they were still making monolithic products while AMD was going chiplets, Intel decides to prove the naysayers wrong by going all in on chiplets and advanced packaging. They were going to out-do what AMD was doing by doing it bigger. But where did they end up? With a horribly late, non-competitive product like Ponte Vecchio. The same feels true about Sapphire Rapids, where they went too far to the right on the monolithic-chiplet spectrum and they pulled it back with Emerald Rapids. Similarly, look at how Lunar Lake actually simplifies the number of chips in comparison to Meteor Lake. On the foundry side, Intel is trying to beat TSMC by going all in on High NA EUV because of their earlier mistake of not committing to EUV. Again, trying to correct for their earlier mistake by overreacting and swinging the other way without much thought on if it really makes sense (consider that TSMC have said they don't need High NA EUV even for their newly announced N16 node).Pat said that the bottleneck for Meteor Lake production is "wafer assembly". It came a bit out of the left field.
I have speculated that the overcomplicated Meteor Lake design and partition will be an albatross around its neck (from cost POV) and now it is manifesting itself in other ways (beyond cost).
View attachment 97865
Pretty specifically Gaudi 3. but yea.
Well yeah FCS is a GPU. An actual SIMD machine.
But it's also late-late'25 aka basically 26.
bwah.
It does seem like for the last few years Intel has been hamstrung by poor decisions from former management, decisions in my opinion that were an overreaction from what the competition was doing or where the industry was heading. For example, this whole issue about MTL being volume limited by wafer assembly strikes me as being the same mistake as Ponte Vecchio: in a time where Intel was getting screwed because they were still making monolithic products while AMD was going chiplets, Intel decides to prove the naysayers wrong by going all in on chiplets and advanced packaging. They were going to out-do what AMD was doing by doing it bigger. But where did they end up? With a horribly late, non-competitive product like Ponte Vecchio. The same feels true about Sapphire Rapids, where they went too far to the right on the monolithic-chiplet spectrum and they pulled it back with Emerald Rapids. Similarly, look at how Lunar Lake actually simplifies the number of chips in comparison to Meteor Lake. On the foundry side, Intel is trying to beat TSMC by going all in on High NA EUV because of their earlier mistake of not committing to EUV. Again, trying to correct for their earlier mistake by overreacting and swinging the other way without much thought on if it really makes sense (consider that TSMC have said they don't need High NA EUV even for their newly announced N16 node).
Now that they are behind the 8-ball on the AI gold rush, Intel is going all in on NPUs and AI GPUs. Given their bad luck, I would not even be surprised if by the time they catch up the AI boom would be over.
I have little to add except that my true cause of worry wrt Intel is that nothing seems to panic over there.It does seem like for the last few years Intel has been hamstrung by poor decisions from former management, decisions in my opinion that were an overreaction from what the competition was doing or where the industry was heading. For example, this whole issue about MTL being volume limited by wafer assembly strikes me as being the same mistake as Ponte Vecchio: in a time where Intel was getting screwed because they were still making monolithic products while AMD was going chiplets, Intel decides to prove the naysayers wrong by going all in on chiplets and advanced packaging. They were going to out-do what AMD was doing by doing it bigger. But where did they end up? With a horribly late, non-competitive product like Ponte Vecchio. The same feels true about Sapphire Rapids, where they went too far to the right on the monolithic-chiplet spectrum and they pulled it back with Emerald Rapids. Similarly, look at how Lunar Lake actually simplifies the number of chips in comparison to Meteor Lake. On the foundry side, Intel is trying to beat TSMC by going all in on High NA EUV because of their earlier mistake of not committing to EUV. Again, trying to correct for their earlier mistake by overreacting and swinging the other way without much thought on if it really makes sense (consider that TSMC have said they don't need High NA EUV even for their newly announced N16 node).
Now that they are behind the 8-ball on the AI gold rush, Intel is going all in on NPUs and AI GPUs. Given their bad luck, I would not even be surprised if by the time they catch up the AI boom would be over.
I believe the term that explains their line of thinking is leapfrogging. I've talked before about their collective thinking on winning through sheer technology superiority, and leapfrogging is the concept that springs to mind when you realize the competition has pulled ahead. They cannot just emulate what the competition does, it must be bigger, better, faster all at once. This often fails, especially when the collective is overestimating their current capabilities.It does seem like for the last few years Intel has been hamstrung by poor decisions from former management, decisions in my opinion that were an overreaction from what the competition was doing or where the industry was heading.
That word works if you actually leapfrog.I believe the term that explains their line of thinking is leapfrogging.
Have they realised it? Cause to me it sems like they're mostly busy trying to act like nothing has changed since 2016 and they just need to pour more money until they win.I've talked before about their collective thinking on winning through sheer technology superiority, and leapfrogging is the concept that springs to mind when you realize the competition has pulled ahead.
Actually you can emulate just fine until you've cleared out your main problems. Doing it all at once just piles the problems on. Which is what we're seeing here.They cannot just emulate what the competition does, it must be bigger, better, faster all at once. This often fails, especially when the collective is overestimating their current capabilities.
Intel posted the 10-Q... Notebook volume was up 39% YoY and Desktop 29%. ASPs were flat... which is kind of interesting given Meteor Lake should be more expensive.
Server OTOH had a 13% drop in volume but 25% higher ASPs.
MTL was always going to be low volume so this isn't surprising. The server volume drop was expected given the current environment but that much increase in ASP is a nice surprise.
no AIbuxx.Seems the leadership wasn't able to calm people's response.
Volume is still pretty much Raptor.which is kind of interesting given Meteor Lake should be more expensive.
Volume is still pretty much Raptor.
Yeah, pricier products are PITA to ramp.Which is to be expected. Hell, it's going to be that way in 2025 too.
Yeah, pricier products are PITA to ramp.
On a normal day, the costs are not fully accounted for. Intel just pretends that some other department is making a loss selling them.
Process development should be financed by the margin of the foundry, no? That such a margin doesn't exist to fund future development suggests something bad but I am not sure.They are not. After all, the Foundry got 4.4 billion in "Revenue". You're just seeing what it costs to fund new nodes.
Process development should be financed by the margin of the foundry, no? That such a margin doesn't exist to fund future development suggests... something bad.
I doubt that there's a sucker with enough money and enough stupidity to pay for this.Yeah. Spinning off the fabs seems inevitable. Just gotta find a sucker.
I doubt that there's a sucker with enough money and enough stupidity to pay for this.
One with the stupidity but not the money or one with the money but not stupid enough are already hard enough.
Mmmmmh.Uncle Sam
Then you haven't being paying attention. As is said, there are many ways to skin a cat.Mmmmmh.
I think not.
Explaining why would require an answer in the Politics section.
To what?Then you haven't being paying attention.
Who are you?As is said
That's highly specific, thanks., there are many ways to skin a cat.
Would they need a buyer? Can't they just do some sort of share split, and leave Intel shareholders with a mix of shares in the two separate companies? (Proportional to the relative value of the two entities at the time of the split.)I doubt that there's a sucker with enough money and enough stupidity to pay for this.
One with the stupidity but not the money or one with the money but not stupid enough are already hard enough.
The UAE's sovereign fund is already one of the least dumb national investors out there, and all they could afford was GlobalFoundries, which isn't even going to attempt jumping into 7nm.
Someone with enough funds to buy IFS would require what, 10x the money GlobalFoundries was worth?