News Intel 3Q23 Earnings

[Saylick]

I wouldn't even be surprised if at some point in the future Apple becomes an IFS customer, assuming IFS gets their act together node-wise. With Tim Cook, a supply chain guru, at the helm of Apple, they'd desire a second supplier for their chips so that they can de-risk sole sourcing from TSMC.

 

[H433x0n]

QC and nVidia are like the least likely to bother with 18A. It's probally the usual suspects like Erickson or perhaps some DoD stuff.

I think most people are looking at this incorrectly and treating it as if it's like partisan politics where you have to pick a side. We already know that both Qualcomm and Nvidia have tested the waters with IFS from previous press material. I could absolutely see both companies performing tape-outs at both IFS & TSMC. These companies were exclusive to TSMC at times in the past mainly because TSMC had a monopoly on leading edge process tech and there weren't any other competitive options. There's no reason to assume they wouldn't use both simultaneously for a competitive price advantage and to ensure they've always got products on the best process tech.

When will they reveal their foundry customers?

The customer doesn't benefit by allowing Intel to use you for publicity - so we likely won't hear about the customers until things have progressed to the point where it can't really be hidden.

 

[jpiniero]

Why wouldn't Nvidia want it? They can sell all the AI stuff they can make so a second supplier - especially if it is was the best performing node at the time it starts mass production - would be something I'd think they'd want if Intel is able to pull off the claimed schedule.

So far, NV is doing retcle sized monolithic dies which is... not exactly compatible with 18A. I forgot about the QC rumors but I don't think they were really serious and Intel missed the metrics anyway so they bailed.

The ideal customer would be AMD ironically enough.

 

[H433x0n]

So far, NV is doing retcle sized monolithic dies which is... not exactly compatible with 18A. I forgot about the QC rumors but I don't think they were really serious and Intel missed the metrics anyway so they bailed.

The ideal customer would be AMD ironically enough.

Why wouldn’t it be compatible? Have you seen Xeon tiles? All of these nodes will be used for Xeon dies at the reticle limit. Afaik GNR tile is ~600mm2 and that’s going to be on Intel 3.

 

[adroc_thurston]

So far, NV is doing retcle sized monolithic dies which is... not exactly compatible with 18A.

?
No.
Intel has a history of shipping dies every bit as chungus, see HSW-EP/EX or Tukwila (or the recent SPR-MCC, it's like 770mm^2).

The ideal customer would be AMD ironically enough.

never ever until Intel fully spins the fabs out

Afaik GNR tile is ~600mm2 and that’s going to be on Intel 3.

Yes Xeons have been chungus since like Tulsa (if not earlier).

 

[jpiniero]

Why wouldn’t it be compatible? Have you seen Xeon tiles?

It helps that the tile is just Sea of Cores. Intel should be able to salvage most CPU tiles regardless of yield.

 

[H433x0n]

It helps that the tile is just Sea of Cores. Intel should be able to salvage most CPU tiles regardless of yield.

The same holds true with a GPU die, if anything it’s even easier to bin than a CPU tile.

 

[Doug S]

I wouldn't even be surprised if at some point in the future Apple becomes an IFS customer, assuming IFS gets their act together node-wise. With Tim Cook, a supply chain guru, at the helm of Apple, they'd desire a second supplier for their chips so that they can de-risk sole sourcing from TSMC.

Tim Cook is very conservative about Apple's supply chains and Apple would require a truly massive number of wafers. I am pretty sure Intel would not have nearly the capacity to take them on until the Ohio fabs are ready.

I could see them dual sourcing Apple Silicon perhaps as a test to see how working with Intel goes versus working with TSMC before committing their most important product to Intel foundries. Since Apple is practically a partner in node design and are responsible for 25% of TSMC's revenue they have of power in that relationship. They would not want to go to playing second fiddle behind what will always be Intel's most important customer unless they spin off the foundry business.

 

[mikk]

MTL took 1.5 years after booting up windows to launch

Which is why it makes sense that the first ARL chips, I mean ARL-S 8+16 tile, are on N3B instead of 20A. It's similar with Lunar Lake by the way. Lunar Lake on N3B booted earlier than ARL 20A by at least 2 months. For those who are still expecting an Intel node for Lunar lake compute chiplet, just ask why Intel completely ignores Lunar Lake when they are talking about future process nodes. They have mentioned 20A+ARL, 18A+PTL and other future chips with the exception of Lunar Lake. ARL is lead vehicle on 20A and just booted later than Lunar Lake and PTL will be the first 18A client chip. Intel already gave us the answer.

 

[jpiniero]

Intel said notebook volume was up 8% YoY but ASPs were down 5%. Desktop volume was down 19% but ASPs were up 6%.

Data Center is interesting... volume was down 35% but ASP was up 38%. That's wild. "higher mix of core count products"

 

[Abwx]

Data Center is interesting... volume was down 35% but ASP was up 38%. That's wild. "higher mix of core count products"

They were selling their DC chips at a loss the prvious quarters, currently they are sold at 0% net margin, that s how they kept AMD from gaining marketshare in this segment.

 

[jpiniero]

They were selling their DC chips at a loss the prvious quarters, currently they are sold at 0% net margin, that s how they kept AMD from gaining marketshare in this segment.

ASPs were up in Q2 too. They aren't giving the products away. It's more likely that Sapphire Rapids yields (for the products that customers are buying) are so terrible they are spending a crazy amount of wafers to satisfy the demand. Which they have the 10 nm capacity for since volume is so low.

Edit: I guess they could be offering deals on the high core count products, which causes the need to burn more wafers. But they are still getting higher ASPs out of what they are selling.

 

[Abwx]

ASPs were up in Q2 too. They aren't giving the products away. It's more likely that Sapphire Rapids yields (for the products that customers are buying) are so terrible they are spending a crazy amount of wafers to satisfy the demand. Which they have the 10 nm capacity for since volume is so low.

They were making big losses in DC, something like 500M in Q2, wich mean that the SKUs were sold at a loss, last quarter they were back to break even in this segment, wich mean that they sold them at 0% net margin.

If prices increased in Q2 it simply mean that they sold them at less loss during this quarter than during the previous ones.

 

[jpiniero]

They were making big losses in DC, something like 500M in Q2, wich mean that the SKUs were sold at a loss

It's not so simple with Intel... since (IIRC) they used to hide the R&D costs of future nodes within the Client and Data Center businesses. They may not be selling enough server product to overcome that.