Discussion Intel current and future Lakes & Rapids thread

[shady28]

Aurora

I think what Pat was obliquely referring to with his positive comments on Sapphire Rapids was Aurora and Nvidia's H100. Those are pretty narrow use cases, as opposed to general purpose compute, so probably weren't affected by the defects they found.

 

[mikk]

Seems like TSMC denied the rumor as well.

https://twitter.com/x/status/1555401797166313472

Wccftech got another Intel statement

https://twitter.com/x/status/1555554802087743488

 

[shady28]

Seems like TSMC denied the rumor as well.

https://twitter.com/x/status/1555401797166313472

Wccftech got another Intel statement

https://twitter.com/x/status/1555554802087743488

I really don't think this was about Intel. There was plenty of news about other customers pulling back on orders and trying to re-negotiate. There's probably just not as much 3nm demand as TSMC anticipated, which shouldn't be real surprising given macro factors. PC sales crated, phone sales cratered, general electronics sales crated (see Best Buy), and there are reports of oversupply everywhere.

Canalys: Smartphone sales fell 9% in Q2 2022 due to oversupply and soaring inflation

Apple retakes its second place after steady iPhone 13 sales while Chinese companies struggled. Canalys posted a troubling report on the smartphone market...

www.gsmarena.com

I should probably put this out as a reminder too. The inflection point on cost per chip is at 7nm. The development cost for 3nm, before you even produce anything, is astronomical and very few can afford it :

 

[igor_kavinski]

Yikes! Half a billion+ for 5nm. And if it gets worse, say bye-bye to affordable chips.

 

[Exist50]

It was one of the many "Good Ideas" from their last CEO...

It was BK that laid off the server validation team, not Swan.

I've been thinking that Since SPR have been on an Eternal ES Cycle for a few years now. Is Intel actually closer to releasing Emerald Rapids than Sapphire Rapids? Please don't tell it's the same Team working on both

Emerald Rapids is basically a Sapphire Rapids refresh. It's probably delayed almost 1:1 with SPR, but even if not, it's not going to budge the needle regarding the performance gap.

 

[shady28]

Yikes! Half a billion+ for 5nm. And if it gets worse, say bye-bye to affordable chips.

Yeah, this is why sometimes I mention 'bifurcation' in the market - between the haves (latest node) and have-nots (older nodes) due to cost of designing and then making these sub 5nm chips.

We are already seeing it with Apple where next year they are expected to have the normal iPhone on 5nm still while the Pro and Pro Max might be on 3nm.

 

[Exist50]

Yeah, this is why sometimes I mention 'bifurcation' in the market - between the haves (latest node) and have-nots (older nodes) due to cost of designing and then making these sub 5nm chips.

We are already seeing it with Apple where next year they are expected to have the normal iPhone on 5nm still while the Pro and Pro Max might be on 3nm.

Eh, I'm not sold on some of these numbers. Sure, maybe costs start that high, but they come down over time with cost reductions in the nodes and cheaper IP.

Also, the Apple situation is quite different. They're paying to design and tape out a 3nm SoC regardless. Just for one reason or another, they're not using it across the lineup.

 

[nicalandia]

Emerald Rapids is basically a Sapphire Rapids refresh. It's probably delayed almost 1:1 with SPR, but even if not, it's not going to budge the needle regarding the performance gap.

Emerald Rapids is scheduled to be released in the second half of 2023, if they are 1:1 with SPR, this may well end up being second half 2024

 

[Exist50]

Emerald Rapids is scheduled to be released in the second half of 2023, if they are 1:1 with SPR, this may well end up being second half 2024

Well if we take Igor's dates, +1 year would still be H2'23, and it wouldn't be unreasonable to shave a bit of time off for such a simple iteration. If it takes them till H2 2024, with the accompanying implication that GNR is also delayed, well then just write off Intel's future in servers entirety.

 

[nicalandia]

Well if we take Igor's dates, +1 year would still be H2'23, and it wouldn't be unreasonable to shave a bit of time off for such a simple iteration. If it takes them till H2 2024, with the accompanying implication that GNR is also delayed, well then just write off Intel's future in servers entirety.

When can we expect to see ES Samples of Emerald Rapids if they are going to be releasing them at the end of 2023?

 

[Hitman928]

Eh, I'm not sold on some of these numbers. Sure, maybe costs start that high, but they come down over time with cost reductions in the nodes and cheaper IP.

Also, the Apple situation is quite different. They're paying to design and tape out a 3nm SoC regardless. Just for one reason or another, they're not using it across the lineup.

I'm not going to comment on the absolute number estimate, but the relative growth seems to match what I know about design costs on these processes. If you notice, the 'software' and 'verification' categories are the two highest contributors, not the cost of the physical masks/wafers. This is because the tool vendors force you to pay more money to support lower geometry nodes with their tool sets (usually because it requires additional functionality at lower nodes) and it's not just a little more. The lower nodes you go, the number and complexity of rules also increases and sometimes model complexity increases, which increases the time it takes for simulation and verification. You may also have to start bringing in all new tools into your flow for certain checks because of how hard the physics of everything are being pushed and the complexity of all the new packaging techniques.

It's not just the fab build outs themselves that are increasing in costs, but the cost to design on those processes is increasing significantly as well. I know of several design houses that are well funded (not the big boy Samsung/AMD/Intel type but still pretty well funded) that are trying to stay on older nodes as long as they possibly can because the design cost to jump to the n+1 (for them) node is brutal and so they are trying to make the nodes they are on now work for them as long as possible.

 

[Exist50]

When can we expect to see ES Samples of Emerald Rapids if they are going to be releasing them at the end of 2023?

I'd assume we should see ES chips popping up around early next year, but that's a guess on my part.

 

[tomatosummit]

I'm not going to comment on the absolute number estimate, but the relative growth seems to match what I know about design costs on these processes. If you notice, the 'software' and 'verification' categories are the two highest contributors, not the cost of the physical masks/wafers. This is because the tool vendors force you to pay more money to support lower geometry nodes with their tool sets (usually because it requires additional functionality at lower nodes) and it's not just a little more. The lower nodes you go, the number and complexity of rules also increases and sometimes model complexity increases, which increases the time it takes for simulation and verification. You may also have to start bringing in all new tools into your flow for certain checks because of how hard the physics of everything are being pushed and the complexity of all the new packaging techniques.

It's not just the fab build outs themselves that are increasing in costs, but the cost to design on those processes is increasing significantly as well. I know of several design houses that are well funded (not the big boy Samsung/AMD/Intel type but still pretty well funded) that are trying to stay on older nodes as long as they possibly can because the design cost to jump to the n+1 (for them) node is brutal and so they are trying to make the nodes they are on now work for them as long as possible.

"Software" is a pretty vague term here. While the licensing costs are probably eye watering does it include the man hours designing and egnineering your designs?
With time even those costs will drop, licences are a prime spot for negotiation, using mature third party IP blocks, reusing your own IP blocks, tool development to reduce hours of simulation or automated designing and even open specifications so companies can share costs.
A lot of the talk from chip companies moving forward is about how they're compartmentalising their design efforts so a soc can be made in a more mix and match style reducing costs. So a new product on a new node will certainly cost the earth but that can be amortised even early on across multiple products and then the node will mature over time, see N7 to N6, tsmc themselves say they want people to move to N6 and a good way to that is to offer better value to their customers so I would bet it is significantly cheaper than the day1 N7 ~$300M figure we have for a new design.

 

[Doug S]

Yikes! Half a billion+ for 5nm. And if it gets worse, say bye-bye to affordable chips.

I saw that graph elsewhere and a guy in the industry refuted that, saying he's got in depth knowledge of several 7nm SoC designs (he didn't say for what) that were all in at the $75 - $100 million range, nowhere near the claimed $300 million.

 

[pakotlar]

Not sure if this posted here as well, so here it is with a slight risk of redundancy:

https://twitter.com/x/status/1555297656183762944

Trendforce insights are 💩

 

[pakotlar]

It was one of the many "Good Ideas" from their last CEO... Hence he got Fired and I am sure now there is a Non-Rehirable Warning wiht Huge Red Caps on his file....

I've been thinking that Since SPR have been on an Eternal ES Cycle for a few years now. Is Intel actually closer to releasing Emerald Rapids than Sapphire Rapids? Please don't tell it's the same Team working on both, I always thought they had different team working on separate projects....

I have yet to se Emerald Rapids A0 or early ES sample in the wild.

Kzranich should be hired to break up monopolies, he’s really good at that.

 

[pakotlar]

It was one of the many "Good Ideas" from their last CEO... Hence he got Fired and I am sure now there is a Non-Rehirable Warning wiht Huge Red Caps on his file....

I've been thinking that Since SPR have been on an Eternal ES Cycle for a few years now. Is Intel actually closer to releasing Emerald Rapids than Sapphire Rapids? Please don't tell it's the same Team working on both, I always thought they had different team working on separate projects....

I have yet to se Emerald Rapids A0 or early ES sample in the wild.

Id be willing to bet 5 internet points that Emerald Rapids will be released same year as SPR, maybe as a 64 core high end offering in the same generation of processors.

 

[nicalandia]

Id be willing to bet 5 internet points that Emerald Rapids will be released same year as SPR, maybe as a 64 core high end offering in the same generation of processors.

So it will be short lived Waste of Sand like Rocket Lake and others in Intel past?

 

[jpiniero]

Trendforce insights are 💩

It is consistent with what Digitimes had reported earlier (ie: Meteor Lake has been pushed from a paper launch in Q2 to 2023 because of Intel 4, and they don't know what to do with the TSMC wafers they ordered and now don't need)

 

[Exist50]

It is consistent with what Digitimes had reported earlier (ie: Meteor Lake has been pushed from a paper launch in Q2 to 2023 because of Intel 4, and they don't know what to do with the TSMC wafers they ordered and now don't need)

That's not quite what the Digitimes report said, and either way, N3 wasn't in the cards for MTL.

And as a reminder of some historical predictions...

TSMC to Kick off Mass Production of Intel CPUs in 2H21 as Intel Shifts its CPU Manufacturing Strategies, Says TrendForce

Intel has outsourced the production of about 15-20% of its non-CPU chips, with most of the wafer starts for these products assigned to TSMC and UMC, according to TrendForce’s latest investigations. While the company is planning to kick off mass production of Core i3 CPUs at TSMC’s 5nm node in...

www.trendforce.com

 

[jpiniero]

That's not quite what the Digitimes report said, and either way, N3 wasn't in the cards for MTL.

So what do you think these N3 wafers were for if it wasn't for Meteor's IGP? Unless you don't believe that Intel had bought N3 wafers for 2H 22/1H 23 period.

 

[FangBLade]

here is another positive news for @Henry swagger to talk about: https://www.theregister.com/2022/08/05/intel_is_late_again/

 

[Saylick]

Trendforce insights are 💩

More like Trendfarce, amirite?

 

[Asterox]

Kzranich should be hired to break up monopolies, he’s really good at that.

Old art, but still.........................................

 

[Exist50]

Unless you don't believe that Intel had bought N3 wafers for 2H 22 period.

Well that's a likely possibility in its own right, given that N3 availability de facto starts in 2023.

But to address the broader point, they could have wanted N3 for server (good match for the Forest line in particular), Arrow Lake (as leaked), Lunar Lake (still planned?), and/or dGPUs. If anything, using limited leading node capacity on a mainstream iGPU seems like an incredible waste, wouldn't you agree?