Discussion Intel current and future Lakes & Rapids thread

[Ajay]

I agree. But, I do still find it quite enjoyable to cobble together the different pieces of verifiable data and wild rumors to try to get the best estimate of the future. No estimate is infallible though.

I find DrMrLord's conjecture laughable. He is essentially claiming that since one die might be 6+8 therefore desktop Meteor Lake does not exist. As a thought comparison, how many Alder Lake dies are there? At least 3 (desktop, mobile, and ultramobile). Plus a rumored Alder Lake-X which might make 4 dies if that rumor turns out to be true. So, how on earth does the presence of one Meteor Lake die prevent other Meteor Lake dies from existing?

I imagine that @DrMrLordX is operating on the understanding that Intel have very few of the EUV machines needed to produce those compute dice for ML. This was based on a report from a well respected analyst in the semiconductor field. ASML is filling orders on a first come first served basis. Intel, under Krzanich cancelled orders (!!) and they are now playing catch-up. Seems like they won’t be in a good position WRT advanced node production capacity till around 2025 (in house capacity).

 

[dullard]

I imagine that @DrMrLordX is operating on the understanding that Intel have very few of the EUV machines needed to produce those compute dice for ML. This was based on a report from a well respected analyst in the semiconductor field. ASML is filling orders on a first come first served basis. Intel, under Krzanich cancelled orders (!!) and they are now playing catch-up. Seems like they won’t be in a good position WRT advanced node production capacity till around 2025 (in house capacity).

What that is all likely to be true, it is also fairly irrelevant.

1) EUV is only used on a small portion of the Meteor Lake back end of line. Most of the rest of the CPU tile is made with conventional equipment.
2) The majority of the tiles (and thus the majority of the area) are not manufactured at Intel.
Combine #1 and #2 and there just isn't much EUV needed for Meteor Lake. Even if Intel cancelled a lot of ASML machines, they still have enough for Meteor Lake.

3) Intel is most likely not launching mobile and desktop Meteor Lake at the same time. Meaning their limited EUV capability will be stretched out over more time. It looks like they will manufacture the mobile tiles first and then the desktop tiles later.
4) Intel has the opportunity to utilize TSMC for the CPU tile if necessary.
5) There is an intentional overlap of Raptor Lake, Meteor Lake, and Arrow Lake. Meteor Lake is being jammed in for a short period of time. Thus, it won't face the pressure of trying to meet all of the market demand on just Meteor Lake. Meaning if there is overwhelming market demand, their limited EUV capability does not need to fulfill it all.

Note: this is not true for Intel 3 where lots of Intel EUV is necessary as they are trying to use EUV extensively. That is most likely leading to their rumored reshuffling of schedules amongst client and server chips.

 

[igor_kavinski]

I think Intel isn't in any trouble if it can ride out 2023 with just Raptor Lake as their desktop offering.

 

[Saylick]

I think Intel isn't in any trouble if it can ride out 2023 with just Raptor Lake as their desktop offering.

Intel has a lot of market share they can afford to lose before it becomes a big deal. Even when Zen 3 was beating 14nm+++ Intel month after month in sales, it didn't leave a permanent mark.

 

[JasonLD]

I think Intel isn't in any trouble if it can ride out 2023 with just Raptor Lake as their desktop offering.

I am sure they will release something, and mostly like it seems there will be a Meteor Lake desktop. If they were going to release something else, we would have known already. Comet Lake and Rocket Lake desktop chips leaked out years ago befor launch.

 

[biostud]

I think Intel isn't in any trouble if it can ride out 2023 with just Raptor Lake as their desktop offering.

I still don't see how they in any way can compete in performance/watt with Intel 7 vs TSMC 5nm.

 

[dullard]

I still don't see how they in any way can compete in performance/watt with Intel 7 vs TSMC 5nm.

Intel can't compete on that metric. Moving more and more to the E-cores will help a bit, but not enough. Intel can compete on other aspects though: single threaded or lightly threaded applications and availability (since TSMC seems to be reaching their limits to provide chips).

 

[igor_kavinski]

I still don't see how they in any way can compete in performance/watt with Intel 7 vs TSMC 5nm.

They don't have to compete. People are still buying their 10th gen CPUs all around the world, mostly the i3s and i5s. They will manage just fine.

 

[biostud]

Intel can't compete on that metric. Moving more and more to the E-cores will help a bit, but not enough. Intel can compete on other aspects though: single threaded or lightly threaded applications and availability (since TSMC seems to be reaching their limits to provide chips).

Apple will move away from 5nm freeing up some capacity, but since both AMD and Nvidia is also moving their GPUs to 5nm it might still be constrained. And I agree we're not going to see a marketslide.

 

[biostud]

They don't have to compete. People are still buying their 10th gen CPUs all around the world, mostly the i3s and i5s. They will manage just fine.

Yeah, I have no doubt about that. Most people don't have 10k+ posts on computer forums and will buy whatever the sales person tells them.
But for someone as me who actually cares, I see no compelling reason to choose Raptor lake over zen4.

 

[DrMrLordX]

So, you have nothing but your own hopes to base your posts on. I see.

Draw whatever conclusion you want; meanwhile, Intel has only shown one die and has released no data to the contrary. Go on being blind if you like.

@Ajay

It's been rumoured (yes, rumoured!) that Meteor Lake was at least considered for N3 some time ago, a consideration that was initially dismissed by those hoping for better performance from their own manufacturing. If they really want to launch a desktop product in 2023 then N3 continues to stand ready. They could pivot, and potentially do so quickly by just adding another compute tile, but from what everyone's said the current 6+8 tile that they were apparently showing off to the public already is the highest core-count tile they've got. That being said, declaring "woops!" and trying to port the compute tile over to N3 on a moment's notice would be neither simple nor quick, unless they may this decision 6+ months ago.

But yes otherwise, it does look like Intel is struggling with Intel 4 volume enough that a mobile-only Meteor Lake seems plausible.

 

[Doug S]

Apple will move away from 5nm freeing up some capacity, but since both AMD and Nvidia is also moving their GPUs to 5nm it might still be constrained. And I agree we're not going to see a marketslide.

The next iPhone will use N4, which is still part of the N5 family and not freeing up capacity. Plus Apple keeps selling older models (i.e. currently alongside iPhone 13 they are selling iPhone 12s at $100 off, and one model of iPhone 11 at $200 off) so all of those older iPhones still being sold will be made using N5. Then there's the new SE which is also N5 and will stick around for 2-3 years. The two year old iPhone 11 dropping off for a two year old iPhone 12 will shift some N7 to N5.

In addition, if rumors are true that Apple will use A15 for the non-Pro versions of iPhone 14 then even if you want to count N4 as separate (it isn't) Apple will be overwhelmingly N5 based for iPhone 14. I'm skeptical about those rumors since N4 wafers should cost less than N5 wafers, but if Apple has made the SoC larger via bigger cores or more cache then maybe there's enough of a cost advantage to using one year old SoCs in non-Pro phones.

TSMC is still expanding N5 generation capacity, that's where you will get your new capacity, you won't get it from Apple until they start shipping stuff made using N3/N3E.

 

[IntelUser2000]

How many test chips do they usually go through before it's ready for production?

I didn't like them using tape-in. Tape-outs are better, and usually takes about a year before launch to consumers.

 

[nicalandia]

I think Raptor Lake on desktop will have a good chance of competing favorably with Zen 4 in single threaded workloads, and edging it in gaming.

I believe you are being overoptimistic. Raptor Lake is an enhancement of Alder Lake, any IPC gains will be under 10%


Also..

Intel Fishhawk Falls HEDT 'Sapphire Rapids-X' Platform Detailed: 24 Cores at 5 GHz For Mainstream, Up To 112 Cores With 8-Channel Memory For Expert Workstations

The latest information regarding Intel's Fishhawk Falls HEDT platform which will feature Sapphire Rapids-X 'Xeon-W' CPUs has been leaked

wccftech.com

 

[Mopetar]

Intel has a lot of market share they can afford to lose before it becomes a big deal. Even when Zen 3 was beating 14nm+++ Intel month after month in sales, it didn't leave a permanent mark.

AMD is limited in how much they can produce and if Intel has less compelling products it will just mean that AMD grows most in the high margin areas while Intel sells more chips in the market segments that AMD can't fully supply or where they're more price competitive.

Intel's overall share won't change much, but their margins would go down. Even then they'd still be a massively profitable company.

 

[exquisitechar]

I believe you are being overoptimistic. Raptor Lake is an enhancement of Alder Lake, any IPC gains will be under 10%

Probably, but Alder Lake is significantly ahead of Zen 3 and Raptor Lake will probably have a bit of an advantage in clock speeds. I don't think it's particularly overoptimistic to assume that Raptor Lake may have the ST performance crown. That said, I think they will be close either way. As for gaming performance, Zen 4 certainly has a lot of ground to cover considering how much faster than Zen 3 ADL is already. Who knows how the memory subsystem/caches of each of them will affect gaming performance at this point?

 

[Timmah!]

I believe you are being overoptimistic. Raptor Lake is an enhancement of Alder Lake, any IPC gains will be under 10%


Also..

Intel Fishhawk Falls HEDT 'Sapphire Rapids-X' Platform Detailed: 24 Cores at 5 GHz For Mainstream, Up To 112 Cores With 8-Channel Memory For Expert Workstations

The latest information regarding Intel's Fishhawk Falls HEDT platform which will feature Sapphire Rapids-X 'Xeon-W' CPUs has been leaked

wccftech.com

that 24-core SPR interests me very much. If its 4,5 all core and 5+ on 2~8 of them, while the price will be no more than what 3960x did cost (around 1500), then count me in.
Alternatively to that 24~32 core Ryzen 5, whatever is better value and comes sooner.

 

[nicalandia]

that 24-core SPR interests me very much. If its 4,5 all core and 5+ on 2~8 of them, while the price will be no more than what 3960x did cost (around 1500), then count me in.
Alternatively to that 24~32 core Ryzen 5, whatever is better value and comes sooner.

The 24 Core SPR will be a MCM with 12 cores enabled on two tiles, the other two tiles will be dummy silicon for support.

 

[Timmah!]

The 24 Core SPR will be a MCM with 12 cores enabled on two tiles, the other two tiles will be dummy silicon for support.

Zen4 based TR is going to be released at the end of 2023. In the mean time ThreadRipper PRO 5000 will be wrecking all sorts of havoc. 8 Channel memory, 64 cores on 2022

they claim that 24 core chip will be monolithic mcc die, not mcm. But its just a rumor for sure.
64 core threadripper does not interest me too expensive and dont need that many cores. 24/32 core ryzen on other hand…

 

[nicalandia]

they claim that 24 core chip will be monolithic mcc die, not mcm. But its just a rumor for sure.

Not a chance of being a single Monolith chunk of silicon, it's going to be like this, SPR is limited by 2 channels per compute tile(So four channel will be two compute tiles).. The "Monolithic" term is used as "Software Monolithic"

 

[igor_kavinski]

If they get that chip onto a workstation with >100MB shared cache, it would make a really interesting (and expensive) gaming machine.

 

[nicalandia]

If they get that chip onto a workstation with >100MB shared cache, it would make a really interesting (and expensive) gaming machine.

Two tiles will give it a maximum of 50MiB per CPU. The only way to make that work(100MiB shared cache) would be to have all of the 4 tiles active and that would be quite expensive.

 

[Timmah!]

Not a chance of being a single Monolith chunk of silicon, it's going to be like this, SPR is limited by 2 channels per compute tile(So four channel will be two compute tiles).. The "Monolithic" term is used as "Software Monolithic"

View attachment 61185

why would it be no chance?

 

[nicalandia]

why would it be no chance?

It's the way Sapphire Rapids was designed. The Maximum core count Sapphire Rapids will have is 56 cores, 8 memory channels and 100 MiB. This means that each tile is capable of 14C/28T, 2 memory channel and 25 MiB per compute tile.

The leak suggest 24 Cores and 4 memory channels. This means that Intel is using two tiles on that Sapphire Rapids CPU. And while the Leak does not mentions the Cache, we can assume it will not exceed 50 MiB

 

[Timmah!]

It's the way Sapphire Rapids was designed. The Maximum core count Sapphire Rapids will have is 56 cores, 8 memory channels and 100 MiB. This means that each tile is capable of 14C/28T, 2 memory channel and 25 MiB per compute tile.

but they imply in their rumor or leak whatever it is, there is another SPR die, one not meant to be used as a tile, but regular monolithic chip. Surely thats not impossible?