Discussion Intel current and future Lakes & Rapids thread

[eek2121]

I think lunar lake will use tsmc n3e

No. Intel is not planning to put it's core IP on a non-Intel node at this point in time.

The slide did say MTL were having yield issues: Really meh, after so many years in 7nm process, they still cannot get past 6 big cores in MTL. So my question is how long Intel going to bring 44 p cores in Intel 3 to acceptable yields and ship it?

Come on Intel, I don't want to lose the bet, please ship GNR before end of 2025

Intel is not having yield issues on Intel 4. He is full of it, as usual. MTL is a mobile first chip and one of the reasons for sticking to 6 cores is die sizes. It will take Intel a while to ramp up N4 production due to the increased use of EUV as well as other changes. Because of this, it is in their best interest to target only a specific market segment, while using Intel 7 to target the rest of the market.

 

[Timmah!]

Do we know the pricing for the Xeons 2400/3400 yet?

 

[mikk]

No. Intel is not planning to put it's core IP on a non-Intel node at this point in time.

Arrow Lake-S uses TSMC 3nm according to Raichu and both Arrow and Lunar seems to use Lion Cove, Intel likely could use both TSMC 3nm or Intel node. Don't forget that Arrow Lake-P was planned to use TSMC 3nm as well, we know this from an early roadmap.

 

[nicalandia]

Do we know the pricing for the Xeons 2400/3400 yet?

Not yet, but the 13900KS is a $700 Part, so at the least I would say $1500 for the 16C part

 

[IntelUser2000]

What's the point of Emerald Rapids having only 64 cores? So they are creating an entirely new layout just for 1 more core per tile?

It's more logical to me that Sapphire Rapids will get a revision and small update like prefetchers and use the same 15 core die to get 60 cores, or 18 core die they couldn't manufacture will come back and we'll see 72 cores.

Either way they can release Emerald Rapids similar to Cooper Lake where there was no high end, and when Granite Rapids arrives it can take over the high end.

 

[Exist50]

What's the point of Emerald Rapids having only 64 cores? So they are creating an entirely new layout just for 1 more core per tile?

It's more logical to me that Sapphire Rapids will get a revision and small update like prefetchers and use the same 15 core die to get 60 cores, or 18 core die they couldn't manufacture will come back and we'll see 72 cores.

Either way they can release Emerald Rapids similar to Cooper Lake where there was no high end, and when Granite Rapids arrives it can take over the high end.

It should make much more sense once die shots leak.

 

[A///]

My guess is specific industries or markets where they need that core count? AMD designs the silicon, they don't create it or package it. A silicon flaw would be widespread, not 3 seemingly dead cpus sent to a dipstick youtuber to talk about.

 

[semiman]

What's the point of Emerald Rapids having only 64 cores? So they are creating an entirely new layout just for 1 more core per tile?

It seems that Sapphire rapid is delayed too much. Probably they've been working on Emerald rapids in parallel, but SPR's too late.

 

[IntelUser2000]

It seems that Sapphire rapid is delayed too much. Probably they've been working on Emerald rapids in parallel, but SPR's too late.

You can't parallel the teams the same way as you parallel something else like say, roads for example. You need the previous gen to work on the last gen. It could be argued that Sapphire Rapids was so delayed the emergency backup came too close to the original product so it looks like "parallel".

Either way 64 cores is nothing. They pretty much need to do the Cooper Lake strategy where there's no high end line up at all, just mid end.

@A/// Huh? It's Intel not AMD. Intel plans, builds, and manufactured Sapphire Rapids.

Also it almost sounds like AMD has very little to do with chip making. TSMC is pretty much a printer in this case. AMD does all the work.

 

[A///]

You can't parallel the teams the same way as you parallel something else like say, roads for example. You need the previous gen to work on the last gen. It could be argued that Sapphire Rapids was so delayed the emergency backup came too close to the original product so it looks like "parallel".

Either way 64 cores is nothing. They pretty much need to do the Cooper Lake strategy where there's no high end line up at all, just mid end.

@A/// Huh? It's Intel not AMD. Intel plans, builds, and manufactured Sapphire Rapids.

Also it almost sounds like AMD has very little to do with chip making. TSMC is pretty much a printer in this case. AMD does all the work.

It's simple. I'll repeat it. TSMC sends engineers to AMD to work with AMD to get their designs working with the nodes chosen. TSMC does this with Apple, too. TSMC is doing more work for Zen 4 due to their tech being used for more than the core chiplets. Final buildup and packaging (not the box) is done by TSMC too according to what I'd read months before Zen 4 came out. I worked at Broadcom for a number of years and we'd get engineers in from whatever fab we were using at the time to make our designs.

Intel while floundering out of sea and roughly out of mind can still build and package everything in house for specific cloud customers, for example.

 

[DrMrLordX]

It seems that Sapphire rapid is delayed too much. Probably they've been working on Emerald rapids in parallel, but SPR's too late.

That's definitely part of it. But Emerald Rapids was supposed to go higher than 64c.

 

[mderbarimdiger]

i dont think EMR was ever supposed to go higher than 64 cores,
To (over)simplify things you might consider EMR to be to SPR the same as Raptor is to Alder lake.

Performance per watt should be better (it is using the same process node as RPL), more native memory, higher clock speeds, and a it have an extra nice feature for cloud providers to to manage virtual machines more secured.

techpowerup sums it up this way:
Built on the same LGA4677 package as "Sapphire Rapids," the new "Emerald Rapids" MCM packs up to 64 "Raptor Cove" CPU cores, which support higher clock-speeds, higher memory speeds, and introduce the new Intel Trust Domain Extensions (TDX) instruction-set. The processor retains the 8-channel DDR5 memory interface, but with higher native memory speeds. The chip's main serial interface is a PCI-Express Gen 5 root-complex with 80 lanes. The processor will be built on the last foundry-level refinement of the Intel 7 node (10 nm Enhanced SuperFin); many of these refinements were introduced with the company's 13th Gen Core "Raptor Lake" client processors

 

[mderbarimdiger]

by the way, since EMR is already in testing phase with select costumers, i would expect some benchmark leaks or pre production photos to show up soon.

 

[semiman]

That's definitely part of it. But Emerald Rapids was supposed to go higher than 64c.

Probably not really useful even if they pack up more cores. They made AVX powerhouse with enhanced Golden Cove cores but still have 8 memory channels. Awkward times indeed...

 

[Markfw]

Probably not really useful even if they pack up more cores. They made AVX powerhouse with enhanced Golden Cove cores but still have 8 memory channels. Awkward times indeed...

But who cares about 8 memory channels. When you are facing a CPU core thats almost as fast per core, takes less watts to run, AND has 50-100% more cores and 50% more memory channels, you have no chance.

 

[LightningZ71]

The HBM parts should do very well with AVX-512 code vs. The field. I suspect that those EMR parts won't care about AMD's dram channel advantage or higher core counts too much .

 

[mderbarimdiger]

But who cares about 8 memory channels. When you are facing a CPU core thats almost as fast per core, takes less watts to run, AND has 50-100% more cores and 50% more memory channels, you have no chance.

AMD is not going to have 30% market share in 18 months from now, regardless how good of a product they have and regardless of how much wafers they can get from TSMC. you seem to think that everyone is as excited as you to jump for genoa because its a better product, but thats not the case, AMD does not expect EPYC to grow by 40% YoY, please accept the fact that intel will outsell AMD at least for another few years.

reasons for that are manyfold, including some stupid data center managers who are not ready to pay the higher cost upfront (genoa does have a higher cost upfront) there are channel relationships, there are process and procedures for entire big teams that is based on intel's products, thats not going to change overnight, some folks heavily rely on intel accelerators, (yups! not everyone has the same needs for some folks SPR is the better choice, Nvidia and Meta are known to chose intel over AMD for performance reasons) and the list goes on and on and is irrelevant regardless, the fact is just that Xeon will still outsell EPYC 3X by the end of the year.

So, those folks care about intel having 8 channels vs 12 channels.

 

[Hitman928]

AMD is not going to have 30% market share in 18 months from now, regardless how good of a product they have and regardless of how much wafers they can get from TSMC. you seem to think that everyone is as excited as you to jump for genoa because its a better product, but thats not the case, AMD does not expect EPYC to grow by 40% YoY, please accept the fact that intel will outsell AMD at least for another few years.

reasons for that are manyfold, including some stupid data center managers who are not ready to pay the higher cost upfront (genoa does have a higher cost upfront) there are channel relationships, there are process and procedures for entire big teams that is based on intel's products, thats not going to change overnight, some folks heavily rely on intel accelerators, (yups! not everyone has the same needs for some folks SPR is the better choice, Nvidia and Meta are known to chose intel over AMD for performance reasons) and the list goes on and on and is irrelevant regardless, the fact is just that Xeon will still outsell EPYC 3X by the end of the year.

So, those folks care about intel having 8 channels vs 12 channels.

Revenue share wise, AMD's datacenter group is already at 27% share compared to Intel's and this will only continue to grow. It's not a perfect comparison as both have other products besides just server CPUs in their groups, but I think AMD is doing much better in server CPU market share than you realize. It is very possible that AMD could be at 30% volume share in 18 months.

 

[mderbarimdiger]

No doubt AMD has a super great product that is far better than Intel is offering, i am not sure if they gain more in revenue share but definitely they have a higher profit margin, intel is trying to compete with AMDs superior product with aggressive pricing tactics.

However my comment was to markfw who seems to think of intel server chips as a irrelevant matter entirely besides for the history of server chips, but the matter of fact is that Xeon is selling 3X the sales of EPYC, and that will still be the case by the end of this year, so dismissing intel news is perhaps an inaccurate assumption that because AMD is having a better product they will just put intel to rest.

 

[Saylick]

No doubt AMD has a super great product that is far better than Intel is offering, i am not sure if they gain more in revenue share but definitely they have a higher profit margin, intel is trying to compete with AMDs superior product with aggressive pricing tactics.

However my comment was to markfw who seems to think of intel server chips as a irrelevant matter entirely besides for the history of server chips, but the matter of fact is that Xeon is selling 3X the sales of EPYC, and that will still be the case by the end of this year, so dismissing intel news is perhaps an inaccurate assumption that because AMD is having a better product they will just put intel to rest.

This. Even as an AMD investor who is well aware of the superiority of Genoa, I have to acknowledge that Intel has a ton of institutional backing that makes it hard for AMD to claw market share. The same goes for Nvidia vs. AMD in the GPU space. There's an implicit moat that one benefits from when one supplies the vast majority of the market. Even if one's competitor has a product that is 2x better than yours, most customers are set in their ways and don't have the incentive to make the switch. Of the ones that do decide to switch, your loss in market share is further limited by how much supply your competitor can deliver. If you already owned the vast majority of the market, your competitor likely cannot scale supply by a significant amount within a short period of time because they likely don't have the financial means to do so, nor would it be feasible even if they had unlimited money. We all wish the free market can reward better products and promote competition overnight, but it simply doesn't work that way, especially not in an industry known for exceptionally long lead times that require forecasting years out in advance.

 

[jpiniero]

Interesting... the loss associated with the Intel 4 ramp is not very much for client versus server or even the Network/Edge. Maybe it suggests Intel really isn't serious about Client products on Intel 4 for the time being and is mainly focusing on getting the server products working. Which kinda makes sense from a priority perspective but Granite Rapids better be very yieldable if this is true though.

 

[Markfw]

No doubt AMD has a super great product that is far better than Intel is offering, i am not sure if they gain more in revenue share but definitely they have a higher profit margin, intel is trying to compete with AMDs superior product with aggressive pricing tactics.

However my comment was to markfw who seems to think of intel server chips as a irrelevant matter entirely besides for the history of server chips, but the matter of fact is that Xeon is selling 3X the sales of EPYC, and that will still be the case by the end of this year, so dismissing intel news is perhaps an inaccurate assumption that because AMD is having a better product they will just put intel to rest.

I am not sure exactly what you mean, but I am saying that AMD has far superior server chips, and has for years, and this is finally supported by their ever increasing market share. I don't see that changing until Intel finally has a decent server ship. And I do think they have a good chance of hitting 30% in 18 months. I don't see it ending there either.

 

[Exist50]

Interesting... the loss associated with the Intel 4 ramp is not very much for client versus server or even the Network/Edge. Maybe it suggests Intel really isn't serious about Client products on Intel 4 for the time being and is mainly focusing on getting the server products working. Which kinda makes sense from a priority perspective but Granite Rapids better be very yieldable if this is true though.

The bulk of MTL shipments probably won't show up till next year, and it uses minimal Intel 4 silicon anyway. Not to mention questions about which markets they actually ship it in.

Granite, however, is going to take a ton of Intel 3 silicon, so yeah, they better hope it's in good shape by then.

 

[Saylick]

The bulk of MTL shipments probably won't show up till next year, and it uses minimal Intel 4 silicon anyway. Not to mention questions about which markets they actually ship it in.

Granite, however, is going to take a ton of Intel 3 silicon, so yeah, they better hope it's in good shape by then.

Do we have a latest update on how much EUV wafers Intel can produce? As you mentioned, GNR will use an EUV node and knowing how much of the market Intel needs to supply, it will take a decent number of EUV machines to produce all the compute tiles.

 

[Abwx]

And I do think they have a good chance of hitting 30% in 18 months. I don't see it ending there either.

This quarter eventually, the last numbers from AMD and Intel put the former at 27.7% revenue marketshare, and it would require 18 months to get to 30%..?...

Their sales amount to 38.45% of Intel s ones, that is 1.655 bn and 4.305 bn respectively.