Discussion Intel current and future Lakes & Rapids thread

[semiman]

At least Intel caught up quite fast in every metric. We already know that Alder Lake 12900K with 65W is on par with Rocket Lake 11900K with 250W envelope(11900K scored 5900~) from Intel's presentation so 6+8 will be also quite nice.

If we create a hypothetical 12+16 complex, then its core complex will be a little bit bigger than 2 x Zen 3 full chiplet(80mm2 x 2). It'll consume 130W of power with 1.7~2.0 better than 11900K@250W. If we compare it with results from Tiger Lake, then we can see that Intel caught up really fast.

Intel will enjoy better market shares in laptops and desktops except for pseudo-HEDT 5950X range for a while(Maybe 6 months).
Zen 3D will not help AMD much due to massive silicon consumption. Then Zen 4 will be there with TSMC 5nm, then Intel will have a tough time again due to manufacturing inferiority. A hoard of small cores(8+16) will negate some of its disadvantages but a better Zen 4 IPC, combined with new manufacturing will hit intel really hard. The only thing Intel can do that time is pray. Just hope for a less space for TSMC 5nm.
Then there will be Intel mix-and-match chips with new Intel 4 manufacturing, negating AMD chiplet advantages and manufacturing superiority while keeping big.LITTLE approach only in Intel. A nice seesaw game awaits us.

 

[eek2121]

At least Intel caught up quite fast in every metric. We already know that Alder Lake 12900K with 65W is on par with Rocket Lake 11900K with 250W envelope(11900K scored 5900~) from Intel's presentation so 6+8 will be also quite nice.

If we create a hypothetical 12+16 complex, then its core complex will be a little bit bigger than 2 x Zen 3 full chiplet(80mm2 x 2). It'll consume 130W of power with 1.7~2.0 better than 11900K@250W. If we compare it with results from Tiger Lake, then we can see that Intel caught up really fast.

Intel will enjoy better market shares in laptops and desktops except for pseudo-HEDT 5950X range for a while(Maybe 6 months).
Zen 3D will not help AMD much due to massive silicon consumption. Then Zen 4 will be there with TSMC 5nm, then Intel will have a tough time again due to manufacturing inferiority. A hoard of small cores(8+16) will negate some of its disadvantages but a better Zen 4 IPC, combined with new manufacturing will hit intel really hard. The only thing Intel can do that time is pray. Just hope for a less space for TSMC 5nm.
Then there will be Intel mix-and-match chips with new Intel 4 manufacturing, negating AMD chiplet advantages and manufacturing superiority while keeping big.LITTLE approach only in Intel. A nice seesaw game awaits us.

Zen 4 is competing with Raptor Lake (Raptor Cove + double or possibly triple Gracemont). AMD may still have an edge there (or possibly not), but once Meteor Lake launches they will have a tough time.

If AMD wants to stay competitive they will either need to hit harder or move to a faster cadence.

 

[ondma]

Zen 4 is competing with Raptor Lake (Raptor Cove + double or possibly triple Gracemont). AMD may still have an edge there (or possibly not), but once Meteor Lake launches they will have a tough time.

If AMD wants to stay competitive they will either need to hit harder or move to a faster cadence.

Assuming Intel will meet it timeline targets. I hope they do, but certainly am not counting on it.

 

[DrMrLordX]

but once Meteor Lake launches they will have a tough time.

Meteor Lake may be very niche. There are huge questions about volume for anything on Intel 4.

 

[sierpp]

Hi everyone, I'm new here. Mainly lurking the forum but i found something interesting in Intel slides and would like to share it with you. Don't know if it wasn't already posted though.
So it's reasonable to keep expectations regarding ADL-S in check.
Sorry for my English, it's not my native language. If i do some mistakes please correct me Thanks!

 

[JoeRambo]

That is damn generuos of Intel to use 3200C14 flat for 5950X, that is a nice boost over run of the mill CL18 junk. And also generous to use only DDR5 4400C36 for their own system. Not the worst case for ADL mem config, but still.

 

[pakotlar]

Hi everyone, I'm new here. Mainly lurking the forum but i found something interesting in Intel slides and would like to share it with you. Don't know if it wasn't already posted though.
So it's reasonable to keep expectations regarding ADL-S in check.
Sorry for my English, it's not my native language. If i do some mistakes please correct me Thanks!

View attachment 52255

View attachment 52256

105W PL1 is the TDP for 5950x, and is not the max power draw (142W)
241W PL1 for ADL is the max power draw, as PL1=PL2.

Intel configured the 5950x correctly for stock.

 

[Abwx]

105W PL1 is the TDP for 5950x, and is not the max power draw (142W)
241W PL1 for ADL is the max power draw, as PL1=PL2.

Intel configured the 5950x correctly for stock.

Nothing in the slide says how PL2 is configured for the 5950X, it could be PL1 = PL2 , like in their own setting...

 

[sierpp]

Nothing in the slide says how PL2 is configured for the 5950X, it could be PL1 = PL2 , like in their own setting...

Also those tests were done using Win 11 compilation without AMD fixes, so it looks like it's some form of disadvantage as well as mentioned above vague statements regarding power limits.

 

[semiman]

Zen 4 is competing with Raptor Lake (Raptor Cove + double or possibly triple Gracemont). AMD may still have an edge there (or possibly not), but once Meteor Lake launches they will have a tough time.

If AMD wants to stay competitive they will either need to hit harder or move to a faster cadence.

Raptor Lake will confront tough competition for sure since an extra 20~30% ish MT improvement(some from Cove IPC, the other from extra Gracemont) can be easily achieved by rumored Zen 4 IPC improvement alone(20%) while enjoying power saving from N5 which Raptor Lake cannot really enjoy.
If needed, AMD can make 3-chiplet variant for desktop CPU to maintain leadership in Desktop. Intel EMIB will not be in a desktop segment to parry this attack.

After that, Meteor lake will be quite competitive due to superior interconnect technology but Intel 4 volume will not be enough. Intel will need to put Raptor Lake Refresh and Meteor Lake together at that timeframe. But still better position than what they suffered in the 2018~2019 timeframe.

If everything goes as planned, then Intel will get nice manufacturing facility for post Intel 4, combined with EMIB and Foveros and new "Monts" architecture in 2024. So I think it's post Meteor lake where Intel really poses threat to AMD, not by volume but by technology. Pretty sure Monts will be the real workhorses of x86 in near future.

 

[pakotlar]

Nothing in the slide says how PL2 is configured for the 5950X, it could be PL1 = PL2 , like in their own setting...

There is no PL2 for AMD processors.


Sent from my iPhone using Tapatalk

 

[eek2121]

The 12700k scores 22,689 points in CBr23 @160W. I found that interesting.

Meteor Lake may be very niche. There are huge questions about volume for anything on Intel 4.

Meteor Lake would be Core 14xxx series. I doubt we will see further delays. Intel has already produced and powered on the early silicon with no issues.

 

[nicalandia]

Meteor Lake would be Core 14xxx series. I doubt we will see further delays. Intel has already produced and powered on the early silicon with no issues.

There is a Huge difference between Powering On Early Silicon and full fledge production. Everything we have read so far is pointing to a limited supply of Intel 4 Products

 

[dmens]

The 12700k scores 22,689 points in CBr23 @160W. I found that interesting.



Meteor Lake would be Core 14xxx series. I doubt we will see further delays. Intel has already produced and powered on the early silicon with no issues.

LOL. I debugged a Cannonlake A0 in the lab in 2016. Hit the market right on time.

 

[eek2121]

There is a Huge difference between Powering On Early Silicon and full fledge production. Everything we have read so far is pointing to a limited supply of Intel 4 Products

Such as? Keep in mind we are around 2 years from production at this point and that Intel has already begun shipping chips, despite Intel 4 being pre-production. Meteor Lake has been taped out and Pat Gelsinger said all planned future nodes were on or ahead of schedule just 2 days ago.

 

[TESKATLIPOKA]

The 12700k scores 22,689 points in CBr23 @160W. I found that interesting.

There is also 8658 points in R20 with PL1: 241W but 160W package power.
From Lenovo leak It was 8677 points in R20 with 241W.

 

[TESKATLIPOKA]

Core i9 12900k
218 Watts in Cinebench 20/23.
Twitter

Core i5 12600K
115 Watts in Cinebench 20/23.
Twitter

 

[dullard]

Everything we have read so far is pointing to a limited supply of Intel 4 Products

So far, I've seen three pieces of data. Eek2121 mentioned two of them (Loihi 2 preproduction and Meteor Lake Power On). The third piece of information is that Intel does not have many EUV lithography instruments (Samsung and TSMC have most). Have you read other information on this?

Meteor Lake is supposed to have only some EUV. It is Arrow Lake on Intel 3 that has a lot more EUV. Plus only the CPU tile of Meteor Lake is supposed to be on Intel 4. The GPU and the rest of the SOC chip portions could be produced at TSMC or on a different Intel node (such as Intel 7). Having only 1 out of 3 tiles on Intel 4 will help with the Meteor Lake production bottleneck.

 

[Zucker2k]

Core i9 12900k
218 Watts in Cinebench 20/23.
Twitter

Core i5 12600K
115 Watts in Cinebench 20/23.
Twitter

So, the 12600k is 50% faster than the 5600x in this test while using 30% more power, and beats both the 10900k and 11900k while using half the power of those chips. That's freaking insane efficiency!

 

[Hulk]

Zen 4 is competing with Raptor Lake (Raptor Cove + double or possibly triple Gracemont). AMD may still have an edge there (or possibly not), but once Meteor Lake launches they will have a tough time.

If AMD wants to stay competitive they will either need to hit harder or move to a faster cadence.

The fact that the V-Cache on Zen 3 seems to show a significant performance increase leads me to believe that the Zen 3 microarchitecture is not being sufficiently "fed." Zen 3 is crazy wide so the additional 15% AMD claims with the additional cache seems plausible. Now I'm wondering if Zen 3 with the V-Cache will benefit much when it goes to DDR5?

 

[Ajay]

Meteor Lake is supposed to have only some EUV. It is Arrow Lake on Intel 3 that has a lot more EUV. Plus only the CPU tile of Meteor Lake is supposed to be on Intel 4. The GPU and the rest of the SOC chip portions could be produced at TSMC or on a different Intel node (such as Intel 7). Having only 1 out of 3 tiles on Intel 4 will help with the Meteor Lake production bottleneck.

Well, Meteor Lake will be for laptop and desktop (not sure when Sapphire Rapids drops). A reliable industry source** indicated that Intel would be limited to 20K WPM. I assume they made that calculation with the expected number of EUV passes in mind. Intel's typical run rate peaks over 65K WPM on it's leading node (some sources list higher rates than that, IDK). So, Meteor Lake and related products will be in short supply for the first year. Since Intel supposedly cancelled some EUV shipments under Brian Krzanich's tenure - it will take a while for Intel to get enough units in from ASML to compete more fully (hence the use of TSMC 3N). I hope they have a lot more coming in 2024 and 2025. I'm sure chiplets will help and it's the right way to go as evidenced by AMD's success.


** Mizuho Securities projection:

 

[andermans]

So, the 12600k is 50% faster than the 5600x in this test while using 30% more power, and beats both the 10900k and 11900k while using half the power of those chips. That's freaking insane efficiency!

I think that is "just" because more and stronger cores (remember we go 6->10 cores and Golden Cove has ~40% more IPC than Comet Lake) allowing things to run at way more efficient clock frequencies. So a strong achievement from Intel but I don't think it is anywhere unexpected or insane.

 

[DrMrLordX]

@Ajay

Last I heard, Intel will be supplementing Intel 4/Intel 7nm with TSMC N3 but may be limited to 20kwpm from them as well, for an indeterminent period of time (having 20kwpm from your own fabs is different from having that allocation from TSMC for only a few months). And Intel is going to have to share Intel 4 and TSMC N3 wafers across their entire product lineup: Granite Rapids, Xe, Meteor Lake, and other crap.

Meteor Lake may not even show up on the desktop. Some were saying it was mobile-first or potentially mobile-only.

 

[dullard]

Well, Meteor Lake will be for laptop and desktop (not sure when Sapphire Rapids drops). A reliable industry source** indicated that Intel would be limited to 20K WPM. I assume they made that calculation with the expected number of EUV passes in mind. Intel's typical run rate peaks over 65K WPM on it's leading node (some sources list higher rates than that, IDK). So, Meteor Lake and related products will be in short supply for the first year. Since Intel supposedly cancelled some EUV shipments under Brian Krzanich's tenure - it will take a while for Intel to get enough units in from ASML to compete more fully (hence the use of TSMC 3N). I hope they have a lot more coming in 2024 and 2025. I'm sure chiplets will help and it's the right way to go as evidenced by AMD's success.


** Mizuho Securities projection:

If suppose the CPU portion of the chip is 1/3rd of the chip area, then it would require about 1/3rd the rate. 20KWPM is about 1/3rd of 65KWPM. Plus, since it isn't node locked to Intel, TSMC could fill in the gap. I just don't (yet) see the shortage. I'm not saying that it won't happen, I'm just saying that there isn't yet enough data that convinces me that Intel will have a production shortage. Now, suppose that the Meteor Lake GPU tile was also going to be on Intel 4, then you have a strong case.

 

[Ajay]

@Ajay

Last I heard, Intel will be supplementing Intel 4/Intel 7nm with TSMC N3 but may be limited to 20kwpm from them as well, for an indeterminent period of time (having 20kwpm from your own fabs is different from having that allocation from TSMC for only a few months). And Intel is going to have to share Intel 4 and TSMC N3 wafers across their entire product lineup: Granite Rapids, Xe, Meteor Lake, and other crap.

Meteor Lake may not even show up on the desktop. Some were saying it was mobile-first or potentially mobile-only.

That's what I am wondering. Oh, and thanks for the correction (Granite, not Sapphire Rapids).

If suppose the CPU portion of the chip is 1/3rd of the chip area, then it would require about 1/3rd the rate. 20KWPM is about 1/3rd of 65KWPM. Plus, since it isn't node locked to Intel, TSMC could fill in the gap. I just don't (yet) see the shortage. I'm not saying that it won't happen, I'm just saying that there isn't yet enough data that convinces me that Intel will have a production shortage. Now, suppose that the Meteor Lake GPU tile was also going to be on Intel 4, then you have a strong case.

Well, guess we don't know what the compute tile will be - how many cores, cache size, etc. Intel seems to be going larger on P Cores.