Discussion Intel current and future Lakes & Rapids thread

[mikk]

Between Tiger Lake to Alder Lake, Alder Lake to Raptor Lake and Raptor Lake to Meteor Lake the second is by far the smallest 1T performance improvement.

Your second and third option is unknown at this point and not more than a random guess from your side. Meteor Lake might not bring a major CPU IPC upgrade if it's a typical Intel tick core upgrade like Ivy Bridge, Broadwell, Cannonlake. And it remains to be seen if Intels first 7nm/4 product is able to reach the 1T clock speeds of the matured 10ESF/7.

Nobody expects an Alder Lake like IPC upgrade from Raptor Cove. The thing is jpiniero claims zero IPC upgrade which is in conflict with Intels Raptor Lake comment as well as MLIDs expectation.

 

[uzzi38]

Your second and third option is unknown at this point and not more than a random guess from your side. Meteor Lake might not bring a major CPU IPC upgrade if it's a typical Intel tick core upgrade like Ivy Bridge, Broadwell, Cannonlake. And it remains to be seen if Intels first 7nm/4 product is able to reach the 1T clock speeds of the matured 10ESF/7.

Hey, at least your last point is on the right line of thought. It's something, I guess.

 

[Hulk]

According to the Anandtech ADL architecture day article there will be three ADL dies. 8+8, 6+8, and 2+8.

Preliminary leak data is showing 12900 as 8+8, 12700 as 8+4, and 12600 as 4+4. https://wccftech.com/intel-12th-gen...k-for-394-euros-core-i5-12600k-for-287-euros/

If these configurations are the real deal then do you think the defect rate for the Gracemont cores is high? I'm wondering why the +4's? Seems like gimping the P's to 6 and 4 makes more sense than cutting back Gracemont, from the 12900 I mean. Unless of course defects are high in the Gracemont cores?

Also, is it all or nothing for Gracemont defect-wise? Meaning either it works (4 cores) or it doesn't?

 

[eek2121]

Between Tiger Lake to Alder Lake, Alder Lake to Raptor Lake and Raptor Lake to Meteor Lake the second is by far the smallest IPC improvement.

Depends on which rumors are true. I've only digested a bit of Raptor Lake, but any IPC increase is likely going to severely eat into thermals at this point. AMD did an amazing job with Zen2 -> Zen3, but I absolutely do NOT expect Intel to do the same with Golden Cove to Raptor Cove. Intel is overdue for a reboot IMO.

The real reason why Raptor lake is going to shine is double the "small" cores.

 

[eek2121]

According to the Anandtech ADL architecture day article there will be three ADL dies. 8+8, 6+8, and 2+8.

Preliminary leak data is showing 12900 as 8+8, 12700 as 8+4, and 12600 as 4+4. https://wccftech.com/intel-12th-gen...k-for-394-euros-core-i5-12600k-for-287-euros/

If these configurations are the real deal then do you think the defect rate for the Gracemont cores is high? I'm wondering why the +4's? Seems like gimping the P's to 6 and 4 makes more sense than cutting back Gracemont, from the 12900 I mean. Unless of course defects are high in the Gracemont cores?

Also, is it all or nothing for Gracemont defect-wise? Meaning either it works (4 cores) or it doesn't?

If you look closely at the specs, you'll see Intel is artificially segmenting things. There does not seem to be much in the way of actual defects. Note that I'm not saying Intel won't have defects, just that if they did (have a lot of defects), you'd see SKUs that are a bit more finer grained.

EDIT: clicked save too soon.

 

[moinmoin]

Also, is it all or nothing for Gracemont defect-wise? Meaning either it works (4 cores) or it doesn't?

Appears to work that way on a per module level indeed. But that does seem like a waste, especially considering how AMD handles CCXes in comparison.

 

[jpiniero]

If these configurations are the real deal then do you think the defect rate for the Gracemont cores is high? I'm wondering why the +4's? Seems like gimping the P's to 6 and 4 makes more sense than cutting back Gracemont, from the 12900 I mean. Unless of course defects are high in the Gracemont cores?

Didn't want to regress on the number of Big Cores on desktop. Also I saw stuff suggesting that it is possible to have partial cluster, they just aren't likely to do that.

 

[geegee83]

Didn't want to regress on the number of Big Cores on desktop. Also I saw stuff suggesting that it is possible to have partial cluster, they just aren't likely to do that.

Isn’t 8+4 the area of 9 cores? They wouldn’t have been able to sell 9 cores previously

 

[jpiniero]

Isn’t 8+4 the area of 9 cores? They wouldn’t have been able to sell 9 cores previously

The die physically is 8+8. If that's what you are asking.

 

[geegee83]

The die physically is 8+8. If that's what you are asking.

I mean the active die area is equal to 9 cores. They never had this kind of config before.

(Edit) Normally at least two cores are fused off.

 

[Hulk]

Appears to work that way on a per module level indeed. But that does seem like a waste, especially considering how AMD handles CCXes in comparison.

The desktop part is 8+8 and they are (supposedly) binning to 8+8, 8+4, and 4+4. From an outsider point of view (outside the industry) it looks like Gracemont defects are more problematic than Golden Cove. I guess the good news for many here that are not hyped on the Gracemont cores is that you will be able to purchase 8 Golden Coves without having to pay the top-of-the stack premium.

Also, seems like they should have an 6+4 desktop part as well, right? I mean they've gotta have a lot of dies with 6 or 7 working Coves I would think?

 

[jpiniero]

The i5 K is 6+4 and maybe some other i5 will be that too. There is also potentially for locked 8+0, 6+0, 4+0 and 2+0.

 

[dullard]

If these configurations are the real deal then do you think the defect rate for the Gracemont cores is high? I'm wondering why the +4's? Seems like gimping the P's to 6 and 4 makes more sense than cutting back Gracemont, from the 12900 I mean. Unless of course defects are high in the Gracemont cores?

I have no insider knowledge of the answer. But my main question is why Intel didn't go for 6+16 or 4+24 right away. They've got the core designs, it would be the same amount of physical space, it would be the same hybrid problems to solve. If the Gracemont cores are yielding a bit less, then that would be a good answer to both of our questions and it would be something that Intel could solve with a bit of time and experience.

 

[eek2121]

I have no insider knowledge of the answer. But my main question is why Intel didn't go for 6+16 or 4+24 right away. They've got the core designs, it would be the same amount of physical space, it would be the same hybrid problems to solve. If the Gracemont cores are yielding a bit less, then that would be a good answer to both of our questions and it would be something that Intel could solve with a bit of time and experience.

They didn’t have time. They are trying to release new architectures every month. (I kid, but if you look at their cadence recently it has been ridiculous).

 

[jpiniero]

I have no insider knowledge of the answer. But my main question is why Intel didn't go for 6+16 or 4+24 right away.

Page 517 - Discussion - Intel current and future Lakes & Rapids thread

Page 517 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

 

[TESKATLIPOKA]

I have no insider knowledge of the answer. But my main question is why Intel didn't go for 6+16 or 4+24 right away. They've got the core designs, it would be the same amount of physical space, it would be the same hybrid problems to solve. If the Gracemont cores are yielding a bit less, then that would be a good answer to both of our questions and it would be something that Intel could solve with a bit of time and experience.

Why would a much smaller core have yield problems when the big doesn't have?
I think the reason for 8+8 is gaming and maybe marketing. Let's be honest, not many people are that much in favour of this big+little concept, so I don't think they wanted to sacrifice too many big cores in favour of the smaller ones.
I think that's also a reason why the segmentation is mostly done by disabling Gracemont cores first.

 

[dullard]

Didn't want to regress on the number of Big Cores on desktop. Also I saw stuff suggesting that it is possible to have partial cluster, they just aren't likely to do that.

But, they are going to be selling 6+8, 2+8, etc. right from the start. I see the desire to go for the 8+8 chip. But, I could really see a 35W to 65 W mobile 4+16 chip going right into SFF computers and companies snatching them up. The entire company that I work at uses 4 core 35 W AMD chips. I could really use more cores for computations.

 

[TESKATLIPOKA]

But, they are going to be selling 6+8, 2+8, etc. right from the start. I see the desire to go for the 8+8 chip. But, I could really see a 35W to 65 W mobile 4+16 chip going right into SFF computers and companies snatching them up.

6+8 and 2+8 are for mobile. They also have bigger IGP.

 

[dullard]

Why would a much smaller core have yield problems when the big doesn't have?
I think the reason for 8+8 is gaming and maybe marketing. Let's be honest, not many people are that much in favour of this big+little concept, so I don't think they wanted to sacrifice too many big cores in favour of the smaller ones.
I think that's also a reason why the segmentation is mostly done by disabling Gracemont cores first.

I don't know the answer to your question. Are they really the same density? Is there something more difficult to produce about Gracemont than Golden Cove? I'm not in the chip production business, so I'm just curious.

I'd like to see your actual consumer data about people not being in favor of big+little. Or are you just going by forum posters who hate Intel, haven't actually seen any real data, and are quite vocal?

6+8 and 2+8 are for mobile. They also have bigger IGP.

Exactly. A mobile 4+16 would be exactly what fits well for me. Intel will be making and selling it, but I have to wait a generation or two.

 

[TESKATLIPOKA]

I don't know the answer to your question. Are they really the same density? Is there something more difficult to produce about Gracemont than Golden Cove? I'm not in the chip production business, so I'm just curious.

We won't find out their density unless Intel says how many transistors each core has.
I am not in the semiconductor business either, but I don't see what would be more difficult about producing Gracemont than Golden Cove, not to mention they are made at the same time, since they are a part of a single CPU.

I'd like to see your actual consumer data about people not being in favor of big+little. Or are you just going by forum posters who hate Intel, haven't actually seen any real data, and are quite vocal?

I don't have any consumer data to back up my claim(speculation), but most likely you also don't have any data saying the exact opposite.
If someone isn't in favour of big-little Alder Lake, he shouldn't be automatically labelled as an Intel hater, in my opinion.

Exactly. A mobile 4+16 would be exactly what fits well for me. Intel will be making and selling it, but I have to wait a generation or two.

Intel will release 2+8 for ULV, so maybe Raptor Lake will have such a model.

 

[LightningZ71]

It could be that they have no ability, currently, with Alder Lake specifically, to run a 'mont cluster with a member disabled. It could be a scheduler issue, or a cache management issue, whatever. So any sort of error in a mont cluster kills the whole cluster.

 

[Hulk]

I have no insider knowledge of the answer. But my main question is why Intel didn't go for 6+16 or 4+24 right away. They've got the core designs, it would be the same amount of physical space, it would be the same hybrid problems to solve. If the Gracemont cores are yielding a bit less, then that would be a good answer to both of our questions and it would be something that Intel could solve with a bit of time and experience.

I love the idea of 4+24, especially if Gracemont=Skylake proves to be true.

 

[TESKATLIPOKA]

I am quite content even with the current Alder Lake line-up, but I am more mobile oriented and would never need a 4+24 MT monster.
What is a bit sad is that Intel didn't increase the EU count to 128EU.

 

[LightningZ71]

I think it would make more sense to make a 6+16/128eu chip for mobile now that we're entering the DDR5 era. I realize that that's a "bigger" chip than the existing largest chip, but, after a node shrink, it should be manageable.

 

[Hulk]

I am quite content even with the current Alder Lake line-up, but I am more mobile oriented and would never need a 4+24 MT monster.
What is a bit sad is that Intel didn't increase the EU count to 128EU.

Looks like we're still stuck at 32EU's for desktop too.

I'm not sure if I would need 4+24 either, but it sounds good! My main compute heavy apps are video editing and multitrack audio mixing. Much as I wish they were super multithreaded, I have a feeling I'm going to get the most benefit from the upgraded Cove cores.