Discussion Intel current and future Lakes & Rapids thread

[ondma]

Yes, I did. Funny you should call it a "delay from years ago" when it was about their Intel 4 process.

Anyway, my point was related to Intel's consumer desktop roadmap and their use of Intel 4 in their mix. If the node is in such good shape as you say, why are we getting a Raptor Lake refresh instead of a proper tick?

Obviously, there could be other reasons for delay than problems with the process. ML is a totally new (for Intel), product, so there could be problems with communication between the chiplets.

 

[nicalandia]

The question is how will they "refresh" RL. It is already pushed to the max in frequency and power consumption.

You would have think that the 12900KS was pushed beyond sense... And then we got the 13900K, So I fully expect Raptor Lake Refresh will have better binned dies that will allow at least 100 mhz bump and also DLVR. Also the non-K version might be bumped to real raptor cove cores

 

[Exist50]

Perhaps MTL is a Mobile Only CPU?

Well there were the rumors that MTL-S and ARL-S are landing around the same time. This would lend credence to that. If the mobile chips only barely make it by the end of '23, it would make sense for the desktop chips with a different SoC (and compute die?) to be pushed into '24.

Or they split the lineup. RPL-S for high end/gaming, and MTL-S for OEM.

 

[coercitiv]

Perhaps MTL is a Mobile Only CPU?

ML is a totally new (for Intel), product, so there could be problems with communication between the chiplets.

MTL is hopefully proof that Intel 4 is not vaporware, and I hope it comes out in better relative shape than Icelake, because it's an important stepping stone for Intel.

However, I purposely left out MTL from the desktop discussion because Raptor Lake refresh by itself signals a problem, a hole in their roadmap. That hole would be better filled with a proper tick, even if that means a Raptor Cove shrink. Meteor Lake on the desktop might as well be planned for the next desktop cycle, to be used as the cheaper die for value parts.

 

[jpiniero]

Or they split the lineup. RPL-S for high end/gaming, and MTL-S for OEM.

If you look at the slide, it shows the entire desktop lineup is getting the Raptor Lake Refresh. So no. But the timing suggests that whatever they are going to do afterwards would be released in Q3 2024.

 

[Exist50]

If you look at the slide, it shows the entire desktop lineup is getting the Raptor Lake Refresh. So no. But the timing suggests that whatever they are going to do afterwards would be released in Q3 2024.

They'd done weird, non-yearly steps before, notably with Kaby Lake->Coffee Lake and Rocket Lake->Alder Lake. But agreed it probably makes more sense to think of MTL-S as an ARL contemporary.

 

[Exist50]

On the topic, has this been posted yet?

https://www.tomshardware.com/news/intel-to-disclose-client-pc-roadmap-next-week

Seems like we'll get some answers soon enough.

 

[nicalandia]

On the topic, has this been posted yet?

https://www.tomshardware.com/news/intel-to-disclose-client-pc-roadmap-next-week

Seems like we'll get some answers soon enough.

I think that's were the leaked slides are from. We are just missing the slides for Mobile

 

[DrMrLordX]

TSMC 3nm and Intel 4 products could co-exist to meet volume. Depends on whether Intel has enough TSMC 3nm wafer allocation.

If Arrow Lake is still on-track for 2024, why bother releasing Meteor Lake-S at all? It should be an inferior product vs. Arrow Lake.

 

[ashFTW]

Well there were the rumors that MTL-S and ARL-S are landing around the same time. This would lend credence to that. If the mobile chips only barely make it by the end of '23, it would make sense for the desktop chips with a different SoC (and compute die?) to be pushed into '24.

Or they split the lineup. RPL-S for high end/gaming, and MTL-S for OEM.

If you pay attention to the colors in the figure below, ARL will have the same SOC and IOE tiles but different CPU and GPU tiles presumably on Intel 20A and some External process respectively.

 

[Exist50]

If you pay attention to the colors in the figure below, ARL will have the same SOC and IOE tiles but different CPU and GPU tiles presumably on Intel 20A and some External process respectively.

I was speaking entirely of Meteor Lake. They probably can't reuse the mobile SoC die for desktop, even within the Meteor Lake family, so that's another die/product/platform to develop. And if they're actually making a 6+16 desktop compute tile, that too.

 

[ashFTW]

I was speaking entirely of Meteor Lake. They probably can't reuse the mobile SoC die for desktop, even within the Meteor Lake family, so that's another die/product/platform to develop. And if they're actually making a 6+16 desktop compute tile, that too.

I expect all (10W to 100W) of MTL and ARL to use the same SOC tile; it’s the “central hub“ in the disaggregated design. The extra IO for desktop will be on a larger IOE tile. But the design does allow for variable SOC tiles.

 

[nicalandia]

Intel 13900KS

ASUS System Product Name - Geekbench

Benchmark results for an ASUS System Product Name with an Intel Core i9-13900KS processor.

browser.geekbench.com

 

[witeken]

In case the folks around here need a reminder of what Intel's plans looked like just 1 year ago, here's their roadmap as described at Intel's Accelerated event in the summer of 2021. Intel 4 was supposed to ramp up in 2022 and deliver products in H1 2023. And that's the updated timeline, after they had already announced a ~12 month delay at the end 2020.

View attachment 72529

It feels like 2017 all over again, only this time at least they have better contingency plans both in terms of design and manufacturing. Looking forward to Intel 7+ in 2023.

This AnandTech slide is an utter misrepresentation of what Intel actually announced in mid-2021. TLDR: Intel's roadmap has remained the same sind mid-2020, there have been 0 delays and 1 pull-in.

*The HVM window for all nodes has always been 6 months. For Intel 4 it was H2'22... And Intel just announced that it has started manufacturing on Intel 4, so it is on schedule.
*Intel is known for its annual cadence. Meteor Lake in H2'23 is consistent with Raptor Lake in H2'22 (even if Bob Swan did initially say early 2023. Pat Gelsinger has never said this).
*20A has always been H1'24 (in time for Arrow Lake launch in H2'24), so the AnandTech timeline is wrong.
*18A was initially early 2025 (not mid-2025 like the AnandTech slide), but Intel has since pulled this in to H2'24.

 

[jpiniero]

This AnandTech slide is an utter misrepresentation of what Intel actually announced in mid-2021. TLDR: Intel's roadmap has remained the same sind mid-2020, there have been 0 delays and 1 pull-in.

*The HVM window for all nodes has always been 6 months. For Intel 4 it was H2'22... And Intel just announced that it has started manufacturing on Intel 4, so it is on schedule.
*Intel is known for its annual cadence. Meteor Lake in H2'23 is consistent with Raptor Lake in H2'22 (even if Bob Swan did initially say early 2023. Pat Gelsinger has never said this).
*20A has always been H1'24 (in time for Arrow Lake launch in H2'24), so the AnandTech timeline is wrong.
*18A was initially early 2025 (not mid-2025 like the AnandTech slide), but Intel has since pulled this in to H2'24.

In case you haven't been paying attention, Meteor Lake is looking like a 2024 product in reality (although maybe they will launch something paperish at the end of 2023) and presumably Arrow Lake would be a year later... and that's also in reality fabbed at TSMC.

 

[DrMrLordX]

TLDR: Intel's roadmap has remained the same sind mid-2020, there have been 0 delays and 1 pull-in.

LOL doesn't make it any better that Meteor Lake won't show up in any form until H2 2023, possibly late Q4.

 

[eek2121]

We’ve seen no indications Intel is porting it’s core design for desktop or mobile to TSMC. I mentioned this i. the other thread, but this is what I strongly suspect we will see:

2023 Desktop: Raptor Lake Refresh
2023 Mobile (possibly q1 2024): Mostly Raptor Lake with some Meteor Lake parts for premium “thin and light” laptops. Possibly a NUC release later on

2024 Desktop: Arrow Lake, Maybe some Meteor Lake, and possibly some Raptor Lake covering high volume SKUs.

2024 Laptop: Meteor Lake/Arrow Lake.

Things will be similar to what Intel did with Raptor Lake and Alder Lake for 13th gen with some SKUs.

EDIT: Also note that the competition is doing something similar on mobile as well.

Expect multiple designs on multiple nodes for a single generation to be a strategy used widely in the future as better processes cost significantly more and may have lower volume.

 

[nicalandia]

Current 13900KS numbers.

Geekbench Search - Geekbench

 

[Joe NYC]

Did you see the updated version that came out this week? Not much change, but 18A was moved forward from Q3 2025 to H2 2024. Yet, you needed to highlight a delay from years ago in your post?
View attachment 72540

Q3 2021 was when Alder Lake (on Intel 7) was released, with processors widely available in Q4 2021

Regarding Intel 4, it says it is ready Q3 2022 / Today, but no product is anywhere close to wide availability, so we have to take Intel's word for it, that it is ready "today" while something else is holding up Meteor Lake release.

 

[nicalandia]

Intel N95. All e core CPU

Default string Default string - Geekbench

Benchmark results for a Default string Default string with an Intel N95 processor.

browser.geekbench.com

 

[clemsyn]

Q3 2021 was when Alder Lake (on Intel 7) was released, with processors widely available in Q4 2021

Regarding Intel 4, it says it is ready Q3 2022 / Today, but no product is anywhere close to wide availability, so we have to take Intel's word for it, that it is ready "today" while something else is holding up Meteor Lake release.

That 7nm (Intel 4) must be a big headache for Intel. Would be very surprised it they hit the target with Intel 3, 20A and 18A.

 

[nicalandia]

This is the first Apples to Apple comparison of Intel Tremont Cores vs Gracemont Cores 4C/4T vs 4C/4T

Default string Default string vs WTM W-N95 - Geekbench

 

[Exist50]

This is the first Apples to Apple comparison of Intel Tremont Cores vs Gracemont Cores 4C/4T vs 4C/4T

View attachment 72890

WTM W-N95 vs Default string Default string - Geekbench

Something's clearly off with this comparison. Gracemont should have a much larger margin over Tremont. Maybe wildly different power limits?

 

[nicalandia]

Something's clearly off with this comparison. Gracemont should have a much larger margin over Tremont. Maybe wildly different power limits?

Both are 15 Watt SKUs,

I see a 25% boost in ST Performance at about the same speed(2.8 Ghz) and about 2% faster on MT despite a 18% speed deficiency

Perhaps this is a better graph

Default string Default string - Geekbench

Benchmark results for a Default string Default string with an Intel N95 processor.

browser.geekbench.com

So all in all, it's a great IPC Jump of about 20-25% at ISO TDP

 

[Exist50]

I see a 25% boost in ST Performance at about the same speed(2.8 Ghz) and about 2% faster on MT despite a 18% speed deficiency

We know from ADL silicon that Gracemont clocks way higher than Tremont, in addition to the IPC boost. It's completely nonsensical for the two to perform similarly in MT at iso-power, and some previous results support this conclusion. So either the result is just garbage, or they're not being run at the same power limits.

Where are you seeing that they're both running at 15W?