Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads

[Tigerick]

As Hot Chips 34 starting this week, Intel will unveil technical information of upcoming Meteor Lake (MTL) and Arrow Lake (ARL), new generation platform after Raptor Lake. Both MTL and ARL represent new direction which Intel will move to multiple chiplets and combine as one SoC platform.

MTL also represents new compute tile that based on Intel 4 process which is based on EUV lithography, a first from Intel. Intel expects to ship MTL mobile SoC in 2023.

ARL will come after MTL so Intel should be shipping it in 2024, that is what Intel roadmap is telling us. ARL compute tile will be manufactured by Intel 20A process, a first from Intel to use GAA transistors called RibbonFET.





Intel Core Ultra 100 - Meteor Lake



As mentioned by Tomshardware, TSMC will manufacture the I/O, SoC, and GPU tiles. That means Intel will manufacture only the CPU and Foveros tiles. (Notably, Intel calls the I/O tile an 'I/O Expander,' hence the IOE moniker.)

 

[SiliconFly]

Both 680M and 780M Radeon iGPUs are much faster than 50% of MTLs supposed performance.

780M, for example scores 2800 pts in 3D Mark Time Spy Graphics. MTL-P 128 EU iGPU is supposed to score here around 3500 pts. 680 in TS Graphics got 2400 pts, average.

You are vastly underestimating the performance of iGPUs these days.

Not really. 780M is a very slow GPU. Runs at less than half the performance of a RTX 4050 mobile, generally speaking. Also, it doesn't include ray-tracing.

On the other hand, MTL's GPU is expected to be slightly better than a 780M in raw performance. And it also has ray-tracing.

Thats a huge difference.

 

[AMDK11]

Moving the new core to a lower process brought really minor architectural changes, and they were even cosmetic in terms of logic complexity.Broadwell introduces 5% higher IPC, IvyBridge 3% according to Intel, RaptorCove +5%.

Historically, RaptorCove should have been in Intel4, but it stayed in Intel7, which is why Intel believes Raptor should not be there (in this form).The codenames themselves don't really guarantee whether the solutions under the codename now are the same as they were supposed to be earlier.Anyway, a demo and more details about RedwoodCove will be coming soon so you can verify the changes in x86.

 

[Glo.]

Not really. 780M is a very slow GPU. Runs at less than half the speed of a RTX 4050 mobile, generally speaking. Also, it doesn't include ray-tracing.

On the other hand, MTL's GPU is expected to be slightly better than a 780M in raw performance. And it also has ray-tracing.

Thats a huge difference.

What do you mean "not really"?

I've told you exactly how it performs. You use now mental gymnastics to justify your argument.

And yes, 780M DOES have Ray Tracing.

 

[AMDK11]

Being more careful with your forecasts would mean following the general sentiment of the leakers, who all claim ARL has LNC. Not a single one think ARL uses RWC+. It's fine to have less popular and more speculative forecasts, but one should advertise it as such.
Your assumption that LNL is coming 2025 on Intel 18A is wrong. 18A vs N3 is all rumors (it's n3) but Intel themselves confirmed LNL will launch in 2024 numerous times this year.
I'm pretty sure the "ARL has RWC+" rumors started since ARL performance projections got leaked. Guys, LNC might just be a dud of an architecture. It happens. SNC wasn't all that great either, remember? I mean I hope those perf projections aren't true for the sake of competition in the market, but no leaker seems to be disputing it either...

x86 SunnyCove failed? What? SunnyCove is a redesigned and expanded core that has 38% more transistors than SkyLake. Despite still having 4-way decoding and 4 ALUs, it has an average of 18% higher IPC. Zen 3 with 19% IPC increase is a huge profit, and SunnyCove with 18% higher IPC is already a failed architecture? Really?

 

[SiliconFly]

What do you mean "not really"?

I've told you exactly how it performs. You use now mental gymnastics to justify your argument.

And yes, 780M DOES have Ray Tracing.

You're right. MTL's tGPU is expected to be a lot faster than 780M in raw performance. And it has ray-tracing.

Notebookcheck:
(In theory) "780M supports hardware raytracing, although the performance is too slow to use it in current games."

Even, even an old 80386 can do ray-tracing. In theory. Just sayin'

 

[SiliconFly]

x86 SunnyCove failed? What? SunnyCove is a redesigned and expanded core that has 38% more transistors than SkyLake. Despite still having 4-way decoding and 4 ALUs, it has an average of 18% higher IPC. Zen 3 with 19% IPC increase is a huge profit, and SunnyCove with 18% higher IPC is already a failed architecture? Really?

Actually, Sunny Cove was a good product, but had yield issues (on the original 10nm node). And it got overshadowed by Comet Lake (14nm++++).

 

[Geddagod]

x86 SunnyCove failed? What? SunnyCove is a redesigned and expanded core that has 38% more transistors than SkyLake. Despite still having 4-way decoding and 4 ALUs, it has an average of 18% higher IPC. Zen 3 with 19% IPC increase is a huge profit, and SunnyCove with 18% higher IPC is already a failed architecture? Really?

Lil bro I said it "wasn't all that good". And where did you get that transistor count number?
And ye SNC wasn't that good lol. Mid core. SNC didn't improve perf/watt vs CML mobile, had a ST freq regression, and increased area by like 25%. Zen 3 increased perf/watt by like 15-20%, increased ST freq, increased freq/power across the curve , and increased area by 10-15%. So ye, Zen 3 is a "huge profit" and SNC is pretty bad tbh.
And before we blame 10nm for SNC too much, we should remember than Cypress Cove fixed the ST freq regression, but still consumed more power iso frequency vs comet lake, and by a good chunk iirc. Zen 3 is incredibly impressive with the fact that it didn't. And oh ye, CYPC is a whopping 33% larger per core vs SKL.

Actually, Sunny Cove was a good product, but had yield issues (on the original 10nm node). And it got overshadowed by Comet Lake (14nm++++).

SNC shouldn't be considered a good product if it got overshadowed by a product on the previous node.

 

[AMDK11]

Actually, Sunny Cove was a good product, but had yield issues (on the original 10nm node). And it got overshadowed by Comet Lake (14nm++++).

10 core (x86 Skylake) CometLake overshadowed RocketLake but only because the latter had only 8 CypressCove cores.I'm not surprised at all by Intel because CypressCove is Sunny in 14nm and the area of 8 cores takes up 11 Skylake cores.After all, CypressCove is still more advanced cores with a much higher IPC.

 

[AMDK11]

Lil bro I said it "wasn't all that good". And where did you get that transistor count number?
And ye SNC wasn't that good lol. Mid core. SNC didn't improve perf/watt vs CML mobile, had a ST freq regression, and increased area by like 25%. Zen 3 increased perf/watt by like 15-20%, increased ST freq, increased freq/power across the curve , and increased area by 10-15%. So ye, Zen 3 is a "huge profit" and SNC is pretty bad tbh.
And before we blame 10nm for SNC too much, we should remember than Cypress Cove fixed the ST freq regression, but still consumed more power iso frequency vs comet lake, and by a good chunk iirc. Zen 3 is incredibly impressive with the fact that it didn't. And oh ye, CYPC is a whopping 33% larger per core vs SKL.

SNC shouldn't be considered a good product if it got overshadowed by a product on the previous node.

What does this have to do with IPC?You are confusing the final product with the x86 core microarchitecture.

Jim Keller revealed the numbers: x86 Skylake 217 million transistors and SunnyCove 300 million transistors or +38%.Too bad there is no data for GoldenCove.

https://pbs.twimg.com/media/E2GW2NWXIAUbdBK?format=jpg&name=large

 

[AMDK11]

Lil bro I said it "wasn't all that good". And where did you get that transistor count number?
And ye SNC wasn't that good lol. Mid core. SNC didn't improve perf/watt vs CML mobile, had a ST freq regression, and increased area by like 25%. Zen 3 increased perf/watt by like 15-20%, increased ST freq, increased freq/power across the curve , and increased area by 10-15%. So ye, Zen 3 is a "huge profit" and SNC is pretty bad tbh.
And before we blame 10nm for SNC too much, we should remember than Cypress Cove fixed the ST freq regression, but still consumed more power iso frequency vs comet lake, and by a good chunk iirc. Zen 3 is incredibly impressive with the fact that it didn't. And oh ye, CYPC is a whopping 33% larger per core vs SKL.

SNC shouldn't be considered a good product if it got overshadowed by a product on the previous node.

What does this have to do with IPC?You are confusing the final product with the x86 core microarchitecture.Jim Keller revealed the numbers: x86 Skylake 217 million transistors and SunnyCove 300 million transistors or +38%.Too bad there is no data for GoldenCove.

 

[SiliconFly]

Lil bro I said it "wasn't all that good". And where did you get that transistor count number?
And ye SNC wasn't that good lol. Mid core. SNC didn't improve perf/watt vs CML mobile, had a ST freq regression, and increased area by like 25%. Zen 3 increased perf/watt by like 15-20%, increased ST freq, increased freq/power across the curve , and increased area by 10-15%. So ye, Zen 3 is a "huge profit" and SNC is pretty bad tbh.
And before we blame 10nm for SNC too much, we should remember than Cypress Cove fixed the ST freq regression, but still consumed more power iso frequency vs comet lake, and by a good chunk iirc. Zen 3 is incredibly impressive with the fact that it didn't. And oh ye, CYPC is a whopping 33% larger per core vs SKL.

SNC shouldn't be considered a good product if it got overshadowed by a product on the previous node.

We shouldn't compare clock frequencies of Sunny Cove with Cypress Cove. The former was on 10nm+ node while the latter was on a highly refined 14nm+++++ node.

 

[Geddagod]

What does this have to do with IPC?You are confusing the final product with the x86 core microarchitecture.

You do realize stuff like perf/watt, ST peak freq, and area are all heavily influenced by architecture as well, right?
A high IPC increase doesn't help perf/watt much if power consumption iso frequency also increases a lot as well.
I mean, the only thing that 10nm could be blamed for with SNC was the peak ST freq. SNC on 14nm showed that area increase and frequency iso power were still terrible compared to SKL. SNC, as an architecture, was not that good.
Just comparing IPC numbers is stupid.

What does this have to do with IPC?You are confusing the final product with the x86 core microarchitecture.Jim Keller revealed the numbers: x86 Skylake 217 million transistors and SunnyCove 300 million transistors or +38%.Too bad there is no data for GoldenCove.

Can you provide a link or something? I have not seen that.
Also that picture you provided shows die area, not transistor count. It shows +37%, which is pretty close to the +33% larger cypress cove vs skl area I was talking about before (i used a different source). But ye, that level of area increase, for the IPC increase it provided, is laughably bad in comparison of Zen 3's improvement vs the area increase it has.
Also do we know how many more transistors Zen 3 uses vs Zen 2? You know, so we can have a comparison point?

We shouldn't compare clock frequencies Sunny Cove with Cypress Cove. The former was on 10nm+ node while the latter was on a highly refined 14nm+++++ node.

Ye, I conceded that in my paragraph comparing cypress cove and skylake. It hurt ICL as a product, but it would be a bit unfair to blame it on the arch when we have seen CYPC clock as high as SKL.

 

[AMDK11]

You do realize stuff like perf/watt, ST peak freq, and area are all heavily influenced by architecture as well, right?
A high IPC increase doesn't help perf/watt much if power consumption iso frequency also increases a lot as well.
I mean, the only thing that 10nm could be blamed for with SNC was the peak ST freq. SNC on 14nm showed that area increase and frequency iso power were still terrible compared to SKL. SNC, as an architecture, was not that good.
Just comparing IPC numbers is stupid.

Can you provide a link or something? I have not seen that.
Also that picture you provided shows die area, not transistor count. It shows +37%, which is pretty close to the +33% larger cypress cove vs skl area I was talking about before (i used a different source). But ye, that level of area increase, for the IPC increase it provided, is laughably bad in comparison of Zen 3's improvement vs the area increase it has.
Also do we know how many more transistors Zen 3 uses vs Zen 2? You know, so we can have a comparison point?

Ye, I conceded that in my paragraph comparing cypress cove and skylake. It hurt ICL as a product, but it would be a bit unfair to blame it on the arch when we have seen CYPC clock as high as SKL.

Jim Keller, when he was still working at Intel, talked about future generations of very large cores.Then he talked about the fact that in the future there would be a huge x86 core with 8-way decoding and ROB ~800.He provided data on Skylake whose x86 core consists of 217 million transistors and the next one after it, i.e. SunnyCove, is logically larger by 38%.217 + 38% = 300 million transistors.An average of 18% higher IPC with a 38% more complex core is a good increase and rather expected.Unfortunately, in the case of CypressCove, it was not balanced by the lower technological process.

Rocketlake has compromises made by 14nm, including the RAM controller, which does not ensure full use of SunnyCove's capabilities.As far as I know, CypressCove is an exact copy of SunnyCove ported to 14nm with optimizations for path length.Regardless, I would still choose Rocketlake instead of CometLake due to its more advanced x86 core.

 

[Geddagod]

Then he talked about the fact that in the future there would be a huge x86 core with 8-way decoding and ROB ~800.

Can you send me a link of that? Did he provide those numbers? Just curious

He provided data on Skylake whose x86 core consists of 217 million transistors and the next one after it, i.e. SunnyCove, is logically larger by 38%.217 + 38% = 300 million transistors.

That's not how it works lol

An average of 18% higher IPC with a 38% more complex core is a good increase and rather expected.

Do we have any data of previous architectures to support that claim?
Compared to Zen 3, which only increased area by 10-15%, that level of area increase for the same amount of IPC increase looks terrible.

 

[FangBLade]

Really? Tell me one existing Ryzen mobile chip that has a GPU that's even half as good as MTL's.

You show me the results of the MTL graphics card testing first, and then I'll respond with the Phoenix.

 

[FangBLade]

Not really. 780M is a very slow GPU. Runs at less than half the performance of a RTX 4050 mobile, generally speaking. Also, it doesn't include ray-tracing.

On the other hand, MTL's GPU is expected to be slightly better than a 780M in raw performance. And it also has ray-tracing.

Thats a huge difference.

As far as I know, the 780m has ray tracing. Secondly, when you mention ray tracing, you'll surely be satisfied with integrated graphics, especially with 6P cores. Someone earlier wrote that 95% of people who buy a laptop without an external graphics card don't buy it for gaming, which is logical. So, it's pointless to talk about ray tracing. If they buy one with an external graphics card, they will use that graphics card, not the integrated one. There's a lot of contradiction in this topic. The more comments you write, the less serious you seem to me. Take a break.

 

[Dayman1225]

Well, the jury is still out in this matter. It may have RWC+ (likely) or the 1st iteration of LNC (more likely). But most "believable" rumors suggest more in favor of LNC rather than RWC+

AFAIK ARL is Lion cove and Skymont, with some RWC+CMT in the bottom of the stack on Intel3(maybe only mobile) Not 100% on that though, so don’t quote me.

 

[AMDK11]

Can you send me a link of that? Did he provide those numbers? Just curious

That's not how it works lol

Do we have any data of previous architectures to support that claim?
Compared to Zen 3, which only increased area by 10-15%, that level of area increase for the same amount of IPC increase looks terrible.

I can't find it now, but when I do, I'll post the link.I am 100% sure about these numbers because I was very intrigued by them at the time and based on his words I had quite a heated discussion on another forum.Such numbers always remain in memory because they have been rarely made available for a long time.It's possible that he quoted 300 million transistors for SunnyCove and said it was 38% more than Skylake.As I've already written, I'm more confident in these numbers than anything else.

As for Zen3, it is a new design almost from scratch in terms of 4-way decoding and 4 ALU.A big feature in the development of IPC is also the transition from two separate parts of L3 to shared between 8 cores. According to AMD IPC growth curve from +9% to even 39% depending on the type of data with an average of +19%. According to Anandtech, from 1-2% to even over 100%.

IPC SunnyCove vs Skylake:

 

[Geddagod]

It's possible that he quoted 300 million transistors for SunnyCove and said it was 38% more than Skylake

I'm calling cap on all that lol. I'll remain skeptical until a link is posted.

As for Zen3, it is a new design almost from scratch in terms of 4-way decoding and 4 ALU.

Ok?

A big feature in the development of IPC is also the transition from two separate parts of L3 to shared between 8 cores.

Eh. Look at the Zen 3 IPC breakdown.

IPC growth curve from +2% to even 40-60% depending on the type of data with an average of +19%.

I mean that's true for like every architecture.

Literally that entire third edit: second paragraph was nontent. Sunny Cove was a mid architecture, too large and power hungry for what it was. Zen 3 was just much better, both as an improvement over both companies previous architectures, but also in comparison with eachother.
I have high hopes for LNC still, despite the perf projections, but I really hope Intel's core architecture team can get it together for a win. They haven't had one in a while.

 

[AMDK11]

I'm calling cap on all that lol. I'll remain skeptical until a link is posted.

Ok?

Eh. Look at the Zen 3 IPC breakdown.

I mean that's true for like every architecture.

Literally that entire third edit: second paragraph was nontent. Sunny Cove was a mid architecture, too large and power hungry for what it was. Zen 3 was just much better, both as an improvement over both companies previous architectures, but also in comparison with eachother.
I have high hopes for LNC still, despite the perf projections, but I really hope Intel's core architecture team can get it together for a win. They haven't had one in a while.

SunnyCove is a very successful microarchitecture with significant IPC increases.Personally, I would choose Rocketlake over Cometlake, despite the lower core count and disproportionately higher power consumption.Unfortunately, SunnyCove was released too late in the obsolete 14nm when it should have been intended for at least 10nmSF.

 

[Geddagod]

SunnyCove is a very successful microarchitecture

Ok you keep on saying this, but I literally listed out all the reasons it's bad. Compared to what Intel's own last gen architecture provides in perf/watt it's bad, it's bad compared to the competition, it's just not good. If by successful you mean, "hey it works and sold products", I mean, I guess that's a really low bar, but looking at Palm Cove, maybe Intel needs that low of a bar?

with significant IPC increases

Intel brute forced it. Threw transistors at the problem lol.

Personally, I would choose Rocketlake over Cometlake, despite the lower core count and disproportionately higher power consumption.

Most people wouldn't. The subpar cache subsystem and weird memory stuff esentially castrates the higher PPC potential of the core in gaming.

Unfortunately, SunnyCove was released too late in the obsolete 14nm when it should have been intended for at least 10nmSF.

Willow Cove wasn't too bad tbh. At least it caught up in performance. Whatever DTCO magic it took Intel, it was a really good improvement over SNC using what is esentially the same core microarchitecture. Still a worse Zen 3 tho lol.

 

[AMDK11]

Ok you keep on saying this, but I literally listed out all the reasons it's bad. Compared to what Intel's own last gen architecture provides in perf/watt it's bad, it's bad compared to the competition, it's just not good. If by successful you mean, "hey it works and sold products", I mean, I guess that's a really low bar, but looking at Palm Cove, maybe Intel needs that low of a bar?

Intel brute forced it. Threw transistors at the problem lol.

Most people wouldn't. The subpar cache subsystem and weird memory stuff esentially castrates the higher PPC potential of the core in gaming.

Willow Cove wasn't too bad tbh. At least it caught up in performance. Whatever DTCO magic it took Intel, it was a really good improvement over SNC using what is esentially the same core microarchitecture. Still a worse Zen 3 tho lol.

Yes, RocketLake as a final product was below expectations.What I care about all the time is the microarchitect which had potential, but this potential was not used mainly due to problems with the production process.Despite all this, CypressCove is a new core and not the 5-year-old Skylake.

WillowCove even recorded a drop in IPC due to a different L2-L3 memory subsystem, but it achieved higher clock speeds than SunnyCove in Icelake.

 

[Geddagod]

What I care about all the time is the microarchitect which had potential, but this potential was not used mainly due to problems with the production process

No. The area increase between RKL and CML was not due to a production process issue. That's an iso node comparison, the area increase was solely due to how many extra transistors the larger CYPC (backported SNC) architecture took.
You see the same thing between PLMC and SNC. SNC there is like 25% larger than PLMC.
Zen 3 iso process was only 10-15% larger. AMD is getting the same IPC increase for less area.
The power increase, can also be compared iso process with RKL and CML. RKL consumes way more power for the same frequency, across the curve. It's less efficient than comet lake in some instances even at the same core count. It's that ridiculous.
Even if we assume 10nm+ has no energy efficiency gains over 14nm+++, it's still pretty sad to see ICL products have ~the same energy efficiency as CML does. And I'm not talking about 125 watt desktop chips where ICL might have issues scaling due to the node, no, I'm talking about when it's at the 15 watt mobile range.
Even accounting for the process node, Sunny Cove is just a bad architecture. There's really nothing more to it. I've said the same stuff numerous times now. I'm accounting for the process node. That's the entire reason why I brought up RKL as well, it's because you can compare the two architectures on the same 14nm process.
Sunny Cove is just a worse Zen 3. Worse in performance, worse in power, worse in area. Willow Cove on Intel 10SF shows what Sunny Cove would look like on a better node. And guess what? It caches up in performance- still takes the L in power and area. It took all the way to Golden Cove for Intel to catch up to Zen 3 in energy efficiency.
I really do hope Intel revives their core architecture team, because they have not had a really good core architecture in a while. LNC looks like a dud too based on perf projections, but I really do still have hope for the area and power side of things.
Edit: just a bonus piece of speculation, I think RWC should at the very least, be competitive in power with Zen 4. Ties in performance it looks like, but is a bigger core. Prob one of the best cores Intel has released versus AMD in a long time. Too bad Zen 5 drops only like half a year after RWC does.

 

[AMDK11]

Of course, RocketLake is a problem with the technological process because it should never have been released at 14nm.SunnyCove with 38% more core complexity should be on 10nm and then the core would occupy a smaller area than the Skylake cores in CometLake.Then Intel could easily give 10-12 cores, and SunnyCove in 14nm was large, which forced Intel to cut 2 cores.

If Intel didn't have problems with the process, PalmCove would have been released first and then Sunny in the form of Icelake.By the way, PalmCove does not contain anything particularly interesting compared to Skylake and the IPC increase may be a few%.

 

[Henry swagger]

x86 SunnyCove failed? What? SunnyCove is a redesigned and expanded core that has 38% more transistors than SkyLake. Despite still having 4-way decoding and 4 ALUs, it has an average of 18% higher IPC. Zen 3 with 19% IPC increase is a huge profit, and SunnyCove with 18% higher IPC is already a failed architecture? Really?

Lol.. comments are crazy.. sunnycove has same ipc as zen 3 on a crap 14nm node. Imagine if its on intel 7