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.)

 

[igor_kavinski]

They'd have been better off doing a large monolithic die on Intel 3 with updated P and E cores. But by how much we'll never know.

They probably had to rule out a large die on Intel 3 because their testing showed that the yields would be abysmal. Not worth it for the prices they have to sell consumer chips at. The crazy thing about Pat's 5N4Y plan was that they barely got any time to refine the nodes to increase the yields. It was just burning massive amounts of money to hit goals for better optics and best yield outcome wasn't one of those goals.

 

[511]

They probably had to rule out a large die on Intel 3 because their testing showed that the yields would be abysmal

GNR has 600mm2 Intel 3 dies lol and Intel 3 Yield is pretty good it already matched Intel 7 yields last year

 

[igor_kavinski]

GNR has 600mm2 Intel 3 dies lol and Intel 3 Yield is pretty good it already matched Intel 7 yields last year

And your source is Intel Public Lies

 

[Win2012R2]

GNR has 600mm2 Intel 3 dies lol and Intel 3 Yield is pretty good it already matched Intel 7 yields last year

How do you know if yield is pretty good or not? Gross margins are way down in DC, and AMD is not exactly selling cheap server chips.

 

[511]

And your source is Intel Public Lies

for the 600mm2 it's 598mm2 off by 2mm2

https://twitter.com/x/status/1699962079774462309

for the 2nd part it was leaked some time ago

https://twitter.com/x/status/1798678499747836002

 

[511]

How do you know if yield is pretty good or not? Gross margins are way down in DC, and AMD is not exactly selling cheap server chips.

that's due to selling SPR/EMR Cheap to maintain market share and GNR/SRF is in ramp mode

 

[Win2012R2]

that's due to selling SPR/EMR Cheap to maintain market share and GNR/SRF is in ramp mode

How do you know they are sold cheap vs yield issues make gross margins tank?

I certainly don't observe cheap Turins or Xeons - in retail and OEM channels (such as SuperMicro).

 

[511]

How do you know they are sold cheap vs yield issues make gross margins tank?

recent report also Intel 7 is a highly mature node there are no yield issues on Intel 7 and Intel 3 otherwise we won't be having both high clocking/large dies products

I certainly don't observe cheap Turins or Xeons - in retail and OEM channels (such as SuperMicro).

https://www.newegg.com/intel-xeon-6980p/p/N82E16819118532

 

[igor_kavinski]

https://www.newegg.com/intel-xeon-6980p/p/N82E16819118532

https://www.newegg.com/amd-epyc-9555-socket-sp5/p/N82E16819113857

Turin is cheaper.

Even 96 core is cheaper that beats the Xeon's ass: https://www.newegg.com/amd-epyc-9645-socket-sp5/p/N82E16819113855

 

[511]

https://www.newegg.com/amd-epyc-9555-socket-sp5/p/N82E16819113857

Turin is cheaper.

Even 96 core is cheaper that beats the Xeon's ass: https://www.newegg.com/amd-epyc-9645-socket-sp5/p/N82E16819113855

no? in what multi core benchmark i would like to see

 

[OneEng2]

Intel 3 is okay, but for some reason they skipped over it for desktop CPUs

I am wondering if it isn't price as mentioned above?

Why aren't they frickin' backporting ARL to Intel 7 and running it at 6+ GHz? Sure, it will run a lot hotter but plenty of folks don't mind the heat generated by 14900K

A lot of folks weren't too happy with returning the 14900K when it burnt up though.

How do you know if yield is pretty good or not? Gross margins are way down in DC, and AMD is not exactly selling cheap server chips.

I think this is likely the problem.

I am guessing that Intel has been creating designs that rely on incredible feats of engineering from the foundry. This worked really well for decades .... until it didn't.

It is my guess that Intel must change their entire philosophy from design up.

For Intel, they need to start with the available foundry services and price point and then design around that IMO.

I have been saying for years that it is easy for Intel engineers to best AMD when they have such a big process advantage to work from.

Turns out that once this advantage went away, so did the performance advantage.

Even today, ARL can't best Zen 5 from a full process advantage (N3B vs N4P). Imagine how badly ARL would be pummeled by an N3E Zen5 desktop processor?

Hopefully this is the kind of things being discussed at Intel.

 

[511]

Even today, ARL can't best Zen 5 from a full process advantage (N3B vs N4P). Imagine how badly ARL would be pummeled by an N3E Zen5 desktop processor?

i think it's more of a their chiplet issue than a design issue. ARL still beats Zen5 in many benchmarks and LNL is a very good product on N3B using the same cores but different Uncore also N3B is more like half node ahead vs N4P. Their axing of HT still baffles me

 

[igor_kavinski]

no? in what multi core benchmark i would like to see

AMD EPYC 9655 vs. Intel Xeon 6980P Benchmarks - OpenBenchmarking.org

openbenchmarking.org

 

[igor_kavinski]

Even today, ARL can't best Zen 5 from a full process advantage (N3B vs N4P). Imagine how badly ARL would be pummeled by an N3E Zen5 desktop processor?

Supporting graph because people will be like, WHAT??

 

[511]

Supporting graph because people will be like, WHAT??

View attachment 128866

Phoronix bench has AVX-512 which increases the average it's not that far apart

AMD EPYC 9655 vs. Intel Xeon 6980P Benchmarks - OpenBenchmarking.org

openbenchmarking.org

View attachment 128865

and for this Phoronix hasn't updated their 6980P Benchmarks since the release there have been fixes just for reference also OpenVino was updated with GNR specific optimization it's not even close in those with updates

 

[igor_kavinski]

@511 What does your spidey sense say about Arrow Lake Refresh hitting 6 GHz or higher?

 

[MS_AT]

OpenVino was updated with GNR specific optimization it's not even close in those with updates

Don't keep us in the dark. Share the numbers.

 

[igor_kavinski]

and for this Phoronix hasn't updated their 6980P Benchmarks since the release there have been fixes just for reference also OpenVino was updated with GNR specific optimization it's not even close in those with updates

One would think that Intel would beg Larabel to retest since Linux is what most server CPUs run. How incompetent can Intel be for letting their prized server CPU appear castrated in Linux?

 

[511]

@511 What does your spidey sense say about Arrow Lake Refresh hitting 6 GHz or higher?

i am not even hoping for 5.8Ghz. Most like validated D2D/NGU/L3 Clocks for 5-7% perf improvments that's it

One would think that Intel would beg Larabel to retest since Linux is what most server CPUs run. How incompetent can Intel be for letting their prized server CPU appear castrated in Linux?

well Intel be Intel

Don't keep us in the dark. Share the numbers.

 

[DrMrLordX]

I am wondering if it isn't price as mentioned above?

Maybe. Intel 3 should have been mature enough (remember it's just a + node for Intel 4) that yield issues weren't going to plague a desktop chip. 8P + 16e on Intel 3 - even if it was just a shrink on Raptor Lake or maybe an update with slightly different cores (Redwood Cove + ???) - wouldn't be that big. The main thing would be to get rid of the voltage problems in Raptor Lake that caused problems there, update the product stack, and maybe maintain the high clocks of Raptor at slightly lower power. Arrow Lake's performance regressions (there were some) made it uncomfortable.

That being said, Intel 3 relies on EUV meaning that Intel is capacity-constrained to some extent. Such a hypothetical desktop chip would be vying for wafers with Granite Rapids and Sierra Forest. Intel probably gets better margins on Granite Rapids, and in turn that product needs to pull in as much cash as it can to make up for the firesale on Sapphire/Emerald Rapids.

A lot of folks weren't too happy with returning the 14900K when it burnt up though.

Yup, an updated LGA1700 product on Intel 3 would at least have distracted people from those problems.

 

[MS_AT]

Well, you could have picked up better numbers.

The ones you have chosen look poor. 128/96, 33% more cores for Intel makes it complete the benchmark faster by 25%. In the 6781P vs 9575F it does take 8% less time but uses 25% more cores. The FPS scores where it scores 2x as fast as Turin would have been better showing... just saying