Question Raptor Lake - Official Thread

Page 95 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Hulk

Diamond Member
Oct 9, 1999
4,191
1,975
136
Since we already have the first Raptor Lake leak I'm thinking it should have it's own thread.
What do we know so far?
From Anandtech's Intel Process Roadmap articles from July:

Built on Intel 7 with upgraded FinFET
10-15% PPW (performance-per-watt)
Last non-tiled consumer CPU as Meteor Lake will be tiled

I'm guessing this will be a minor update to ADL with just a few microarchitecture changes to the cores. The larger change will be the new process refinement allowing 8+16 at the top of the stack.

Will it work with current z690 motherboards? If yes then that could be a major selling point for people to move to ADL rather than wait.
 
  • Like
Reactions: vstar

DrMrLordX

Lifer
Apr 27, 2000
21,583
10,785
136
And if it gets them some extra performance?

At that point they'd be better off just going full-bore Raptor Cove on the new part. There's really not much to be gained from cutting back on the L2 for mobile Raptor Lake. Very soon we'll see the parts for real, and we'll know for sure if they actually made yet another core design just for Raptor Lake mobile.
 

dullard

Elite Member
May 21, 2001
24,998
3,327
126
Up to certain amount of CPU cores yes... But that is no longer the case when you cross the 18 Core CPU barrier. You can bet that those monolithic CPUs will have Subnuma domains(to alleviate the latencies of such large Mesh), the power consumption on such large mesh is significant too.

As an example this 28C/56T Xeon CPU with Mesh interconnect and Subnuma clusters(now called Subnuma Domains)
Break that 28C core CPU into separate chiplets and you still have the clusters PLUS longer distances for signals to travel. You are confusing the issue of monolithic/chiplets with the plusses/minuses of different mesh types.
 
  • Like
Reactions: yuri69

naad

Member
May 31, 2022
63
176
66
That single thread score looks like a joke, almost 30% off from desktop chips.
I know they're laptops but geekbench is bordering on microbenchmark territory, so boosting for seconds shouldn't be a problem
 
  • Like
Reactions: Kaluan

nicalandia

Diamond Member
Jan 10, 2019
3,330
5,281
136

tamz_msc

Diamond Member
Jan 5, 2017
3,726
3,560
136
Based on this:

AMD-Ryzen-7000-Performance-Summary.png

from here, the 13900K should have no trouble matching the 7950X in applications and beat it in gaming by 5% by my estimation.

This is because from this image:

2022-06-09%2013_17_33_575px.jpg


the Cinebench R23 performance uplift estimate of >35% turned out to be a very good proxy for overall performance uplift in applications (comparing the 7950X and 5950X). With Intel promising up to 41% MT performance for the 13900K over the 12900K, at similar power consumption levels (253 W vs 241 W), I expect application performance to be a wash between the 13900K and the 7950X.
 
  • Like
  • Haha
Reactions: Geddagod and Kaluan

inf64

Diamond Member
Mar 11, 2011
3,686
3,958
136
Although the 3dcenter effort is commendable and shows something similar to what techpowerup has in their performance summary, I like better the separate MT/ST indexes like this one :

1664616145953.png

7950X is around 52% faster in MT according to this index. Interestingly, Computer base has almost exactly the same difference (~50%) in their own MT benchmark suit:
1664616397987.png

So 1.5x vs 1.4-1.45x, the performance will be close but 7950X should still be faster vs 13900K stock vs stock in MT workloads (average).
ST and gaming is a bit trickier but I guess 13900K might be a percentage or so behind in ST and similarly a few percent ahead in gaming. Overall, it will be very close.
 

Abwx

Lifer
Apr 2, 2011
10,854
3,298
136
from here, the 13900K should have no trouble matching the 7950X in applications and beat it in gaming by 5% by my estimation.

This is because from this image:

2022-06-09%2013_17_33_575px.jpg


the Cinebench R23 performance uplift estimate of >35% turned out to be a very good proxy for overall performance uplift in applications (comparing the 7950X and 5950X). With Intel promising up to 41% MT performance for the 13900K over the 12900K, at similar power consumption levels (253 W vs 241 W), I expect application performance to be a wash between the 13900K and the 7950X.

The 7950X score 50% better than the 12900K overall , but somewhat with up to 41% for the 13900K it will be a tie .?.



Edit : To keep up with the 7950X they ll have to increase TDP well over 253W, expect higher peak power than this number...
 
Last edited:

nicalandia

Diamond Member
Jan 10, 2019
3,330
5,281
136
Intel promising up to 41% MT performance for the 13900K over the 12900K, at similar power consumption levels (253 W vs 241 W), I expect application performance to be a wash between the 13900K and the 7950X.
It's actually a bit less, for example Geekbench.

12900K vs 13900(65 Watts) are identical here:

But the 13900K vs 13900 the MT performance boost is only 32%
 
Last edited:

pakotlar

Senior member
Aug 22, 2003
731
187
116
Raptorlake improvements:
-Improved Intel 7 process with greater than 50mV reduction at the same frequency, 200MHz increase at the same voltage, and 600MHz increase in clocks at peak
-Machine learning prefetcher for the L3 allows INI caching scheme. INI = Inclusive/Non Inclusive, and it can switch between the two on the fly depending on the needs in 200us times. Was in Alderlake but didn't get used. Sounds like what Raichu was talking about regarding "secret sauce"? Though that was about Golden Cove.
-L2 cache in the E cores also use machine learning to improve prefetching. Performance improves from 2% in some circumstances to 16% in others.
-Fabric frequency increased by 900MHz to 5GHz. Apparently in leaks it said 4.6GHz with E cores active?
-The improvement with caching and prefetchers along with caches allow E cores to perform just like the original Skylake.

Quite an impressive change to uarch for 1 year. Intel may be getting its house in order.
 
  • Like
Reactions: Geddagod

pakotlar

Senior member
Aug 22, 2003
731
187
116
@nicalandia Caches are very power efficient and usually result in LOWER power at the chip level not the other way around.



The ideal is skipping Meteorlake on the high end and go straight to Arrowlake like some leakers have said to get 20A as quick as possible.

Agreed, as long as ArrowLake follows as closely as speculated. I doubt Raptor Lake will do well against Zen 5 in 2024.
 

tamz_msc

Diamond Member
Jan 5, 2017
3,726
3,560
136
Although the 3dcenter effort is commendable and shows something similar to what techpowerup has in their performance summary, I like better the separate MT/ST indexes like this one :

View attachment 68526

7950X is around 52% faster in MT according to this index. Interestingly, Computer base has almost exactly the same difference (~50%) in their own MT benchmark suit:
View attachment 68527

So 1.5x vs 1.4-1.45x, the performance will be close but 7950X should still be faster vs 13900K stock vs stock in MT workloads (average).
ST and gaming is a bit trickier but I guess 13900K might be a percentage or so behind in ST and similarly a few percent ahead in gaming. Overall, it will be very close.

Not all applications are multithreaded and even then, the multithreaded scaling varies by application. So selectively focusing on just those applications that show good multithreaded scaling is nothing but cherrypicking.
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Quite an impressive change to uarch for 1 year. Intel may be getting its house in order.

Still too early to say. Raptorlake plans would have been set in stone years ago. It's not until Meteorlake part of the changes start to apply. I assume until then a good management just prevents things from getting worse.

Even Meteorlake should only have a partial effect from new management.

That's why I've always been confused about the resistance to the idea that this product exists. It makes complete economic sense. The only thing even remotely in question was whether Intel could build it.

They might as well go all the way.

4x 34 core MCC die Emerald Rapids-AP. Maybe 4x 24 core Emerald Rapids for regular chips?

:p
 
  • Like
Reactions: Geddagod

Kaluan

Senior member
Jan 4, 2022
500
1,071
96
So Far we have Geekbench and Sisoftware confirmed Intel's Claims about 65 Watt 13900K(Basically 13900 non-k) matching the 12900K in benchmarks.

View attachment 68553

View attachment 68554
13900 is not "13900K @ 65W" lol
Firstly because it has different clocks than 13900K
Secondly because it's not 65W at all (more like 202W or more, going by 12900), hell, expect most upper end to run it PL1=PL2 indefinitely, provided there is proper 200W+ cooling capable cooling on it of course. That's like saying the 12th i9K is scoring 27K in CineBench @ 125W lmao
And thirdly it's not an unlocked part. But this is more about forced SKU differentiation than "stock" behavior.

In essence, you are comparing a 8+16 202W+ Raptor Lake to a 8+8 241W Alder Lake. Which is still not too shabby as a upgrade efficiency-wise, but it's not the misleading garbage Intel is pushing in their slides/product specs. Which sadly I still see people fall for to this day.

Hell, stuff like 12700 and 12900 already did similar things vs the K versions. It's just a bit better now because of the Intel 7 tweaks and E core spam.
 

moinmoin

Diamond Member
Jun 1, 2017
4,934
7,619
136
Anybody know if it's already confirmed that locked 13000 series chips will keep the locked IMC voltage for ADL/12000 series chips, or may we get an improvement back to the old state there?