Question Raptor Lake - Official Thread

Page 23 - 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
4,230
2,016
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

coercitiv

Diamond Member
Jan 24, 2014
6,213
11,957
136
Intel should put this on the box of non-K CPUs.
For now they put it on an official page, which is better than nothing I guess.

And here's a Techpowerup thread discussing this issue with examples. Turns out this got better documented in the months after I built my machine.
Sample 1: My 12400F locked at 0.95v can do 3600 G1 with better cooler
Sample 2: 12400F and 0.89v can't run 3200 G1.
Sample 3: 12100F stock cooler 3600G1, 3800G1 with AIO.
Sample 4: 12400F stock cooler 3500 max, 3600 with AIO.
Sample 5: 12700 cheap tower cooler 3400 max.
 
Last edited:

TheELF

Diamond Member
Dec 22, 2012
3,973
731
126
For now they put it on an official page, which is better than nothing I guess.

And here's a Techpowerup thread discussing this issue with examples. Turns out this got better documented in the months after I built my machine.
Non-K SKU Processor with Select PCH SKUs
The Platform Controller Hub (PCH) is a family of Intel's single-chip chipsets.
So is it the CPU or the chipset that will hinder ram performance?
 

FangBLade

Member
Apr 13, 2022
199
395
106
Looks like 7% ipc.. engineering sample 3 supports ddr5-7200 and beyond
I'm not impressed, Zen4-3d will beat it easily, and you won't need extra expensive RAM to get maximum performance just like with Zen3-3d.
 

coercitiv

Diamond Member
Jan 24, 2014
6,213
11,957
136
Looks like 7% ipc..
More like 0-1% IPC.
  • the 13900K scores 880 @ 5.5Ghz
  • the 12900K scores 834 @ 5.2Ghz
The increase in score is near linear with the increase in clocks.
 

cortexa99

Senior member
Jul 2, 2018
319
505
136
28% boost in multi-threaded performance

It roughly equals to ADL with additional 2 E-core clusters. The last thing we have to care about is power, lets hope it wont burn the socket.

"It's recommended to only buy B(mobo?), I cannot imagine how to cool down the stock-voltage 13900KS if it does exist" ---------- this is the last message I saw....
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
More like 0-1% IPC.
  • the 13900K scores 880 @ 5.5Ghz
  • the 12900K scores 834 @ 5.2Ghz
The increase in score is near linear with the increase in clocks.

Hold that thought! The translated comments say that CPU-Z 2.0 loses scores by little bit.

In addition, we don't know how stable the 5.5GHz boost is. If the average clock is close to 5.4GHz and the ver 2.0 of the software is responsible for 1%, then you have a 3% difference.

I'm guessing the final gains might be 3-5%. In an ES sample without proper normalization of the testing systems that's a margin of error difference.

It roughly equals to ADL with additional 2 E-core clusters. The last thing we have to care about is power, lets hope it wont burn the socket.

Intel documents say the PL2 goes up slightly from 241W to 253W in performance mode and 188W in base mode for both CPUs.

The two competing chips will be closer than now. AMD will reduce the single thread gap a bit and Intel will reduce the MT perf/watt gap a bit.

The "massive gap" that Intel had with process only ever translated into huge gains over competitors only for a fraction of the lifetime. Like how the top runner motivates himself just enough to stay ahead. Even companies like Nvidia that has noticeably better execution in products fall into the same trap if you could it that, where it sits back and relaxes a bit.

An older article I was reading at RealWorldTech talked about how perhaps Intel managers would have posted then-Apple's marketing graph(it predicted the gains over PC competition would grow) everywhere in it's corridors to motivate engineers to do better, just as sports teams do. Similar thing could have happened at AMD when developing the first gen Ryzen, and when the company was at it's lowest point.
 
Last edited:

coercitiv

Diamond Member
Jan 24, 2014
6,213
11,957
136
In an ES sample without proper normalization of the testing systems that's a margin of error difference.
Exactly, but my point was about something else:
  • the poster taking for granted the WCCFTech quoted value of ST performance improvement of 7% (it's not, the scores don't match)
  • the poster incorrectly equating this improvement to IPC gain, ignoring the higher clocks even as someone who frequently advertises improved clocks on RTL over previous gen
We could also argue around estimating IPC gain based on a single benchmark, but I'd rather have a productive day instead.
 

DrMrLordX

Lifer
Apr 27, 2000
21,643
10,860
136
I'm not impressed, Zen4-3d will beat it easily, and you won't need extra expensive RAM to get maximum performance just like with Zen3-3d.

In games it might still lose to Zen3d/Vermeer-X.

In addition, we don't know how stable the 5.5GHz boost is.

13900k will probably be binned close to the 12900KS. Probably. Time will tell how much easier it is for Intel to put out working dice at that bin level now that they've had time to (maybe) tweak 10ESF a little.
 

coercitiv

Diamond Member
Jan 24, 2014
6,213
11,957
136
In games it might still lose to Zen3d/Vermeer-X.
It's reasonable to expect Zen3D to lose the "technical" gaming crown to the new boys in town, probably both.

I haven't followed Raptor Lake leaks in terms of improvements to the core, but my money is on what @JoeRambo envisioned: improvements to the uncore that will finally make the changes to cache structure and IMC worth the trade-off in Alder Lake. In essence we get bigger cache with little penalty in terms of speed and a better tuned IMC with lower latency (among other improvements that we care less for now). In some benchmarks this will scale poorly, in others it will shine. Gaming will fall in the latter category.

That being said, expect the end of the year to be all about power supplies (falling like flies): between gaming cards taking transient spikes to eleven and both CPU vendors pushing for 200W+ as the norm for sustained power in flagship CPUs, I expect quite the number of PSU related threads on this forum. So everyone should start brushing up their PSU knowledge, memory and cache is so last year!
 

pakotlar

Senior member
Aug 22, 2003
731
187
116
I'm not impressed, Zen4-3d will beat it easily, and you won't need extra expensive RAM to get maximum performance just like with Zen3-3d.
Note that 12900K underperforms 5950X in cpuz bench (13,000 vs 11,720), but beats it pretty handily on average. A 28% uplift for the ES3 variant, and we may see better in retail, should be enough to exceed 5950X by around 35% on average.
Also worth keeping in mind that L2 cache is nearly doubled in Raptor Cove. CPUZ bench likely does not take advantage of this extra cache.
 
  • Like
Reactions: controlflow

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,573
14,526
136
Note that 12900K underperforms 5950X in cpuz bench (13,000 vs 11,720), but beats it pretty handily on average. A 28% uplift for the ES3 variant, and we may see better in retail, should be enough to exceed 5950X by around 35% on average.
Also worth keeping in mind that L2 cache is nearly doubled in Raptor Cove. CPUZ bench likely does not take advantage of this extra cache.
You should note that the only benchmarks the 12900k wins are ones that are more single threaded. All the FULLY multi-threaded benchmarks are won by the 5950x. And the single thread benchmark differences are not in the double-digit percentages.
 
  • Like
Reactions: Drazick

pakotlar

Senior member
Aug 22, 2003
731
187
116
You should note that the only benchmarks the 12900k wins are ones that are more single threaded. All the FULLY multi-threaded benchmarks are won by the 5950x. And the single thread benchmark differences are not in the double-digit percentages.

Its not quite that simple. FULLY means nothing, either the test hammers all cores, or not. The performance differencea are going to be down to the instruction mix (int, fp), whether program fits in l2 or l3 cache, whether the branches are well predicted by the branch predictor, whether the program uses operations that can be eliminated by the CPU (like zero-initializing variables, which are eliminated by Alder Lake but not Ryzen) etc. See https://chipsandcheese.com/2021/12/02/popping-the-hood-on-golden-cove/ for more.

12900K looks to be between 3 and 10% faster on average in MT. 9% https://www.computerbase.de/2021-11/intel-core-i9-12900k-i7-12700k-i5-12600k-test/5/
4% (MT) in windows 11: https://www.tomshardware.com/reviews/cpu-hierarchy,4312.html

It seems to underperform in your workloads, but that is an n=1 experience, not the average, at least in the chosen tests.
 

moinmoin

Diamond Member
Jun 1, 2017
4,956
7,675
136
I'm confused, what are you quoting there? Computerbase has 5950X as 9% faster than 12900K, not the other way around.

bild_2022-06-29_172348tjnq.png


And Tom's Hardware has 5950X at 53.58% with 12900K at 38.39% for MT.

bild_2022-06-29_1728227k0i.png
 

Exist50

Platinum Member
Aug 18, 2016
2,445
3,043
136
I'm confused, what are you quoting there? Computerbase has 5950X as 9% faster than 12900K, not the other way around.

bild_2022-06-29_172348tjnq.png


And Tom's Hardware has 5950X at 53.58% with 12900K at 38.39% for MT.

bild_2022-06-29_1728227k0i.png

For Tom's, the 5950x did better (at review time) under Windows 10, while the 12900k did better under Windows 11. It's hard to tell where that percent came from, as their review includes very different numbers. Certainly most reviews do not have the 5950x 40% faster. That's a huge gap, which makes one think that must be a weird testing setup. E-cores disabled, perhaps?


Bnwo5AwqrbbTLiBuugfRw5-970-80.png.webp


ComputerBase used Windows 10, it looks like, though they say they didn't see a difference between W10 and W11 for Alder Lake.
 
  • Like
Reactions: controlflow

pakotlar

Senior member
Aug 22, 2003
731
187
116
I'm confused, what are you quoting there? Computerbase has 5950X as 9% faster than 12900K, not the other way around.

bild_2022-06-29_172348tjnq.png


And Tom's Hardware has 5950X at 53.58% with 12900K at 38.39% for MT.

bild_2022-06-29_1728227k0i.png
Err, youre right about the Computerbase article, my bad. That was tested on Windows 10, so thread director is useless there, which will drive those down.

For Toms, if you read the article, the numbers you’re quoting numbers are for Windows 10, where Thread Director doesn’t work, making the MT numbers invalid. They speak to this in the article and point to the Windows 11 results in the first pane:


1D6AF7F0-5013-4565-A6D3-5E2735AA15B1.png
 
Last edited:

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,573
14,526
136
Unless you're citing Phoronix, you're not getting very many Linux benchmarks. The Phoronix Test Suite has problems of its own.
I know personally that the more the CPU is used, and for longer, Alder lake loses, but nobody here wants to hear my evidence. This is a raptor lake thread after all. I just suspect that Raptor lake will take after Alder lake in its general behavior. And I am personally not a fan of E-cores.
 
  • Like
Reactions: Drazick

DrMrLordX

Lifer
Apr 27, 2000
21,643
10,860
136
I just suspect that Raptor lake will take after Alder lake in its general behavior. And I am personally not a fan of E-cores.

Welllll we don't know yet exactly what Raptor Lake is bringing to the table. It may be that its only substantial improvement will be extra Gracemont.