Question Raptor Lake - Official Thread

[Hulk]

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.

 

[shady28]

....I agreed with you. The 12600K is faster than all of Zen 3 in those tasks. But it's because of the P-cores, not the E-cores.

Only reason I say this is because that's what Biostud was asking about. What the E-cores provide. The truth is not much, they boost MT perf at the benefit of Intel's bottom line due to BoM. So if you don't really use programs that benefit from more MT performance, the E-cores don't really give you much.

Depends. Most benchmarks are focused on doing a single task, and measure performance doing that one task.

For example, right now I have this browser open with 2 tabs. iTunes is running in the background, MS Teams is up, EverQuest is running in a window, plus there's this Wallpaper Engine running an animated clock background. Then there's icloud, Steam, MS OneDrive, notepad, calculator, calendar, and outlook.

Nobody ever tests like that. The closest is PCMark 10, and it showed an 8% performance loss with e-cores disabled on an 12600K which only has 4 e-cores.

Even for games, which 'suffer' supposedly due to e-cores, it's dubious to disable them unless you know your specific game suffers.

This was what they found on average :



The worst was -11% and the best was +9%.

For an *average user*, I would just tell them to leave the e-cores enabled.

 

[Carfax83]

Well, we know i5 12600K vs R5 5600X they have very similar gaming performance.

Intel Core i5-12600K Review: 5600X Defeated

The Core i5-12600K is Intel's latest mainstream CPU and a direct competitor to AMD's popular Ryzen 5 5600X. The Alder Lake chip packs 6 P-cores and 4...

www.techspot.com

i5 12600K, 6P+4E

i5 13600K, 6P+8E, this is hm Alder Lake with 4mb more Cache. For gaming, you look only at 6P Cores same as i5 12600K

That review used a DDR4 motherboard for the 12600K, which explains the tighter benches.

As I mentioned before, the midrange Alder Lake CPUs have less cache than the high end models; which ironically makes them even more reliant on the more expensive high speed DDR5 memory to compensate.

Raptor Lake should significantly alleviate that problem by increasing not only the capacity of the cache, but the performance as well making it a better gaming CPU than Alder Laker.

 

[lobz]

I don't need to see anything beyond that part of your response because it's a false misdirection, below is the statement I was replying to.

Here I'll bold it for you so you don't miss it again.




If you look at how something like the 12600K performs in average desktop user use cases (which is to say, superbly, crushing much higher tier Zen 3 chips), what anyone 'thinks' about the e-cores kind of pales next to that reality.

Since we're going down this road again, I'll once again show what it is for a top line chip to get smacked down in typical day to day use cases :

12600K 13.5% faster than 5950X in Octane / Web browser benchmark :

View attachment 66386

12600 > 10% faster than 5950X on all MS Office benchmarks :

View attachment 66387

View attachment 66388

Faster in Photoshop :

View attachment 66389

Faster in video editing:

View attachment 66390

Faster in games :
View attachment 66391

Beautiful selection of benchmarks with zero relevance to E-cores.....

it's like you shot yourself in the foot deliberately.

 

[shady28]

That review used a DDR4 motherboard for the 12600K, which explains the tighter benches.

As I mentioned before, the midrange Alder Lake CPUs have less cache than the high end models; which ironically makes them even more reliant on the more expensive high speed DDR5 memory to compensate.

Raptor Lake should significantly alleviate that problem by increasing not only the capacity of the cache, but the performance as well making it a better gaming CPU than Alder Laker.

The problem with Techspot is that you never know what specific RAM they used. They're not consistent. For example, on their Ryzen 5600X and 5800X reviews they specify that they used DDR4-3200 C14.

But they don't tell us anything about the memory used on their Intel 12600K benchmarks, just DDR4-3200. Did they pull an AnandTech and use JEDEC? Or just some crappy 3200 C18 or C16? Who knows. Their entire testing methodology falls apart when they lack transparency like that. And why specify exactly C14 on all of the AMD reviews, but not say anything on the Intel reviews?

TPU has at least adopted a standard of using DDR4-3600 C14. That way you know what you are looking at, even if those speeds are generally accepted top speed for most Zen 3 due to FCLK limit preventing many from reaching 3800, while Intel generally scales up to 4000+.

Nevertheless, DDR4-3600 C14 is probably one of the more common kits for their reader base regardless of platform, so its relevant. Techspot, I have no idea what they tested.

 

[biostud]

I don't need to see anything beyond that part of your response because it's a false misdirection, below is the statement I was replying to.

Here I'll bold it for you so you don't miss it again.




If you look at how something like the 12600K performs in average desktop user use cases (which is to say, superbly, crushing much higher tier Zen 3 chips), what anyone 'thinks' about the e-cores kind of pales next to that reality.

Since we're going down this road again, I'll once again show what it is for a top line chip to get smacked down in typical day to day use cases :

12600K 13.5% faster than 5950X in Octane / Web browser benchmark :

View attachment 66386

12600 > 10% faster than 5950X on all MS Office benchmarks :

View attachment 66387

View attachment 66388

Faster in Photoshop :

View attachment 66389

Faster in video editing:

View attachment 66390

Faster in games :
View attachment 66391

That is all fine and dandy, but how often do an average user actually start a workload that uses all the cores? It's not that I don't know that in benchmarks where you can load all cores, they will show their strengths, but how often is that scenario likely to happen in your daily work? Why not simply go up to 8P cores midrange and keep the e cores for those who specifically wants lots of cores.

I'm basing my speculation on my own use of computers, where I very seldom need heavy MT performance and therefore don't see the need fo E-cores in midrange, but maybe other people use their computers where it makes perfectly sense.

 

[biostud]

I don't know if I'm the average user but I like having the E cores in my 12700K. I always have all 8 P cores available for the foreground application I'm interacting with but I can still complete work (video rendering, audio multitrack rendering, photo RAW conversion, etc..) in the background. Since all of the background tasks are relegated to the E's I have full P power available at my fingertips at all times. It's important when playing back a multitrack audio file in my DAW with lots of plugs, editing photos, and things like that.

My only complaint is that my 12700K doesn't have enough E's to make this really productive. 4 just isn't enough. I think 8 would work really well. 16 would be fantastic.

I'm hoping the 13900K is a simple plug and play operation from my 12700K.

Thanks for the answer, and kind of my point. E-cores makes sense if you have lots of them and run software that is highly Multithreaded.

 

[Saylick]

Speaking of E cores... Raichu apparently was able to determine the core-to-core latency for Raptor Lake. The penalty for switching between P and E cores is much lower than Alder Lake.

https://twitter.com/x/status/1561908308277366785

 

[Rigg]

The problem with Techspot is that you never know what specific RAM they used. They're not consistent. For example, on their Ryzen 5600X and 5800X reviews they specify that they used DDR4-3200 C14.

But they don't tell us anything about the memory used on their Intel 12600K benchmarks, just DDR4-3200. Did they pull an AnandTech and use JEDEC? Or just some crappy 3200 C18 or C16? Who knows. Their entire testing methodology falls apart when they lack transparency like that. And why specify exactly C14 on all of the AMD reviews, but not say anything on the Intel reviews?

TPU has at least adopted a standard of using DDR4-3600 C14. That way you know what you are looking at, even if those speeds are generally accepted top speed for most Zen 3 due to FCLK limit preventing many from reaching 3800, while Intel generally scales up to 4000+.

Nevertheless, DDR4-3600 C14 is probably one of the more common kits for their reader base regardless of platform, so its relevant. Techspot, I have no idea what they tested.

Are you really calling Steve Walton's integrity into question? The same guy who has been review sample blacklisted by AIB's for giving unfavorable yet objective reviews? If the details on timings were omitted in the written review you can rest assured it was an honest oversight. They do all of their standard DDR4 testing on 3200 CL14 for intel and AMD. They have for years. They have been over this many times on the Hardware Unboxed YouTube channel. Steve and Tim are both upstanding and trustworthy reviewers that have built their following based on that fact. Steve is the gold standard for objective gaming PC benchmarks as far as I'm concerned.

 

[HurleyBird]

Speaking of E cores... Raichu apparently was able to determine the core-to-core latency for Raptor Lake. The penalty for switching between P and E cores is much lower than Alder Lake.

https://twitter.com/x/status/1561908308277366785

Can E-cores in the same complex not communicate directly between one another? It looks like a E-cores need to traverse the entire ring bus to communicate with their closest neighbors.

 

[pakotlar]

Can E-cores in the same complex not communicate directly between one another? It looks like a E-cores need to traverse the entire ring bus to communicate with their closest neighbors.

I think they need to traverse the l3?

 

[IntelUser2000]

Can E-cores in the same complex not communicate directly between one another? It looks like a E-cores need to traverse the entire ring bus to communicate with their closest neighbors.

This is difficult to say. The problem could just be the test. It's a shared cache for L2 so if the test is showing the worst case scenario(or a "worse" one) then the numbers are irrelevant. Shared caches show different values depending on where the cores are accessing it.

 

[IntelUser2000]

Only reason I say this is because that's what Biostud was asking about. What the E-cores provide. The truth is not much, they boost MT perf at the benefit of Intel's bottom line due to BoM. So if you don't really use programs that benefit from more MT performance, the E-cores don't really give you much.

Since Alderlake is the first proper implementation, I would say you are dismissing it too early. Raptorlake should fare quite a bit better on the worst case scenarios because of results shown by leakers like Raichu. Ticks are also refinements that flesh out the features introduced with Tock.

Pentium 4's HT had quite a few bad corner case scenarios. Many people who talked about regressions in the Netburst days said with Nehalem it was gone.

Nehalem didn't bring much for people who needed single threaded performance despite the integrated memory controller, and likely partly due to three level cache change. With Sandy Bridge it was great!

 

[coercitiv]

I always have all 8 P cores available for the foreground application I'm interacting with but I can still complete work (video rendering, audio multitrack rendering, photo RAW conversion, etc..) in the background. [...]

I'm hoping the 13900K is a simple plug and play operation from my 12700K.

Imagine the surprise Anandtech got when testing to highlight this advantage of the hybrid setup. They ran different benchmarks in the foreground while a video transcode process was doing work in the background.

First the control run, then the concurrent load run.



The i5 1260P saw a much higher performance loss in the foreground benchmarks than the 5800U. The reviewer could only assume that something was wrong at the software level (Win11 Thread Director, firmware configuration, or a combination of both).

They even provide data for the transcoding performance:



The hybrid approach has high potential for multitasking performance, but those theoretical gains can only materialize with increased complexity in thread management. Up until now that added complexity is messing things up, even if temporarily.

 

[Carfax83]

The problem with Techspot is that you never know what specific RAM they used. They're not consistent. For example, on their Ryzen 5600X and 5800X reviews they specify that they used DDR4-3200 C14.

But they don't tell us anything about the memory used on their Intel 12600K benchmarks, just DDR4-3200. Did they pull an AnandTech and use JEDEC? Or just some crappy 3200 C18 or C16? Who knows. Their entire testing methodology falls apart when they lack transparency like that. And why specify exactly C14 on all of the AMD reviews, but not say anything on the Intel reviews?

I have no issues trusting tech spot's methodology, and I even understand why they used DDR4 for this particular test because someone buying a 12600K is probably more likely to be using a DDR4 motherboard than someone with a 12900K.

That said, it doesn't change anything that I said about low to midrange Alder Lake CPUs benefiting greatly from high speed DDR5. They are leaving a lot of performance on the table by opting for DDR4; especially for gaming.

Nevertheless, DDR4-3600 C14 is probably one of the more common kits for their reader base regardless of platform, so its relevant. Techspot, I have no idea what they tested

I'm all for increased transparency and I personally hate it when reviewers are vague about important details in their test setup. Though that said, I think Techspot/Hardware Unboxed is a trustworthy source for reviews.

 

[Carfax83]

That is all fine and dandy, but how often do an average user actually start a workload that uses all the cores? It's not that I don't know that in benchmarks where you can load all cores, they will show their strengths, but how often is that scenario likely to happen in your daily work? Why not simply go up to 8P cores midrange and keep the e cores for those who specifically wants lots of cores

That's a fair question. I think the answer is because there has been a long time trend towards increased parallelism at both the hardware and software level. It seems to me there are very few pure single threaded apps these days, and all the high performance applications have long ago migrated over to increasingly sophisticated thread level parallelism based programming.

Efficiency cores are a new thing in Windows and Linux, but the new paradigm shift is going to come no matter what as technological progress is unceasing.......barring a planetwide catastrophe.

Efficiency cores can even be leveraged in games which are typically latency sensitive. Hitman 3 apparently uses the efficiency cores and that game has extraordinarily high performance on Alder Lake CPUs compared to Zen 3.

 

[Hulk]

Imagine the surprise Anandtech got when testing to highlight this advantage of the hybrid setup. They ran different benchmarks in the foreground while a video transcode process was doing work in the background.

First the control run, then the concurrent load run.

View attachment 66398 View attachment 66399

The i5 1260P saw a much higher performance loss in the foreground benchmarks than the 5800U. The reviewer could only assume that something was wrong at the software level (Win11 Thread Director, firmware configuration, or a combination of both).

They even provide data for the transcoding performance:

View attachment 66400

The hybrid approach has high potential for multitasking performance, but those theoretical gains can only materialize with increased complexity in thread management. Up until now that added complexity is messing things up, even if temporarily.

Yes, I noticed this behavior with the 1260P in this test as well. Obviously the E cores AND P cores are working on the foreground application in this test. My 12700K seems to behave differently as it seems to keep the foreground to the P's and background to the E's. Perhaps it's due to the 8/4 vs. 4/8 P to E ratio? Meaning the TD behaves differently depending on P/E resources available? I don't know. ADL mobile testing hasn't been nearly as widespread as desktop.

 

[IntelUser2000]

Yes, I noticed this behavior with the 1260P in this test as well. Obviously the E cores AND P cores are working on the foreground application in this test. My 12700K seems to behave differently as it seems to keep the foreground to the P's and background to the E's. Perhaps it's due to the 8/4 vs. 4/8 P to E ratio? Meaning the TD behaves differently depending on P/E resources available? I don't know. ADL mobile testing hasn't been nearly as widespread as desktop.

I wouldn't be surprised the mobile chips have different algorithms. Even back in Core 2 days Intel said the chips have different prefetch optimizations for different markets.

That said, we do have teething pains on Alderlake. A significant progress and a win but lots of rough edges. Comparison using the Framework laptop showed battery life regression.

Raptorlake may address both hybrid performance issues and mobile battery life issues. I don't expect great battery life, but wouldn't be surprised if it gets back to Tigerlake levels. Raichu's tests up few posts are already showing great improvements in the c2c latency figures for example.

 

[igor_kavinski]

Time for some cheese!

A Preview of Raptor Lake’s Improved L2 Caches

Caching is important. We recently got a subset of our test suite run on an engineering example of Raptor Lake. Those results let us take a look at Intel’s changes to their caching hierarchy. …

chipsandcheese.com

 

[igor_kavinski]

My 12700K seems to behave differently as it seems to keep the foreground to the P's and background to the E's.

Would you be able to do a timed benchmark of some sort to test how much slower the foreground app becomes due to the background process running? How much is the slowness percentage? I'm curious if it is less than what ADL mobile suffered in that AT multitasking benchmark.

 

[coercitiv]

Yes, I noticed this behavior with the 1260P in this test as well. Obviously the E cores AND P cores are working on the foreground application in this test. My 12700K seems to behave differently as it seems to keep the foreground to the P's and background to the E's. Perhaps it's due to the 8/4 vs. 4/8 P to E ratio? Meaning the TD behaves differently depending on P/E resources available? I don't know. ADL mobile testing hasn't been nearly as widespread as desktop.

I'm not gonna pretend to know what's happening there. Instead I'll try to make some time next week to test on 12700K, it can easily be done with CB23 + Handbrake.

 

[igor_kavinski]

Sorry. Wrong thread.

 

[DrMrLordX]

I don't need to see anything beyond that part of your response because it's a false misdirection, below is the statement I was replying to.

Sorry, but that entire post of yours was offensively dishonest.

User asks about e-cores. You quote this user, directly, bolded, in enlarged text.

You then go on about the 12600k, which has no e-cores only 4 e-cores (and you used benches that didn't really rely on the e-cores), because apparently you think that's more relevant somehow? And you then accuse an honest respondent of "false misdirection"?

 

[igor_kavinski]

You then go on about the 12600k, which has no e-cores

Umm...12600K does have 4 E-cores.

 

[DrMrLordX]

Umm...12600K does have 4 E-cores.

Yeah my bad. But the tests he showed didn't really showcase their performance.

 

[pakotlar]

Time for some cheese!

A Preview of Raptor Lake’s Improved L2 Caches

Caching is important. We recently got a subset of our test suite run on an engineering example of Raptor Lake. Those results let us take a look at Intel’s changes to their caching hierarchy. …

chipsandcheese.com

~16% fewer l2 misses