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.

 

[igor_kavinski]

That's like 1/3 the price of a laptop with it!

I wouldn't be surprised if the larger OEMs get something like a 40% discount. And it's very possible that near the end of a CPU's lifecycle, Intel may just give them away as "gifts" to their preferred OEMs.

 

[FlameTail]

I wouldn't be surprised if the larger OEMs get something like a 40% discount. And it's very possible that near the end of a CPU's lifecycle, Intel may just give them away as "gifts" to their preferred OEMs.

Wahut.

How does AMD compare?

 

[FlameTail]

I wouldn't be surprised if the larger OEMs get something like a 40% discount. And it's very possible that near the end of a CPU's lifecycle, Intel may just give them away as "gifts" to their preferred OEMs.

Yeah such an explanation is necessary.

Like for instance: the Core i3 1215U's recommended customer price is $300, which is insane because you can find laptops with it for like $500.

 

[igor_kavinski]

How does AMD compare?

@adroc_thurston won't tell

 

[FlameTail]

@adroc_thurston won't tell

So is Adroc like an AMD insider or something?

 

[igor_kavinski]

So is Adroc like an AMD insider or something?

Either that or he is sleeping with someone from AMD

Hella stimulating bed talk

 

[Timur Born]

I observed (and reported earlier) a peculiar behavior of my 13900K, of which I am not sure whether it's BIOS, CPU (microcode) or Windows 11 based.

When I run pure single-core load or low-core load then one of my cores if kept in C1 state even when it's close to 99% idle. Unsurprisingly this affects boosting behavior, either of P or E cores, albeit more often (and more pronounced) on P cores. It is as if the scheduler kept one core "ready" in C1 for faster reaction times.

Example: running single-core Cinebench load on P core 2 may lead to core 3 idling in C1 instead of C6, which in turn means 2 active cores instead of 1.

Here is the part I didn't report earlier:

Disabling "C1 Demotion" via Throttlestop's solves that particular problem, but it does so by creating an even worse problem. When "C1 Demotion" is disabled most cores stay in C6 leading to very noticeable slowdowns as even the most active core will stay over 80% in C6 then while everything gets choppy rather than waking up for work.

 

[igor_kavinski]

Thank you for reporting that. Dang, Intel! Giving us beta CPU architectures without fixing all the errata first.

 

[Timur Born]

I mostly suspect the OS (W11), because I can make this happen/unhappen by running some software. Like making it happen on a P core by running Discord on E cores in the foreground while doing a CB single-core run on a P core and then make it unhappen by minimizing the Discord window (that never ran on P cores anyway). Or running a combination of processes including Share X instead of Greenshot even when both are just idle anyway.

Might be worth playing around with power plans, but none of the preset ones make a difference for this (including power modes), so it would be shooting in the dark within the power plan options.

 

[JoeRambo]

I mostly suspect the OS (W11), because I can make this happen/unhappen by running some software.

it is software based, basically some stuff that runs is requesting so many wakeups that it does no longer make sense for a core to go to sleep to deeper states and that cascades into stupid power usage all over the package that is now unable to go deeper into power savings.

For example with Skype ( ironically made by the devil himself ):
They have dubious distinction of being made by power ignorant persons and make this many context switches while completely idle and closed to taskbar notification icon ( in DND mode even):



Resulting C states look like this:


Notice the almost 20% in C0% for first core and how unhappy with this all are the package C states.
Now with Skype program closed:


Less active core equals more time to save power package wide, simple as that.

This whole "framework" mania, where apps are made with Electron etc is efficiency cancer, while some might argue a gigabyte here or there no longer matter, the impact on planet is in Gigawatt hours for sure.

P.S. not singling out Skype here, some other software is equally bad in this regard.

 

[DAPUNISHER]

Thank you for reporting that. Dang, Intel! Giving us beta CPU architectures without fixing all the errata first.

It isn't just Intel having issues with C states.

 

[Timur Born]

it is software based, basically some stuff that runs is requesting so many wakeups that it does no longer make sense for a core to go to sleep to deeper states and that cascades into stupid power usage all over the package that is now unable to go deeper into power savings.

This is not a problem of specific software. The Discord configuration I listed was just an example. Next reboot or even next restart of Cinebench it can be different. And Discord doesn't do so many context switches compared to other stuff running in the background.

Software *does* affect it, but it's more the total combination of what is running. A few months back it always hit a random core with each new run, P or E core, no matter. Now it always seems to regularly shift to my next priority P core after some time.

 

[Timur Born]

GPU hardware acceleration does seem to affect this, so I speculate that IRQs/DPCs being handled by a core(s) may keep it (them) in C1 despite lower CPU load being measured.

When I do smooth scrolling via middle mouse-button in Discord with GPU hardware acceleration while running CB24 single-core then more cores are kept out of C6 compared to doing the same thing without hardware acceleration.

 

[JoeRambo]

And Discord doesn't do so many context switches compared to other stuff running in the background.

Software *does* affect it, but it's more the total combination of what is running.

it's context switches + time spent doing CPU work i think. They are just too heavy and keep CPU busy for too long and too often and in the end defeat CPU PMU algorithms by presenting enough load for a single core to keep in low C states. If anything i think Windows does stellar job to keep all this localized on "hot" core.
That in turn means package cannot progress to power saving mode where things are powered down more.

 

[Timur Born]

I don't see anything particularly special about Discord in this regard. Also keep in mind that Discord is run on E cores here, while in my example it was a P core that got stuck in C1 when Discord was running.

 

[igor_kavinski]

Very interesting!

Though I wanna punch that guy. Not too hard. Just enough for him to realize that he annoyed me by doing testing involving multiple monitors without saying why he did that. What are the two monitors doing with the game? Is the game running on both monitors? Or one monitor for game and other for desktop? Seriously, more questions than answers.

 

[DAPUNISHER]

Very interesting!

Though I wanna punch that guy. Not too hard. Just enough for him to realize that he annoyed me by doing testing involving multiple monitors without saying why he did that. What are the two monitors doing with the game? Is the game running on both monitors? Or one monitor for game and other for desktop? Seriously, more questions than answers.

From what little I've seen of his channel, he is either shilling for AMD or pandering to the stans.

For the multi monitor explanation, I think you have to watch his vid about switching from Nvidia to AMD because of issues related to them. IIRC Aussie Steve recently said the same thing about switching because of Nvidia issues with his monitors. Anyways, my money is on him when you fight, you little horse jockey. I kid I kid nothin' but love for ya.

hopefully win12 handles these new CPU designs better.

 

[Shmee]

It isn't just Intel having issues with C states.

Interesting, though this is obviously talking about Zen 4 on Win 11, what about Zen 3 on Windows 10? Also, how is power usage affected?

 

[Timur Born]

Concerning the video:

Disabling C-states means to also disable single/low core boosting. So games that run on all cores may benefit from the cores staying awake while games running on few cores may suffer from the lower frequencies.

There is a bit of a mix-up with the Windows power plan and minimum frequency in this video. What the (Ultimate) power plan does is change the ramp up/down reaction/time and what the minimum frequency does is effectively disable P-states (EIST). All of this is less relevant nowadays, because "autonomous mode" is enabled by default in all plans (Speedshift on Intel) and often the CPU enters a C-state rather than clocking down to another P-state (race to halt and C1E).

All that being said, there are different C-states and the video doesn't tell us which were disabled or not. On AMD you get less control over these things in BIOS (basically just one on/off setting), but especially C7+ package C states may matter more than the rest. Starting at C7 package states the L3 cache is flushed. And from 3D cache CPUs we know that L3 cache can make quite a difference in fps. Whether the CPU even reaches PC7 while gaming is another question, though.

 

[igor_kavinski]

I mostly suspect the OS (W11)

Easy to rule that out by trying Win10 or Linux (specifically Clear Linux) on a spare SSD. If Intel's own distro doesn't deal properly with C-states management, then we know what the problem is.

 

[Hulk]

Here is the Turbo Boost Behavior as described by Intel for the 11900K.


Do these look correct for the 13900K and 14900K?

 

[Timur Born]

Thermal Velocity "Boost" is an odd one in these descriptions, since it (seems to be?) more of a limiter than a boost.

 

[youself64]

Here's link for benchmarking boot times in copy.sh/v86:

Also, must include pc specs and screenshot with stopwatch + copy.sh/v86.

This is first benchmark of x86 Emulation performance (copy.sh)

I benchmarked dsl linux in copy.sh v86 boot time for me, it takes about 2 minutes to fully boot (i5-2500K, 4GB RAM). So i need to benchmark this in Raptor Lake CPUs (Including overclocked.)


I am not liking this site, and requires more clearification, as it seems to be accessing hardware without consent and to me feels like its data mining.

Please PM me with more info if this is your site, otherwise i will not approve of it as there is no disclaimer when hitting that link in exactly what its doing.

Moderator Aigo

 

[jpiniero]

The Core (Not Ultra) Raptor Lake U did get a clock speed boost at least... The Core 7 goes up to 5.4 on P and 4 on E.

 

[coercitiv]

Links require commentary, so here it it: hahahahahahahaahhahha

Intel to roll out 14th Gen's game optimization software to older 12/13th Gen hybrid CPUs after all

That's an unexpected change of heart.

www.pcgamer.com

They could have chosen a neutral stance from the start (we're analyzing the possbility...) and then come up with the good news if and when they wanted. They chose the bitter sweet approach instead.