Raptor Lake 6 GHz - Intel Tech Tour

[porina]

Summary

Intel Tech Tour is current running in Israel, and tech journalists have got some news on what the upcoming 13th Generation CPU Raptor Lake will bring. DDR5 5600 support, up from 4800 of Alder Lake.

 

6 GHz single core turbo. 41% improved MT perf over Alder Lake. 15% ST perf relative to Alder Lake. SPECint2017

 

Quotes

Quote

Notably, the peak of 6 GHz is 300 MHz faster than the 5.7 GHz for AMD's Ryzen 7000 processors, but Intel hasn't announced which product will hit that peak speed. We also aren't sure if a 6GHz chip will arrive with the first wave of chips or be a special edition 'KS' model. Intel also claimed that Raptor Lake will have a 15% gain in single-threaded performance and a 41% gain in multi-threaded, as measured by SPECintrate_2017 and compared to Alder Lake, and an overall '40% performance scaling.' 

 

My thoughts

Zen 4 is looking like a strong CPU, but Intel isn't standing still either. Even though Raptor Lake isn't a major architectural change from Alder Lake, they're squeezing out more performance from already known factors such as bigger caches, and more E cores. The clock increase does bring back memories of the P3/P4 vs Athlon era. Interesting that Raptor Lake is stated as a gap filler because Meteor Lake is late.

 

On ram, I don't recall if Zen 4 DDR5 speeds supported have been officially given, with 5200 seen in a leaked mobo diagram, and a stated sweet spot (presumably EXPO) of 6000, like Zen 3 supported 3200 with sweet spot of 3600.

 

Sources

https://www.tomshardware.com/news/intel-teases-8-ghz-raptor-lake-world-record

 

 

 

 

[Motifator]

It's still big.little cores and at an inanely insane power consumption that asks for the form of massive cooling system I had before, which consisted of two EK Phoenix units connected to each other over disconnects, lol.

AMD has only old school big cores yet their power consumption is lower while performing close? At worst their platform asks for D5, but it's not like Raptor Lake doesn't benefit from DDR5 either... D4 seems on its way out with CPUs so modernized to D5.

[genexis_x]

At what cost? Free sauna in your house? Sounds like a good 2in1 deal

[porina]

9 minutes ago, Motifator said:

AMD has only old school big cores yet their power consumption is lower while performing close?

Zen 4 is made on an optimised TSMC N5 process, while Intel is still on their "7" for this generation. The upcoming generation AMD CPUs will have over one process node advantage over Intel. Keep in mind Intel are still on their process recovery plan, which is supposed to catch up and pass TSMC around 2025.

[LAwLz]

19 hours ago, Motifator said:

It's still big.little cores and at an inanely insane power consumption that asks for the form of massive cooling system I had before, which consisted of two EK Phoenix units connected to each other over disconnects, lol.

AMD has only old school big cores yet their power consumption is lower while performing close? At worst their platform asks for D5, but it's not like Raptor Lake doesn't benefit from DDR5 either... D4 seems on its way out with CPUs so modernized to D5.

I think people need to stop focusing on the absolute highest end SKUs and generalizing product stacks based on that.

Yes, the KS version of the i9 processor uses a lot of power, especially when the power limiter is removed. But if we look at more sane workloads (like gaming), with more sane SKUs that people actually buy (like the i3, i5 and i7) then Intel isn't really behind in terms of power consumption or efficiency either:

 

 

 

 

Does anyone actually care that the 5600X uses 5 watts less than the 12600K when gaming?

We are talking about minuscule differences that doesn't matter. People focus too much on these ridiculous scenarios that doesn't matter and miss the truly important things.

We should focus less on

"this 700 dollar CPU uses less power than this 700 dollar CPU when calculating Pi"

and focus more on

"this 300 dollar CPU uses pretty much the same power as this other 300 dollar CPU when playing games or doing web browsing"

 

 

Zen4 seems to be quite a bit more power hungry than Zen3 as well. So I'd be very careful with comparing one unreleased product vs another unreleased product and declaring a winner.

I think both Zen4 and Raptor Lake seem very impressive so far. That's really impressive when taking into account that neither Zen4 nor Raptor Lake are big redesigns. Both of them are rather small changes but with big performance uplifts (probably from increased clock speeds).

[Motifator]

I'd take anything from Igor with a whole bar of ionized legit China sea salt after his buttwipe made up comment about capacitors of GPUs. Don't use him as a source please.

If you go back to TPU, you'll find out that a 12600KF with higher power limits hits 250W easily while that 5600 will pull 50-75W. This doesn't make any sense.

[porina]

2 minutes ago, Motifator said:

If you go back to TPU, you'll find out that a 12600KF with higher power limits hits 250W easily while that 5600 will pull 50-75W. This doesn't make any sense.

Where exactly did you get those power values from? Are they like for like? I can imagine 12600KF possibly hitting over 200W under a heavy load, but under same heavy load I'd expect 5600 to hit 88W which is the stock PPT limit that AMD enforce. Intel gives the system builder the choice of power limit and doesn't enforce one, so then it comes down to comparing how they might perform at a given power limit.

[LAwLz]

17 minutes ago, Motifator said:

I'd take anything from Igor with a whole bar of ionized legit China sea salt after his buttwipe made up comment about capacitors of GPUs. Don't use him as a source please.

If you go back to TPU, you'll find out that a 12600KF with higher power limits hits 250W easily while that 5600 will pull 50-75W. This doesn't make any sense.

Okay, I'll find another source for you.

 

I am not sure where you got that 12600KF number from. I checked out TechPowerUp and they don't seem to have a 12600KF review.

I found their 12600K review and in that their multi-threaded Cinebench workload resulted in these measurements, but please keep in mind that this is for the entire PC not just the PC:

5600X - 126 Watts.

i5-12600K - 189 watts.

 

The i5-12600K however gets a higher score in Cinebench, so the energy efficiency score ends up looking like this (lower is better):

5600X - 9.6kJ

i5-12600K - 10kJ

 

The difference in efficiency is 4%.

We are talking a 0.1W difference when doing a single run in Cinebench, and that's something that pushes the CPU to 100% load. For things like gaming, where the CPU isn't constantly under 100% load, the difference will not be the same. The power consumption will be lower for both CPUs.

 

 

Not sure where you got the 250W number from for the i5, or the 50W number from for the 5600.

 

 

 

Tom's Hardware sadly doesn't have any gaming tests, but they do have an interesting HandBrake test. This is the result from that:

Quote

As you can see, Intel's chips have descended from the undesirable upper right of the chart down to the lower left hand, nearly matching AMD's chips in power consumption while actually being faster. That's an outstanding improvement after six years of power-guzzling 14nm chips. 

 

 

 

According to PCInvsaion's tests, the 12600K will peak at 90 to 110 watts during 144Hz gaming, with some games like Fortnite 144Hz 1080p only using around 65 watts on average (from the CPU, the 12600K).

Even if a processor from AMD is more efficient, we are talking tiny differences. Nobody should choose favourite CPU brand based on a 5 watt difference.

[johnt]

Oooh man the new intel CPUs are arriving just in time to keep my house warm this holiday season

 

6 GHz stock? B to the S.... and if true, then most likely a short term favored core boost. Definitely don't expect this for sustained clock speed.

[Rauten]

[Somerandomtechyboi]

This is just ridicolous, now its back to clock speed wars bullshittery and bye bye efficiency for max performance because haha cpu go vroom vroom and now your pc doubles as a space heater (nice for winter but haha rip for summer and all the folk in tropical countries incl me)

 

looks like this new gen will benifit tons from undervolting

 

Also seems kinda odd with the ludicrous speeds theyre running stock yet still no 8ghz xoc, still nothing to top them cheddar mill celeron d and bulldozer fx maximum xoc freq wise and thatll likely stay that way for a long time

[CephDigital]

Power draw is going to be insane, isn't it? 

[porina]

31 minutes ago, johnt said:

6 GHz stock? B to the S.... and if true, then most likely a short term favored core boost. Definitely don't expect this for sustained clock speed.

Max clocks for any consumer CPU have always been for one core turbo. If it is favoured core or any core doesn't matter. AMD kinda lead the way there since around Zen 2 I think (maybe Zen+?), with each core having different boost characteristics. Traditional Intel (without favoured core) expects single core boost can happen on any core, so an overclocker could sustain that on all cores with sufficient cooling.

 

21 minutes ago, Somerandomtechyboi said:

looks like this new gen will benifit tons from undervolting

Power limit setting is safer. Undervolting is basically trying to trade off stability for lower power usage.

 

 

AMD started the push towards inefficient running of CPUs, with Intel having no choice but to follow. If you want more power efficiency, just set a lower power limit. Keep in mind AMD have also increased power budgets for their upcoming generation, with 170W TDP translating to a power limit of 230W PPT. 

[Somerandomtechyboi]

33 minutes ago, porina said:

Power limit setting is safer. Undervolting is basically trying to trade off stability for lower power usage.

You have to be absolutely joking me

 

Safer? Safer my f ing ass

You cannot kill hardware with low volt, only high volt, and even then deathzone for all modern silicon processes is still 2v+ (like insta die not degrading)

 

power limit is shittier than manual undervolt, you dont optimize the volt (curve) so still higher power consumption vs proper undervolt, you are NOT trading off stability but you ARE trading max clockspeed (usually well worth it since undervolt is really effective on newer cpus to lower power consumption, esp intel with retarded boards running 1.35v+ just for 5ghz when thats acheivable with <1.3v)

[leadeater]

2 hours ago, porina said:

On ram, I don't recall if Zen 4 DDR5 speeds supported have been officially given, with 5200 seen in a leaked mobo diagram, and a stated sweet spot (presumably EXPO) of 6000, like Zen 3 supported 3200 with sweet spot of 3600.

Supported RAM speeds tend to always be whatever the maximum JDEC official speed is at the time of release, that product spec has more to do with that than anything else.

[leadeater]

11 minutes ago, Somerandomtechyboi said:

You have to be absolutely joking me

 

Safer? Safer my f ing ass

You cannot kill hardware with low volt, only high volt, and even then deathzone for all modern silicon processes is still 2v+ (like insta die not degrading)

 

power limit is shittier than manual undervolt, you dont optimize the volt (curve) so still higher power consumption vs proper undervolt, you are NOT trading off stability but you ARE trading max clockspeed (usually well worth it since undervolt is really effective on newer cpus to lower power consumption, esp intel with retarded boards running 1.35v+ just for 5ghz when thats acheivable with <1.3v)

Safer in relation to system stability. If you don't do it well enough then stability will be compromised, hence why power limit is safer. Something being more effective doesn't mean it's safer and easier to do, it can certainly do a better job and have a better outcome when done well but that's not exclusively true always, where as a power limit won't do that (beyond extreme edge case of like setting it to 8W which I've never tried or seen anyone try)

[porina]

9 minutes ago, Somerandomtechyboi said:

Safer? Safer my f ing ass

You cannot kill hardware with low volt, only high volt, and even then deathzone for all modern silicon processes is still 2v+ (like insta die not degrading)

Who said anything about killing? I meant safer in the sense of system stability. The manufacturer chosen volt curve is to ensure it works under any situation. Reducing it may expose weaknesses in some areas e.g. people disregarding prime95 stability. In order of priority I'd always go stability first, then performance, then efficiency. Undervolting is basically another dimension of overclocking, pushing the operating condition outside what is guaranteed to work. That is for an individual user to decide but not for the masses.

 

2 minutes ago, leadeater said:

Supported RAM speeds tend to always be whatever the maximum JDEC official speed is at the time of release, that product spec has more to do with that than anything else.

Was there that much time gap between Zen 4 and Raptor Lake then, given their not too distant expected releases.

[Dracarris]

3 minutes ago, porina said:

Who said anything about killing? I meant safer in the sense of system stability. The manufacturer chosen volt curve is to ensure it works under any situation. Reducing it may expose weaknesses in some areas e.g. people disregarding prime95 stability. In order of priority I'd always go stability first, then performance, then efficiency. Undervolting is basically another dimension of overclocking, pushing the operating condition outside what is guaranteed to work. That is for an individual user to decide but not for the masses.

Just keep the V/f curve from the manufacturer but don't go as high up. Lower V, lower f, higher energy efficiency, much lower power consumption (scales ~ to the third power of voltage).

[Somerandomtechyboi]

1 minute ago, leadeater said:

Safer in relation to system stability. If you don't do it well enough then stability will be compromised, hence why power limit is safer. Something being more effective doesn't mean it's safer and easier to do, it can certainly do a better job and have a better outcome when done well but that's not exclusively true always, where as a power limit won't do that (beyond extreme edge case of like setting it to 8W which I've never tried or seen anyone try)

You got to be an absolute block head to fail undervolting, its literally just setting vcore and cpu multi/freq (other volts just not worth lowering i think), i have mainly only messed around with oc since im on old af hardware and theyre set wayyy below their actual v/f efficiency peak (4.2-4.4g on the 45nm e8x00 and 32nm westmere x56xx cpus), though once i did screw around with vundervolting the crap out of an e5800 just to see if i can run it without a heatsink and just a fan over it, 0.9v 2.1/2.2ghz iirc and ran mid 70s only with a chinese led fan (not even overvolted to 24v) on p95 smallest ffts but didnt take pics cause it was just a stupid experiment

 

i can understand for gpu undervolt cause those use curves and you have to test all the freq/volt points on the curve

[leadeater]

8 minutes ago, porina said:

Was there that much time gap between Zen 4 and Raptor Lake then, given their not too distant expected releases.

 Your guess is as good as mine. Would have to go through the product roadmaps and product lifecycle timelines to figure it out.

 

Quote

Rambus announced a working DDR5 DIMM in September 2017.^[9][10] On November 15, 2018, SK Hynix announced completion of its first DDR5 RAM chip; it runs at 5200 MT/s at 1.1 V.^[11] In February 2019, SK Hynix announced a 6400 MT/s chip, the highest speed specified by the preliminary DDR5 standard.^[12]

Less than a year between 4800 and 5200, so my guess is Raptor Lake settled on their IMC design specifications after Zen 4 sometime after Sept 2017 and before Nov 2018.

[MageTank]

2 hours ago, LAwLz said:

I think people need to stop focusing on the absolute highest end SKUs and generalizing product stacks based on that.

Yes, the KS version of the i9 processor uses a lot of power, especially when the power limiter is removed. But if we look at more sane workloads (like gaming), with more sane SKUs that people actually buy (like the i3, i5 and i7) then Intel isn't really behind in terms of power consumption or efficiency either:

 

Does anyone actually care that the 5600X uses 5 watts less than the 12600K when gaming?

We are talking about minuscule differences that doesn't matter. People focus too much on these ridiculous scenarios that doesn't matter and miss the truly important things.

We should focus less on

"this 700 dollar CPU uses less power than this 700 dollar CPU when calculating Pi"

and focus more on

"this 300 dollar CPU uses pretty much the same power as this other 300 dollar CPU when playing games or doing web browsing"

 

Zen4 seems to be quite a bit more power hungry than Zen3 as well. So I'd be very careful with comparing one unreleased product vs another unreleased product and declaring a winner.

I think both Zen4 and Raptor Lake seem very impressive so far. That's really impressive when taking into account that neither Zen4 nor Raptor Lake are big redesigns. Both of them are rather small changes but with big performance uplifts (probably from increased block speeds).

The biggest problem with Intel's power consumption and heat isn't even related to the processors themselves but their board partners. Intel's whitesheets since ADL gave the go-ahead to run the boards at full power (4096W limits) for an infinite duration. Board partners took this as a means of throwing caution to the wind and shipping with absurdly broken LLC input currents that inflates their power consumption significantly without doing hardly anything in terms of stability or performance. Strongly urge people to read this guide to understand LLC better: https://rog.asus.com/forum/showthread.php?127565-ASUS-MAXIMUS-Z690-EXTREME-amp-i9-12900K-GUIDE-Load-Lines-VF-Curves-Adaptive-Voltage#post854536.

 

Thermals can be tamed on these Intel processors, it just takes a fair bit of effort to do so. The reason why I am less prone to blame Intel is because this behavior isn't exhibited on their test boards, only on board partner boards. This means their implementation of boost/power settings is a bit overzealous compared to what Intel originally intended. Kind of like what we saw back in the day when people loaded DDR3 XMP on Skylake and started pushing aggressive VCCIO/VCCSA voltages, lol.

[leadeater]

20 minutes ago, Somerandomtechyboi said:

You got to be an absolute block head to fail undervolting

It's very simple to set the value to low if you don't actually know what proper values actually are, there's a difference between stabbing in the dark, copying a value someone else used and actually knowing what you are doing.

 

Firing in a vcore of a particular voltage isn't necessarily going to be stable for every application workload and certainly not when combined with other changes like core multipliers.

 

If someone has a tendency to leave their system running for very long times and acknowledge that minor issues are still stability issues then you also have to be willing to accept that it's simply not as safe to do as a power limit setting.

 

What might work for you in your applications may not work as well for someone else even with the same applications, that's the whole issue around silicon quality. I have two 3060 Ti's purchased at the same time that are the same make and model yet one has to run at decently higher fan RPM because the silicon runs hotter due to this. Without any Nvidia safety checks aka "no you're being dumb" the same undervolt on my worse 3060 Ti would not be stable, the same is true for CPUs for the same reasons.

 

20 minutes ago, Somerandomtechyboi said:

i can understand for gpu undervolt cause those use curves and you have to test all the freq/volt points on the curve

Modern CPUs, especially AMD/Zen do have voltage to frequency curves, you just can't directly adjust them as easily. They also have temperature frequency curves too, 25Mhz steps.

Edited September 12, 2022 by leadeater

[Motifator]

8 minutes ago, leadeater said:

Modern CPUs, especially AMD/Zen do have voltage to frequency curves, you just can't directly adjust them as easily. They also have temperature frequency curves too, 25Mhz steps.

With AM4isGOD, you just gotta know how to set up those Precision Boost LLC curves down BIOS.

[Somerandomtechyboi]

3 minutes ago, leadeater said:

It's very simple to set the value to low if you don't actually know what proper values actually are, there's a difference between stabbing in the dark, copying a value someone else used and actually knowing what you are doing.

You could just check stock cpu volt online and what volt you are actually running with cpuz, forgot alot of this stuff is now dynamic and turbo is a mixed bag with absurd voltages for no reason, i always disable any turbo boost speedstep since i just oc

 

Copying someone is also relatively simple to do, just copy it and if not p95 smallest ffts stable just raise it

 

5 minutes ago, leadeater said:

If someone has a tendency to leave their system running for very long times and acknowledge that minor issues are still stability issues then you also have to be willing to accept that it's simply not as safe to do as a power limit setting

Theres prime95 smallest ffts to test cpu stability and you can just add an extra 0.02-0.05v headroom above the lowest stable volt

 

7 minutes ago, leadeater said:

I have two 3060 Ti's purchased at the same time that are the same make and model yet one has to run at decently higher fan RPM because the silicon runs hotter due to this. Without any Nvidia safety checks aka "no you're being dumb" the same undervolt on my worse 3060 Ti would not be stable, the same is true for CPUs for the same reasons

There are some absolutely abysmal samples but those are quite few since most are average - slightly above average, though i forgot theres also binning in the form of heat, some just generate more than others which is something ive noticed with my golden e5400 vs all my other trash e5000 pentiums, thing runs significantly hotter than all of em but also happens to oc the best (4.5g bench stable 1.55v when the others need 1.6v+ just to windows boot 4.5g)

[leadeater]

16 minutes ago, Somerandomtechyboi said:

Theres prime95 smallest ffts to test cpu stability and you can just add an extra 0.02-0.05v headroom above the lowest stable volt

Few hours of p95 isn't the same thing, stability issues like these show up when CPUs and other PCIe devices transition between power states and signaling speeds. All p95 does is check stress loads and nothing else.

 

To be honest the most stability issues I get is not from high loads it from doing basic things with the computer and it going between idle states to various different P states. I also run other PCIe devices like 10/25Gb NICs and RAID/SAS cards which "gamer" boards in general do not play nice with compared to designed workstation and server motherboards that more strictly follow "the rules". You'd expect better from HEDT platforms but alas no, that said I have not tried any TR workstations yet so maybe those are better than the older Intel HEDT (RIP).

 

Really it comes down to the issues of it's all well and good until you're in the middle of Apex/COD etc blah blah and for whatever reason, maybe an excess of some on screen effects your system just nopes and crashes, even if extremely rare. There's no shortage of Twitch/YouTube clips of gamer rage when that type of thing happens haha.

 

Realistically if the goal/objective is to lower the peak or overall power usage of your CPU setting a power limit will achieve that and compared to an undervolt in actual gaming performance you'd be lucky to get outside of margin of error.