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.

 

[A///]

Keep in mind, though, that overclockers drown their CPUs in voltage and water. That Youtuber is using 1.37V fixed Vcore coupled with some LLC increase (impedance decrease).

I opted for water cooling the last year I had my pentium 4 setups, yes plural. I had several high performance p4 builds. feels like a century ago now.

 

[Timur Born]

I opted for water cooling the last year I had my pentium 4 setups, yes plural. I had several high performance p4 builds. feels like a century ago now.

I wasn't arguing against water cooling, though, but against the excess of pushing voltage and cooling.

My current undervolted 13900K OC: 58/59/55/55/56/56/55/55//43/43/45/44. No efficiency loss for these slight overclocks (power increase ~ performance increase). About 253 W for 40k+ scores in CB23 (lower is possible, but not Folding@home stable).

I recently increased Core PLL from 0.900 to 1.020 V, which allowed me to change Vcore offsets to an additional (!) -10 mV up to 56x (all-core to 4-core) and an additional -5 mV at 57x and above (single/dual-core). It also allowed me to change from using cores 2+3+6+7 to cores 0+1+4+5 for highest clocks. The E core overclocking for 4x 45x and 8x 44x is also new under this reign and needs some further stability testing.

I do fear that running the Core PLL at 13% higher voltage might damage it in the long run, even though these things seem to run in the mW range, so no power/temp increase is measurable.

 

[A///]

I wasn't arguing against water cooling, though, but against the excess of pushing voltage and cooling.

p4s ran hot from the factory even the early tamer ones. nothing to do with oc'ing.

 

[Timur Born]

Good story. Thank you.

 

[Timur Born]

Here is one for the experts: How does the screencapturing software ShareX sometimes (but repeatedly) keep my 13900K from boosting properly only by being loaded (trayicon) despite creating zero CPU load?!

 

[A///]

Here is one for the experts: How does the screencapturing software ShareX sometimes (but repeatedly) keep my 13900K from boosting properly only by being loaded (trayicon) despite creating zero CPU load?!

This would a problem the developer needs to solve not intel or microsoft.

 

[Timur Born]

This would a problem the developer needs to solve not intel or microsoft.

How does the problem happen, how does the software even achieve to make it happen? The CPU behaves as if one more core was actively loaded when it is clearly not. And then you just restart the same stress test load and it doesn't happen anymore until it randomly happens again after another test load restart.

 

[igor_kavinski]

Use process affinity to put that naughty misbehaving process on an E-core. That should teach it.

 

[A///]

How does the problem happen, how does the software even achieve to make it happen? The CPU behaves as if one more core was actively loaded when it is clearly not. And then you just restart the same stress test load and it doesn't happen anymore until it randomly happens again after another test load restart.

none of us here are the developer for that program. try their github support board.

 

[dullard]

How does the problem happen, how does the software even achieve to make it happen? The CPU behaves as if one more core was actively loaded when it is clearly not. And then you just restart the same stress test load and it doesn't happen anymore until it randomly happens again after another test load restart.

I cannot say for certain, but as someone who develops some software professionally, I can make guesses.

Every software thread when set up is given a priority. Even if the programmer does not assign a priority, the thread gets one--it will just get the default priority of Normal. If there are not enough cores to run all threads, then the thread(s) with the highest priority would run and the thread(s) with the lowest priority would be suspended until there are cores available. That priority system works quite well with CPUs of only one core type.

But now with hybrid cores, that old system is insufficient. The CPU and Windows combine to attempt to guess which core to run each thread on. That is based on priorities but also based on how long the thread is expected to take and how much slowdown there would be to switch threads from core to core to core. To fix that once and for all, the programmers are supposed to now also set a thread affinity. The thread affinity tells Windows which type of core is preferred (or even required) for each thread. Unfortunately, a lot of software does not do this step. Either it is older software, too much effort, or they don't even care to do so (this is probably not an important step in the rush to get software out the door).

Your ShareX program could probably be programmed to tell Windows to run on the E cores only. My guess is either (1) they haven't yet released a version that properly sets thread affinities to use E cores, or (2) the screen capture might actually require a high priority to guarantee that they capture the most current image. If case (2) is the reason: think of how customers would feel if the screen capture sometimes captured what they were looking at and sometimes captured several images later? Also, only one thread has access to the screen images to avoid conflicts (imagine two software programs trying to display different things full screen at the same time). I could see that ShareX is forced to run on a high priority P core to actually capture the correct image.

 

[Timur Born]

I looked deeper into it. As already suspected ShareX is not the source, just one way to reproduce it and it even happens with ShareX using 0 (zero!) CPU cycles and even being specifically suspended. It also does happen without ShareX, but the ShareX combination is a nice way to turn the issue on/off. I looked deeper into it and here is what happens:

Sometimes a single (!) CPU core fails to enter C6/8 C-state for certain combinations of software running, but no apparent "load" reason, instead it runs in 99.x% C1(E). This also happens when the affected core is specifically (core) parked via Windows 11 power profiles/modes and when Autonomous mode is disable (or even Speedshift and EIST via BIOS). This can happen to P cores or E cores, but I never saw it happen to both at once yet.

This might be a BIOS, CPU or OS scheduler problem, but seeing how the affected core is 99.x% idle in C1 and even specifically parked leaves me to suspect the BIOS and/or CPU. As a result core boosting behaves as if 1 core more was active than there is, so 2 cores being active are identified as 3 cores, which then disables 2-core boosting as a result.

Workaround: Usually this happens when 100% load is put on (like an OCCT or P95 stress test on single cores/clusters), restarting the very same load then might either fix it until a later load restart makes it happen again, or it shifts the problem from P to E core and vice versa. If a process like ShareX is used to reproduce this then unloading the process temporarily solves this, but only until the process is started again.

 

[A///]

I looked deeper into it. As already suspected ShareX is not the source, just one way to reproduce it and it even happens with ShareX using 0 (zero!) CPU cycles and even being specifically suspended. It also does happen without ShareX, but the ShareX combination is a nice way to turn the issue on/off. I looked deeper into it and here is what happens:

Sometimes a single (!) CPU core fails to enter C6/8 C-state for certain combinations of software running, but no apparent "load" reason, instead it runs in 99.x% C1(E). This also happens when the affected core is specifically (core) parked via Windows 11 power profiles/modes and when Autonomous mode is disable (or even Speedshift and EIST via BIOS). This can happen to P cores or E cores, but I never saw it happen to both at once yet.

This might be a BIOS, CPU or OS scheduler problem, but seeing how the affected core is 99.x% idle in C1 and even specifically parked leaves me to suspect the BIOS and/or CPU. As a result core boosting behaves as if 1 core more was active than there is, so 2 cores being active are identified as 3 cores, which then disables 2-core boosting as a result.

Workaround: Usually this happens when 100% load is put on (like an OCCT or P95 stress test on single cores/clusters), restarting the very same load then might either fix it until a later load restart makes it happen again, or it shifts the problem from P to E core and vice versa. If a process like ShareX is used to reproduce this then unloading the process temporarily solves this, but only until the process is started again.

do you use third party av? disable it temporarily and see if it exhibits the same behaviour.

 

[coercitiv]

We have indirect confirmation from Gigabyte that 600 series boards will support Raptor Lake Refresh, based on the notes for a recent B660M AORUS ELITE DDR4 firmware update:

1. Checksum : 196D
2. Addresses Download Assistant Vulnerabilities Reported by Eclypsium Research
3. Supports and powers up Intel next generation processor

 

[igor_kavinski]

Will this be the first time in Intel history that a chipset supports three generations of CPUs?

 

[A///]

Will this be the first time in Intel history that a chipset supports three generations of CPUs?

That I can recall

historically intel has done 2 generations per socket. amd's am4 was long lived. the upside to go with amd here is you pay for a high end board and hopefully get 3 generations out of it based on how am4 was for am5. the complaint about intel here is that while it's one gen less it's only good for 2 years because of intel's cadence. amd's cadence is longer and sees a release every 14-24 months depending on the generation. with am4 there was problems with early boards not being able to support newer processors due to a few reasons including bios storage or the aibs not giving a damn anymore. even amd tried to kill of zen 3 support for some boards.

intel could do 3 or maybe 4 generations with enough forethought put in but what's the point? You'll run into minor issues that may not be so good for the customer in the end. and typically most people aren't upgrading every year or every 2-3 years if we're talking about amd.

my personal method is if I'm going to upgrade im buying a new board because of how fast other features may become obsolete or new features being available on new boards. where intel sucks the most is if you need a board years later it's likely not available whereas there's a chance you can grab a new box for amd. rare but mobos die.

 

[ondma]

Will this be the first time in Intel history that a chipset supports three generations of CPUs?

Well, actually it is 2 "generations". RL refresh is pretty much same old/same old.

 

[A///]

Well, actually it is 2 "generations". RL refresh is pretty much same old/same old.

If you want to go with that line of thinking then 9th generation is the same as 8th generation because 9th is Coffee refresh. If you want to be very technical then 6th through 10th is the same architecture and there's nothing exciting about any of these generations outside slight cache increases, clock increases and core counts going up alongside heat. Other features too but none that stand out to the normal end user. Gens 7 through 10 had a different future ahead of them as we learned over the years but intel's internal battle with 10nm caused a lot of pain in that series progression. 11th was a major departure from Skylake for Intel and also happens to be their red headed bastard stepchild they would prefer to forget. Roll back a couple years before the wuhan flu and intel was sputtering out their lakefield hybrid processors in late 2018 which flew under most people's radar. mobile hardly gets enough love now a days. these were very cool new age processors for the time, and we know regular desktop would get hybrids a few years later. these were the more interesting processors imo after intel's 2014 push out of low tdp mobile processors that apple got to debut first in their beautiful rose gold and other toned macbooks.

 

[Timur Born]

**Efficiency per single core (1 P core = 3 E cores in on-die space) on undervolted 13900K**

CB23 (Floating-Point), single-core:
1x58 P 2299 33.3 W
1x45 P 1783 16.0 W (=22.5% less performance at less than half the power usage)
1x45 E 1258 17.7 W (best cluster)
1x43 E 1213 17.4 W (normal cluster)

7-Zip (Integer), single-core:
1x58 P 10196 28.3 W
1x45 P 8296 13.3 W (=18.5% less performance at less than half the power usage)
1x45 E 7617 16.9 W (best cluster)
1x43 E 7342 16.5 W (normal cluster)

 

[ondma]

**Efficiency per single core (1 P core = 3 E cores in on-die space) on undervolted 13900K**

CB23 (Floating-Point), single-core:
1x58 P 2299 33.3 W
1x45 P 1783 16.0 W (=22.5% less performance at less than half the power usage)
1x45 E 1258 17.7 W (best cluster)
1x43 E 1213 17.4 W (normal cluster)

7-Zip (Integer), single-core:
1x58 P 10196 28.3 W
1x45 P 8296 13.3 W (=18.5% less performance at less than half the power usage)
1x45 E 7617 16.9 W (best cluster)
1x43 E 7342 16.5 W (normal cluster)

Would be interesting to see what the perf/watt for the e cores is at lower frequency.

 

[Timur Born]

Would be interesting to see what the perf/watt for the e cores is at lower frequency.

CB23 (Floating-Point), single-core:
1x43 E 1213 17.4 W
1x33 E 913 4.3 W (=24.7% less performance at a quarter the power usage)

7-Zip (Integer), single-core:
1x43 E 7342 16.5 W
1x33 E 5756 4.3 W (=21.6% less performance at close to a quarter the power usage)

BUT: the CPU Package Power reading at this low load may be be rather incorrect. The Corsair PSU measures a difference of only 6 W instead of 13 W between both CB23 runs.

 

[Timur Born]

CB23 (Floating-Point), multi-core, CPU / PSU Out / PSU In, undervolted (!) 13900K:

55x/43x 40551 253 W 348 W 377 W
45x/33x 32257 121 W 184 W 207 W (=20.5% less performance at close to half the power usage)

 

[Timur Born]

Without Turbo Boost all cores run at 3.0 GHz, use about 62 W power (undervolted) and score 24780 pts.
For comparison: my 5900X scored 23450 at 174 W using fixed 4.6 GHz all-core (21.6k at 134 W stock, 22.2k at 132 W using CO).

 

[A///]

I looked deeper into it. As already suspected ShareX is not the source, just one way to reproduce it and it even happens with ShareX using 0 (zero!) CPU cycles and even being specifically suspended. It also does happen without ShareX, but the ShareX combination is a nice way to turn the issue on/off. I looked deeper into it and here is what happens:

Sometimes a single (!) CPU core fails to enter C6/8 C-state for certain combinations of software running, but no apparent "load" reason, instead it runs in 99.x% C1(E). This also happens when the affected core is specifically (core) parked via Windows 11 power profiles/modes and when Autonomous mode is disable (or even Speedshift and EIST via BIOS). This can happen to P cores or E cores, but I never saw it happen to both at once yet.

This might be a BIOS, CPU or OS scheduler problem, but seeing how the affected core is 99.x% idle in C1 and even specifically parked leaves me to suspect the BIOS and/or CPU. As a result core boosting behaves as if 1 core more was active than there is, so 2 cores being active are identified as 3 cores, which then disables 2-core boosting as a result.

Workaround: Usually this happens when 100% load is put on (like an OCCT or P95 stress test on single cores/clusters), restarting the very same load then might either fix it until a later load restart makes it happen again, or it shifts the problem from P to E core and vice versa. If a process like ShareX is used to reproduce this then unloading the process temporarily solves this, but only until the process is started again.

Any update on this? Worthy of kicking it up support channels online for w11 and see if it lands on the desk of someone who will look into it. if what you say is true and I have nor eason to believe you're lying here, then this is a very problematic issue if it's one of several methods that can trigger an incomplete boost. Very troubling if this is an os issue.

 

[Timur Born]

Any update on this? Worthy of kicking it up support channels online for w11 and see if it lands on the desk of someone who will look into it. if what you say is true and I have nor eason to believe you're lying here, then this is a very problematic issue if it's one of several methods that can trigger an incomplete boost. Very troubling if this is an os issue.

I am in talks with Gigabyte support and unsurprisingly they want me to check their latest unspecific beta BIOS first. Microsoft support chain is less straightforward. At least on the Intel and MS reddits there were only user upvotes, but no answers.

 

[A///]

I am in talks with Gigabyte support and unsurprisingly they want me to check their latest unspecific beta first. Microsoft support chain is less straightforward. At least on the Intel and MS reddits there were only user upvotes, but no answers.

Try reaching out to the person in charge of windows 11 that does the interval updates for beta testers?