New to the Raptor Lake Story? - START HERE

FlyingScot

FlyingScot

Well-known member
Joined
Apr 30, 2024
Messages
1,425
"To those regulars out there, please let me know if I have missed anything or provided inaccurate information so that I can correct it. Thank you."

What do we already know about the Raptor Lake instability problems?
There’s no evidence (so far) that Alder Lake CPUs are affected by either the oxidation issue or the degradation/instability issues.

Intel had an oxidation issue with some batches, but they have not publicly stated how widespread or which CPUs could be affected.

Raptor Lake pushed frequencies way higher than Alder Lake, and voltage went up in lockstep. High voltage has been known to cause degradation.

Raptor Lake P-Cores, E-Cores and Ring share the same voltage rail. See the next point.

On-going concern surrounds the potential voltage sensitivity of the Ring, especially given the increased traffic to and from the additional E-Cores (versus Alder Lake).

High temperatures caused by high voltage (and the issue with the CPU socket locking mechanism) could have been a contributing factor in accelerated degradation.

  • —————

  • Intel had “bugs” in the microcode that allowed unexpectedly high voltage (VID) requests, possibly connected to an algorithm that tries to predict future current and therefore future voltage drop (Vdroop).

  • Intel’s 0x125 microcode includes a fix for bug#1: “eTVB ignored high temperatures when deciding to boost the preferred Single Core / Two-Core frequencies.”

  • Intel’s 0x129 microcode includes a fix for bug#2: “unnecessarily high VID requests above 1.55V”
    • * BIOS now defaults to CPU Cooler Tuning "Intel Default" and renames old presets to "MSI Performance" and "MSI Unlimited". For a complete list of BIOS settings changed by these three modes, refer to this guide. Also, C1E Support is now enabled by default.
    • ** BIOS may default to higher CPU Lite Load value (e.g. old default = 12, new default = 16). This change in default behavior can dramatically increase CPU temperatures. Manually lower the CPU Lite Load value to restore previous voltage behavior.

  • EDIT: [Sept 26, 2024]: Intel’s 0x12B microcode includes more fixes for unnecessarily high VID requests under light load or when cores transition back and forth from a state of idle. We’ll call these bug#3 and #4. See the media thread (here) for more information on what Intel hopes will be the final fix. All Raptor Lake owners are recommended to move to this version of the BIOS microcode as soon as possible.
  • * BIOS includes behavioral changes implemented with 0x129 versions.

  • —————

  • Intel’s Turbo Boost Max 3.0 and Thermal Velocity Boost (TVB) can drive up voltage (Vcore) for all areas of the CPU to satisfy just 1 - 2 cores.

  • The new MSI implementation of “Intel Defaults” appears to alter CPU Lite Load to set AC_Loadline / DC_Loadline to at least 1.1mOhms (110/110) and (for some reason) offsets some of the extra voltage by lowering Loadline Calibration Control to mode 8 (i.e. the most Vdroop). At first glance, this would mean higher voltage for low demand activities but perhaps similar voltage (compared to earlier BIOS releases) for more demanding/multi-core workloads. One 13700K owner with this setup describes his experiences here

  • For the continuing Raptor Lake story (as per media reports) follow it here.

Raptor Lake: Tuning Observations To-date
  • AC_Loadline and DC_Loadline are very important pieces of the puzzle when it comes to reducing the "base" voltage of your CPU, but they are normally hidden from the user. While it's a relatively complex topic that deserves its own guide, knowing where to find this information is the first step. You can find these instructions here. It is advisable to periodically check these values to see if they have changed, especially after updating the BIOS, changing the CPU Cooler preset or changing the CPU Lite Load setting. The AC_Loadline value has a direct impact on your CPU's voltage from idle to peak, although it scales with workload. DC_Loadline has an effect on how your CPU's VID value is reported to you via applications like HWInfo. Remember this information if and when you follow CiTay's power tuning guide linked below.

  • Lowering Core Ratios (frequency) appears to have a positive impact on balancing performance and temperatures, presumably because the auto-throttling mechanisms are too aggressive.

  • Using PL1/PL2 TDP power limits is a useful mechanism for helping you match your CPU thermal behavior to your system's cooling capabilities. See helpful guide here

  • Both CPU Lite Load and negative voltage offsets (e.g. Adaptive Offset) are effective mechanisms for reducing "base" voltage. However, they can only have a partial impact on reducing the scale and frequency of the unusual voltage spikes we see with Raptor Lake CPUs. Microcode fixes are also required.

  • Current Excursion Protection (CEP) should be renamed P.I.A. “Pain in the [***CENSORED***]” because when enabled it can often react violently to your under-volting efforts - what’s referred to as “clock-stretching." EDIT: Further testing by one of our members (Vassil_V) has discover how we can tune CPU voltage and keep CEP happy. For a comprehensive write up on these tests and conclusions, refer to this thread.

  • If you want to reduce the size of voltage spikes further than microcode or under-volting can offer you then consider disabling auto-boosting features that prioritize single/dual-threaded workloads on “preferred cores," such as Intel Turbo Boost Max 3.0.

  • As above. Thermal Velocity Boost (TVB) [All i9's] and Enhanced Thermal Velocity (eTVB) [14th gen i9] frequency boost algorithms can be manually configured to change or "effectively" deactivate their boosting behavior.

  • Knowing that degradation is accelerated by high temperature, reducing TjMax below the default 100ᵒC threshold is highly recommended.

Raptor Lake: Outstanding Research

[13th & 14th gen i9's] What positive (or undesirable) effects occur for voltage spikes, temperatures and performance when users implement OCTVB thresholds to limit the impact of Thermal Velocity Boost (TVB) and eTVB?
  • What is the best balance of AC_Loadline (CPU Lite Load), Loadline Calibration Control (LLC) and maximum current limit (IccMax) for reducing high VID requests?

  • Gain a better understanding of how Intel calculates "predictive current" - which has a direct impact on VID requests to counter anticipated Vdroop. Special focus is on the IccMax setting.

  • What is a “safe” all-core frequency for 14900K, 13900K, 14700K, 13700K in order to maximize lifespan? Higher frequencies usually means higher voltage...

  • Can C-States have a meaningful impact on average voltages?
  • Can C-States slow down the rate of degradation by increasing sleep time for CPU cores?

  • Useful 14900K owner tuning experience (WIP): here
  • Useful 13700K owner tuning experience: here (already linked earlier)
Raptor Lake: Future Tools & Solutions
  • 1) IA VR Voltage Limit - A single BIOS setting to allow a user to "cap" the maximum voltage a CPU can request via VID request. Reddit post here. [EDIT: Svet indicates that, after careful consideration, MSI will now be rolling out this feature with the 0x12B microcode BIOS releases scheduled for October 2024. See announcement.]
  • 2) Further microcode updates [EDIT: now unlikely].
  • 3) Do like the 1950's computer users did and just tune with an oscilloscope. Visit the BuildZoid channel here.
 
Last edited:
Sort by date Sort by votes
I tested, with the help of FlyingScot, a boatload of settings to tame my CPU.

i9 14900k
Msi Mag Z790 Tomahawk Max Wifi
Deepcool LT720

With unlimited PL and iccMax, at LiteLoad Auto, my CPU got 100 degrees and gave me 2207 cinebench 24 scores.

I set the following settings in the BIOS

PL1 and PL2 at 250
IccMax at 307

These settings gave me a cinebench 24 score of 1970 and cinebench r23 of 36500

I then changed these settings:
PL1 and PL2 at 200
P-core multiplier at 56 (instead of 57/60)
E-core multiplier at 41 (instead of 44)
LiteLoad Mode 7 (AC30 DC110)
Intel Turboboost Max 3.0 off
eTVB enabled

This resulted in a Cinebench 24 score of 2007 and Cinebench R23 score of 37500. Lowering the P and E cores helped RAISING the scores.
Temperatures fluctuate between 70/80 celsius for the ring ang package with 60/70 celsius per core.

When I then raised the PL1 and PL2 to 300, It ended up not raising any benchmark score, only the temperatures to 90 degrees Celsius
 
Upvote 0 Downvote
Hey FS, this is a great summary and should be read by anyone visiting this forum and looking to tune their system.

I’m considering capping the frequencies again to stop the two preferred cores from boosting on my 14700KF to 5.6ghz. Would disabling Intel Turbo Boost 3.0 achieve this? Essentially, I want to maintain the low frequency idle behaviour, but cap frequencies at 5.5ghz or lower as I really don’t need this level of performance for my use case.

The other option is locking the p-cores using an offset of -1 like my last thread on this topic.
See here: https://forum-en.msi.com/index.php?threads/locking-p-cores-through-bios-z790.400468/post-2273815

I didn’t like this method because my system behaved weirdly after setting this value - I got a BSOD which I’ve never had before. This could have been caused by anything, but still, odd that a BSOD with no error log happened after a bios setting change.

I’ve since moved back to the following settings:
LiteLoad mode 9. PLs still set to 175w. CEP and enhanced turbo off. Air cooled with Noctua nh-u12a.
Observations following 0x129 bios update:

1) Cinebench gets to 79 degrees in R23 10 min run.
2) TEKKEN 8 hit 83 degrees on one core for a brief moment. Not sure if this is a concern. Most cores reached mid-high 70s. Cooling seems good otherwise and, in the case of Cinebench finishing, quickly dropped to idle temps of low 30z
3) VCORE gets to about 1.37v max now.
4) VIDs request 1.43v max - this seems to be an increase although I’m not sure how much this matters? Also, since the microcode theoretically caps the voltage spikes, I’m assuming I shouldn’t be concerned about this.

Edit: citay considered these good readings, so I could stick with them, but lowering the frequencies is also of interest as I want to ensure longevity for this system.
 
Last edited:
Upvote 0 Downvote
crimson8814f802d4 said:
Hey FS, this is a great summary and should be read by anyone visiting this forum and looking to tune their system.

I’m considering capping the frequencies again to stop the two preferred cores from boosting on my 14700KF to 5.6ghz. Would disabling Intel Turbo Boost 3.0 achieve this? Essentially, I want to maintain the low frequency idle behaviour, but cap frequencies at 5.5ghz or lower as I really don’t need this level of performance for my use case.

The other option is locking the p-cores using an offset of -1 like my last thread on this topic.
See here: https://forum-en.msi.com/index.php?threads/locking-p-cores-through-bios-z790.400468/post-2273815

I didn’t like this method because my system behaved weirdly after setting this value - I got a BSOD which I’ve never had before. This could have been caused by anything, but still, odd that a BSOD with no error log happened after a bios setting change.

I’ve since moved back to the following settings:
LiteLoad mode 9. PLs still set to 175w. CEP and enhanced turbo off. Air cooled with Noctua nh-u12a.
Observations following 0x129 bios update:

1) Cinebench gets to 79 degrees in R23 10 min run.
2) TEKKEN 8 hit 83 degrees on one core for a brief moment. Not sure if this is a concern. Most cores reached mid-high 70s. Cooling seems good otherwise and, in the case of Cinebench finishing, quickly dropped to idle temps of low 30z
3) VCORE gets to about 1.37v max now.
4) VIDs request 1.43v max - this seems to be an increase although I’m not sure how much this matters? Also, since the microcode theoretically caps the voltage spikes, I’m assuming I shouldn’t be concerned about this.
Click to expand...
A couple of quick thoughts. Yes, you could certainly try disabling the Turbo Boost Max 3.0 and see what changes. I can't imagine that there would be any downsides, like instability. The other option is to look at what migrainefilm... did with setting manual core ratios. I've never had much luck with the other methods, e.g. Turbo Offsets, etc. I think because it might be buggy, but maybe just for earlier BIOS versions. Not sure. But I have seen others complain about this issue, too (on Z490/Z590). So it's not just me. Perhaps that's what caused your BSOD.

I'd suggest talking to, or reading, migrainefilm thread on that one. We are still very much in the experimental phase trying different things to see whether we can improve upon his current baseline, which is the manual core ratio capping that you are thinking about. The trick appears to be keeping these CPUs off the power limiters as much as possible to prevent what I believe is wasted time recovering. As you see from his recent post above, if you tune just right, you can improve performance AND reduce temps. Now, who doesn't want that!?
 
Last edited:
Upvote 0 Downvote
Definitely, thanks, I’ll give it a look.

At any rate, my current settings are actually pretty decent, especially if the new microcode stops VCORE jumping too high.
 
Upvote 0 Downvote
crimson8814f802d4 said:
Definitely, thanks, I’ll give it a look.

At any rate, my current settings are actually pretty decent, especially if the new microcode stops VCORE jumping too high.
Click to expand...
Yes...and...no. On the surface, your max observed Vcore is okay. But what we don't know is how bad those spikes are that are too quick for HWInfo. The only thing we can do is (a) definitely try out the new microcode to fix the known bugs - and then re-tune as required (e.g. AC_LL) , and (b) try even harder to get those visible Vcore peaks down even further so you have the maximum headroom for invisible spikes, ideally under 1.45V (for the invisible ones). BuildZoid is always preaching about that need to have as much headroom as possible, because even the older generations have spikes.
 
Last edited:
Upvote 0 Downvote
Back
Top