Intel Statement on Stability Issues: "Motherboard Makers to Blame"

[BoggledBeagle]

He mentioned that all 13700K he tested were stable. They have maximal frequencies 5400/4200 MHz.

I think I may increase limits for my 14900K from 5200/4200 MHz to 5300/4200 without any fear.

 

[Dr. Dro]

He mentioned that all 13700K he tested were stable. They have maximal frequencies 5400/4200 MHz.

I think I may increase limits for my 14900K from 5200/4200 MHz to 5300/4200 without any fear.

You can run it at stock without any fear... just keep a reasonable power limit, or if you insist on extra safety, just disable TVB

 

[BoggledBeagle]

I started running reasonable frequencies because of higher efficiency, and that reason did not change. You need to limit frequencies to force the chip run efficiently, power limit alone does not help. Stability is a bonus for me. ... When I think about it now, I may leave it where it is, because the frequencies I mentioned were my highest acceptable numbers for maintaining "still acceptable, but not great" efficiency.

 

[chrcoluk]

Try to see the bigger picture.

What isn't rational thinking is saying Intel allowed changing Tjmax to 115C because they didn't want anyone using more than 100C!

Many Intel CPU's do not allow changing Tjmax (which is really Tjtarget), if anything only an offset to allow throttling at a lower temp up to about 15C lower.

Board vendors do not write BIOS, BIOS companies do usually with some Intel reference code and use of the "BIOS Writer's Guide" from Intel. The manufactures might customize the BIOS provided or ask for customization.

And IIRC Gigabyte did a similar thing with IMO stupidly high Tjmax. Again, if Intel didn't want to allow this then it would have been locked down as usual.

I am seeing the bigger picture, sadly there is so much hatred for Intel, and for some reason some kind of unexplained loyalty to ASUS and co that people are not seeing things straight. You are right they dont write the bios from scratch, but they do program their own defaults, custom OC options and that kind of stuff.

If someone kills someone by speeding, then is it the car manufacturers fault because they didnt throttle the speed or the driver?

There is a lot of FUD coming from the media right now which I think has skewed some thinking, for example HUB claiming that Intel had a stance of unlimited power is in spec.

I undervolt my CPU out of choice, thats on me if its not stable, But when I see my board out of the box on a out of spec TJMAX, to me thats on the board vendor, its their product, as simple as that really.

 

[AnonymousGuy767]

kmdkai on the chinese forum wrote, that the unstable chip may fail only after a few days of his testing and that his findings are not very relevant for a "casual user".

砸钱试稳13900K14900K。intel的BIOS baseline设定完全没解决任何问题 - 电脑讨论(新) - Chiphell - 分享与交流用户体验

This type of thing is exactly why I switched from a 14900K to a 14700. I think they're trying to drive their top-end SKUs beyond where they're actually stable so they're having to either put so much voltage that the silicon degrades and / or the thing isn't truly 24/7 stable and they rely on the vast majority of people running very little variety of stress testing or for a short duration. (or with more recent "baseline" updates: just clamp down on power limits and hope that drops down the frequency enough to be stable)

And I can tell you for a fact that there's a lot of silicon degradation happening because I had to wait a couple weeks for warranty to have 14900K's in stock.

 

[BoggledBeagle]

If someone kills someone by speeding, then is it the car manufacturers fault because they didnt throttle the speed or the driver?

Well, an automatic transmission car crawls at a walking pace with no user input, it goes faster only if the user commands it so. If intel was selling engines, the default RPM would be maximal attainable and the role of the car manufactures and the driver should be to tame such abomination?

I undervolt my CPU out of choice, thats on me if its not stable

Undervolting is not a good practise, because you decrease the voltage safety margin for stability.

BTW I was surprised to hear recently that MB manufacturers undervolt or intentionaly set the load line calibration pretty weak with Intel CPUs, which may be one of the reasons for instability.

On the other hand, what they are doing could be viewed as a desperate attempt to make the CPUs work close to the Intels INSANE SPECS, because when consumers buy a "5700 MHz CPU", they expect it to work at that frequency. If they see say 5400 MHz, they may think that their MB is broken and MB manufacturers would then need to explain all the time that the specs of the CPUs are not realistic, that they are not attainable, etc.

 

[chrcoluk]

Well, an automatic transmission car crawls at a walking pace with no user input, it goes faster only if the user commands it so. If intel was selling engines, the default RPM would be maximal attainable and the role of the car manufactures and the driver should be to tame such abomination?


Undervolting is not a good practise, because you decrease the voltage safety margin for stability.

BTW I was surprised to hear recently that MB manufacturers undervolt or intentinally set the load line calibration pretty weak with Intel CPUs, which may be one of the reasons for instability.

On the other hand, what they are doing could be viewed as a desperate attempt to make the CPUs work close to the Intels INSANE SPECS, because when consumers buy a "5700 MHz CPU", they expect it to work at that frequency. If they see say 5400 MHz, they may think that their MB is broken and MB manufacturers would then need to explain all the time that the specs of the CPUs are not realistic, that they are not attainable, etc.

The same way as in overclocking is poor practice, yet people do it (it does the same thing pushes clocks closer or even over the limit of expected v/f curve). My point was, its for the consumer to decide if they run out of spec or not, and defaults should never be tuned from spec.

Your first 3 lines make no sense, what makes you think Intel are configuring these bios's?

In terms of attainability, as far as I know, any product that has base speed and turbo clocks, turbo clocks have never ever been guaranteed, on any product. I have never brought a CPU or GPU believing this to be the case, Neither Nvidia, AMD or Intel have misled me on that. If the media give that impression to people thats on them, but I have only seen it stated on both Nvidia and Intel product sheets that only non turbo clocks are assured, the rest is dependent on workload on the chip. So if a workload causes the chip to get too hot or to use too much power, then it may not hit the maximum turbo clocks. Chips are designed to throttle. So if board vendors are as you claim trying to ensure chips hit these speeds, they doing this for their own reasons as I have never seen Intel promise it.

Googled to have something as a reference and here is the first result. Note how the document is worded.

What Is Intel® Turbo Boost Technology? - Intel

How do you use Intel® Turbo Boost Technology to speed up your processor? We’ll explain how it works.

www.intel.com

 

[BoggledBeagle]

After reading this article:

Crash landing with announcement or Ryzen panic? Intel painted the specs as they were needed | igor´sLAB

Some of my articles mature like old wine and it’s certainly not wrong to look at them from time to time, even with current references. This is especially true now with regard to the current “drama”…

www.igorslab.de

and seeing that interesting development of Intel CPU power limits, I just realised that my selected power limits of PL1=PL2=160W are by chance an average of those 125 and 188 original limits...

 

[Solid State Brain]

I think the above article doesn't consider user expectations, which are often unrealistic. It's not just a matter of power limits. The CPUs under stock settings are really only meant to reach their top frequency on 1 or at most 2 cores, not on all cores simultaneously. Motherboard "core enhancements" will often try to increase the number of cores on which the top frequency is reached. Although technically it's not strictly overclocking, it is still out-of-spec operation.

 

[qlum]

Be it motherboard makers or intel specifications being loose, it does not ultimately matter. Intel is in a position to validate and ensure good operation. The fact that this only happens with Intel and happens with all the major motherboard makers. Really it's completely the responsibility of Intel to ensure everything operates correctly.

 

[TheNightLynx]

This issue arose mainly from problems that some users were having with games using the

Imho not "some users" but "many users".

Unreal engine. The hardware involved typically was a 13700K/14700K or 13900K/14900K CPU with a high end NVIDIA card such as the 4080 Super and the 4090. The integral ultra high speed decompressor in Unreal games - Oodle - was detecting errors in decompressed data.

If this was the case than we are talking of a managed error and the relative error messages have been related to decompression not strange generic errors that have nothing to do with the real problem. The behavior and related errors were those of an hardware instability

Epic's policy with Oodle is that decompression errors are fatal errors and users experienced "crashes" where the game terminated with an Oodle error message. Epic say that the problem was caused by overclocked Intel CPUs going out of spec.

To give an example of what was happening, a gamer with an undervolted overclocked 13700K upgraded their graphics card to a 4080 Super. A game using the Unreal engine would not longer run, crashing constantly. They reduced the overclock by 100 MHz - game then runs again without errors. So this seems to be a configuration issue rather than anything else.

If underclocking solve the problem than the problem is that the CPU was running too fast. You can call configuration error but who is to blame ?

This user was able to fix the error. If the cause had been motherboard default settings and they were not familiar with BIOS settings it would have been a different matter.

If user have not touched related motherboard setting than the problem is not the user.

I find it incredible that even after the Intel memo, board makers are still making up values, rather than literally just following the Intel baseline spec.

I totally agree, it is insane that after all the caos and the fact that Intel put responsibility on them, they continue this way.

Another thing to mention, some of the crashing is due to too low voltage. Motherboard makers change the voltage curve when they fiddle with settings, and idle/low load voltages can drop below the Intel spec, enough to cause a crash. It's not as simple as "CPU uses too much power", although you can also crash from the voltage going too high.

Seems to be oscillations in minimum voltage but from what Intel declared I understand that the problem is not totally due to the motherboard setting but is something that can be related to the CPU and is still investigated.

 

[mkppo]

Ah yes, everyone demanding support for Zen 3 was just dying to retain the ability to use those masterpieces of superior AMD engineering, those incredibly high performance, cherished Excavator chips...

It wasn't only excavator, and you seem to discount the older chips like no one was using them anymore. Many motherboards even had to remove zen 1 support and I know I was using one in an older rig.

Massive props to them for adding support which, based on talks with some of them was a massive hurdle because it was a situation which wouldn't make everyone happy. A motherboard which requires a specific BIOS to support older/newer CPU's is a headache for OEM's and end users alike.

However way you spin it, their continued support for AM4 far, far surpasses what intel ever did for any socket ever. In fact they did just the opposite, year after year throughout the generations just to make more money and please motherboard makers while screwing the customers.

 

[Dr. Dro]

It wasn't only excavator, and you seem to discount the older chips like no one was using them anymore. Many motherboards even had to remove zen 1 support and I know I was using one in an older rig.

Massive props to them for adding support which, based on talks with some of them was a massive hurdle because it was a situation which wouldn't make everyone happy. A motherboard which requires a specific BIOS to support older/newer CPU's is a headache for OEM's and end users alike.

However way you spin it, their continued support for AM4 far, far surpasses what intel ever did for any socket ever. In fact they did just the opposite, year after year throughout the generations just to make more money and please motherboard makers while screwing the customers.

If you wanted to use your boring pre-Zen or dirt cheap Raven 1 Athlons (grandpa's HP 12C would be proud) you could very well just not update the BIOS. It's not like it even mattered. There's no use trying to defend AMD's actions here, they fully took advantage of the situation to see if they managed to upsell some customers by actively taking action against any motherboard makers that dared violate their directive of refusing service to 300 series. AsRock had updated the X370 Taichi and got a slap in the wrist for it.

 

[mkppo]

If you wanted to use your boring pre-Zen or dirt cheap Raven 1 Athlons (grandpa's HP 12C would be proud) you could very well just not update the BIOS. It's not like it even mattered. There's no use trying to defend AMD's actions here, they fully took advantage of the situation to see if they managed to upsell some customers by actively taking action against any motherboard makers that dared violate their directive of refusing service to 300 series. AsRock had updated the X370 Taichi and got a slap in the wrist for it.

Sigh. If you read carefully, I mentioned that Zen 1 lost support in the newer BIOSes. Many motherboards even lost support for Zen+. This has been mentioned twice to you but you seem to be stuck in your own thoughts about only pre-zen CPU's being affected which I don't really get. So let's say someone was using a 2700x, which was a perfectly fine CPU, and received no further BIOS updates because of this situation. How is that fair when there might be an important BIOS update along the line which they won't get? Do you see why it was an issue where both parties can't be happy and it's a difficult decision either way?

Secondly, you still didn't address the main point I made, which was it's a nightmare for OEM's and end users alike. Many of those motherboards needed to be flashed with a newer gen CPU because it wouldn't work out of the box. This is a situation they all want to avoid, because , well, things just don't work out of the box and most of those motherboards didn't support BIOS flashing without a CPU installed.

Thirdly, of course ASRock got a slap on the wrist, because if one manufacturer releases a BIOS that makes it lose support for older CPU's but other manufacturers don't follow suit, it's very difficult to track what motherboard supports what CPU etc. Also at that point, no decision was made on how to proceed with the issue of running out of memory and which CPU's should be discontinued support. You think if a haswell motherboard suddenly got support for skylake, intel would sit quiet?

AMD literally went through hoops and hoops because the problem they faced was legitimate. If the issue of running out of memory didn't happen you'd have a point. At this point if you're still stuck in your ways we'll agree to disagree. Just remember that some Ryzen 2xxx CPU's also lost support and it wasn't only pre Zen.

Also to get back on point, more users are experiencing crashes as we speak and it's not even only limited to Unreal engine games. I guess where there's slight instability, somethings bound to crash at some point. Lets see where this all goes

 

[Dr. Dro]

Sigh. If you read carefully, I mentioned that Zen 1 lost support in the newer BIOSes. Many motherboards even lost support for Zen+. This has been mentioned twice to you but you seem to be stuck in your own thoughts about only pre-zen CPU's being affected which I don't really get. So let's say someone was using a 2700x, which was a perfectly fine CPU, and received no further BIOS updates because of this situation. How is that fair when there might be an important BIOS update along the line which they won't get? Do you see why it was an issue where both parties can't be happy and it's a difficult decision either way?

Secondly, you still didn't address the main point I made, which was it's a nightmare for OEM's and end users alike. Many of those motherboards needed to be flashed with a newer gen CPU because it wouldn't work out of the box. This is a situation they all want to avoid, because , well, things just don't work out of the box and most of those motherboards didn't support BIOS flashing without a CPU installed.

Thirdly, of course ASRock got a slap on the wrist, because if one manufacturer releases a BIOS that makes it lose support for older CPU's but other manufacturers don't follow suit, it's very difficult to track what motherboard supports what CPU etc. Also at that point, no decision was made on how to proceed with the issue of running out of memory and which CPU's should be discontinued support. You think if a haswell motherboard suddenly got support for skylake, intel would sit quiet?

AMD literally went through hoops and hoops because the problem they faced was legitimate. If the issue of running out of memory didn't happen you'd have a point. At this point if you're still stuck in your ways we'll agree to disagree. Just remember that some Ryzen 2xxx CPU's also lost support and it wasn't only pre Zen.

Also to get back on point, more users are experiencing crashes as we speak and it's not even only limited to Unreal engine games. I guess where there's slight instability, somethings bound to crash at some point. Lets see where this all goes

Nightmare? Literally not our problem. AMD and the manufacturers should have dealt with it, as they ultimately did. To the best of my knowledge, that early beta BIOS AsRock released at the time did not remove support for any older CPUs, but I no longer have a copy of it, so I can't put it through the microcode tool. Even if pre-Zen was removed we were literally asking for it - i'm sure every buyer of a 5800X or 5950X at the time was deeply concerned about Bristol support, we totally didn't want to make keychains out of them as soon as possible... Really, enough with the excuses, stop thinking that AMD is a small company with limited resources doing us favors, they're a multi-billion-dollar international conglomerate and we as customers must hold them to their word and the highest standards.

To clarify, Zen 1 (Summit Ridge) works on current BIOSes - if the motherboard manufacturer didn't screw up the implementation (again, not the customer's problem). Specifically, it's still supported on the BIOS of the motherboard I had (ASUS Crosshair 6 Hero), one of the supposedly "incompatible" 16 MB BIOS motherboards. No features were removed, no graphics were downgraded, it received practically every feature that was added over time, it's still being updated and has space for future updates, even today.

I'm even going to point out, CPUID 800F00 - that odd microcode that hadn't been maintained since 2016, is actually a pre-release stepping A0 Summit Ridge early engineering sample of the Ryzen 7 1800X. It is still supported even in the latest BIOS revision for this motherboard. A20F12, for example, is Vermeer X3D (used for the 56/7/800X3D), while A50F00 is Cezanne. Every processor of the Zen era for socket AM4 works on this supposedly incompatible motherboard.

AMD did in fact lie, they did attempt to excuse themselves, and they could have handled the situation far more gracefully. It's not a mistake they'll repeat... or is it? Only time will tell. Looking forward to the people on Reddit telling people with X670E motherboards that they should upgrade their early generation garbage "because early designs sucked and you're just being stingy."

 

[mkppo]

Nightmare? Literally not our problem. AMD and the manufacturers should have dealt with it, as they ultimately did. To the best of my knowledge, that early beta BIOS AsRock released at the time did not remove support for any older CPUs, but I no longer have a copy of it, so I can't put it through the microcode tool. Even if pre-Zen was removed we were literally asking for it - i'm sure every buyer of a 5800X or 5950X at the time was deeply concerned about Bristol support, we totally didn't want to make keychains out of them as soon as possible... Really, enough with the excuses, stop thinking that AMD is a small company with limited resources doing us favors, they're a multi-billion-dollar international conglomerate and we as customers must hold them to their word and the highest standards.

To clarify, Zen 1 (Summit Ridge) works on current BIOSes - if the motherboard manufacturer didn't screw up the implementation (again, not the customer's problem). Specifically, it's still supported on the BIOS of the motherboard I had (ASUS Crosshair 6 Hero), one of the supposedly "incompatible" 16 MB BIOS motherboards. No features were removed, no graphics were downgraded, it received practically every feature that was added over time, it's still being updated and has space for future updates, even today.

View attachment 346274

I'm even going to point out, CPUID 800F00 - that odd microcode that hadn't been maintained since 2016, is actually a pre-release stepping A0 Summit Ridge early engineering sample of the Ryzen 7 1800X. It is still supported even in the latest BIOS revision for this motherboard. A20F12, for example, is Vermeer X3D (used for the 56/7/800X3D), while A50F00 is Cezanne. Every processor of the Zen era for socket AM4 works on this supposedly incompatible motherboard.

AMD did in fact lie, they did attempt to excuse themselves, and they could have handled the situation far more gracefully. It's not a mistake they'll repeat... or is it? Only time will tell. Looking forward to the people on Reddit telling people with X670E motherboards that they should upgrade their early generation garbage "because early designs sucked and you're just being stingy."

Funny how we are talking, at present, about both motherboard manufacturers and intel being lazy in this thread but you point out a situation from many years ago when the mobo manufacturers were lazier still. Yes we know all 16MB motherboards support at least zen1 now, this is common knowledge but you're talking about something which happened close to 4 years ago there's no point in posting proof of older CPU support today. BIOS sizes were way larger at the time because of bloat and at the time of release the zen/zen+ CPU's not being supported was 100% real. Sure AMD should have had better control over the BIOS/mobo manufacturers at the time but that, and your facts and predictions at the end of the post are a discussion for another topic.

At least that whole situation arose because they didn't add/remove a pin to make it physically impossible to upgrade to a new CPU. I know which situation i'd prefer more

 

[BoggledBeagle]

Intel wants the baseline profile to become the default profile by the end of May, article from benchlife.info:

Intel Baseline Profile can't solve the problem of 13th and 14th Gen processors, but it is a bigger problem for motherboard manufacturers and users - BenchLife.info

Intel's Baseline Profile can't solve the problem of 13th and 14th Gen processors, but it is a bigger problem for motherboard manufacturers and consumers​

By Chris.L on 2024-05-05 in Hardware Components, Processors

Is it the problem of the motherboard factory, or the problem of the Intel processor itself, in fact...

The problems with Intel 13th and 14th Gen desktop processors on the Raptor Lake-S and Raptor Lake-S Refresh platforms have a tendency to get bigger and bigger, and at this stage, we only feel that Intel has thrown all the problems to the motherboard factory, and has not suspected that it is the problem of its own processors.

Thinking about it from another angle, the 13th and 14th belong to the same architecture, and if there is a real problem, it should have erupted a long time ago, rather than a large-scale problem that has only occurred now, because it is not unreasonable for Intel to think that the BIOS settings of the motherboard factory are too aggressive.

However, Intel's approach to using the Baseline Profile is actually incomprehensible to motherboard manufacturers, and the Intel Baseline Profile will be renamed to Intel Default Settings in the future, but this is another story, and it doesn't have any impact, it's just a name anyway.

Intel® requests system and motherboard manufacturers to provide end users with a default BIOS profile that matches Intel®
recommended settings.
• Suggested profile name “Intel Default Settings”.
• Intel requests customers to implement the “Intel Default Settings” profile as the BIOS default profile by May 31, 2024.

Intel Default Settings is tentatively scheduled to go live on May 31, 2024.

At this stage, Intel Baseline Profile is doing to reduce the PL value. Taking the PL2 = 253W of the Intel Corei9-14900K, you can make the Turbo Boost reach the operating conditions of 6.0GHz, but after applying the Intel Baseline Profile, the PL2 of the Intel Core i9-14900K will be set to 188W The Baseline Profile is all about making the Intel Core i9-14900K's PL2 = 188W.

What's funnier is that Intel's suggestion includes the phrase "For best performance use PL1 = 253W". This means that PL1 = PL2 = 253W, but unfortunately the Intel Baseline Profile will completely limit these settings.

PBP (Processor Base Power) is still at 125W, which has not changed, and we know that the CPU will not run at 125W at all.

On the other hand, according to the Performance setting of Intel specifications, the maximum values of Iccmax and Iccmax.app are 307A and 245A respectively, except for the PL2 = 253W of Intel Corei9-14900K, but after applying the Intel Baseline Profile, the maximum values are reduced to 249A and 200A, which is quite a limit to Intel Core i9-14900K Overall processor performance.

	Baseline	Performance	Extreme
Processor Base Power	125W	125W	125W
Iccmax	249A	307A	400A
Iccmax.app	200A	245A	320A
PL1	125W	125W	253W
PL2	188W	253W	253W
PL4	293W	380W	380W
iPL2	160A	200A	200A

As mentioned earlier, Intel Z790 chip motherboards have high-end, mid-range and entry-level motherboards, which are very different, but in the mid-range and high-end parts, the motherboard manufacturers use higher designs in the design of CPU power supply modules than recommended by Intel, and this time the 13th and 14th processors are not only downgraded to the same level as the non-"K" series processors due to the Intel Baseline Profile, but also make the motherboard manufacturers' motherboards useless.

After all, in addition to Performance and Baseline, the plan given by Intel also has an Extreme setting. In the Extreme Profile, there will be the PL1 = PL2 = 253W setting we mentioned earlier, which is also designed for high-end motherboards, after all, the "K" series processors can get better performance with high-end motherboards, which is one of the things that consumers expect.

The Iccmax and Iccmax.app in the Extreme Profile are 400A and 320A, respectively.

After the Intel Baseline Profile is applied, the Intel Thermal Velocity Boost Frequency time of the Intel Core i9-14900K, that is, the time that can reach 6.0GHz, will decrease, and the overall will be lower than 3s, assuming that we look at the number 3s, the Turbo Boost of the Intel Core i9-14900K It's like waste, useless.

To paraphrase Taiwanese folks, Intel Baseline Profile turns the Intel Core i9-14900K into a three-second man!

The 13th vs. 14th Intel Baseline Profile didn't actually prevent the problem from continuing to happen, but rather masked it.

In the course of our conversations with some motherboard manufacturers, we also learned that each motherboard manufacturer actually introduced Extreme Profile from the 9th Gen (Coffee Lake) platform, and all of them were unharmed until the 13th and 14th Core of the Raaptor Lake generation.

In fact, we also know that the settings of each motherboard manufacturer for high-end motherboards are higher than those of Intel Extreme Profile.

At the same time, there are also motherboard factories for Intel's instructions are not clear and distressed (the baby is bitter, but the baby can only do it in his heart), after all, it is not only Intel who faces consumers on the front line, but employees in various regions of the motherboard, and these practitioners have no way to solve the problems faced by consumers in the first time.

As for whether the problem is whether Intel itself has relaxed the acceptance criteria for commercially available processors, this can only be said to be speculation, after all, most of the processors in the hands of motherboard manufacturers are not commercially available, and most of the problems encountered are commercial versions, so that motherboard manufacturers are quite powerless when encountering feedback.

Fortunately, there is no such thing as 15th Gen in the future, and the new Intel Core Ultra 2 series (Arrow Lake-S) should allow Intel to change its life.

Intel is expected to issue an official statement before the end of May, and then see what different the x86 processor leader that has recently focused on Foundry will say.

​

PBP (Processor Base Power) is still at 125W, which has not changed, and we know that the CPU will not run at 125W at all.

To paraphrase Taiwanese folks, Intel Baseline Profile turns the Intel Core i9-14900K into a three-second man!

I am not sure what a "three-second man" means, but the CPU at 125W certainly runs and with a heavy load runs very efficiently. I am not sure, if Intel in the beginning did not make a mistake (because they have not planned to use that limit anyway), but 125W is really a very low power for a 24 core CPU!

 

[Solid State Brain]

Buildzoid made a video about how to fix the issue and what is its root cause (this one is the short version):

Basically, motherboards like his own Aorus are actually undervolting the CPU under load, as well as running with no current limits with default settings—a recipe for crashes. Some also disable TVB, which may lead to higher idle voltages (although this isn't really a problem).

Setting a shallow LLC, configuring IccMax to a sane level (Intel spec instead of unlimited), enabling CEP (current excursion protection, which is supposed to be enabled anyway—mitigates instabilities due to insufficient voltages at the cost of performance) increasing AC Loadline to the minimum that doesn't cause CEP to trigger, solves all the issues. There are also other tips for people concerned about low-load/idle voltages, but they aren't really related with the instability problems.

 

[BoggledBeagle]

Buildzoid made a video about how to fix the issue and what is its root cause (this one is the short version)

Do not forget that some people report that in the beginning the CPUs were stable and they started acting up after some time.

So there is a problem of degradation, caused by electromigration, which is greatly worsened by higher electric current density and temperature. Intel baseline profile fixes high current density and overheating.

Then there is a second problem, that the MB does not supply the CPU what it momentarilly needs for stable operation.

Intel baseline profile prevents quick degradation and also the second problem, because under that settings the CPUs will not have high demand for current, the power delivery circuits on the motherboard will not be strained too much, etc.

What Buildzoid does is stabilising the CPU, which is still stressed by the high power, currents and temperatures. Stabilising it with higher voltage and causing even quicker degradation.

 

[Solid State Brain]

Degradation—are there credible reports or is it just really users who after testing/toying around with their CPUs in benchmarks they begin to observe instabilities in real-world loads?

In either case, both the first and second problem would be due to incorrect/out-of-spec defaults from motherboard manufacturers.

Buildzoid is simply configuring the Intel-spec IccMax/current limit (which you aren't supposed to exceed anyway) and an AC loadline (i.e. indirectly, voltages) high enough that it doesn't cause CEP (current excursion protection) to trigger, after enabling it. Just these are enough to avoid seeing crashes and operating the CPU within specifications.

What Buildzoid does is stabilising the CPU, which is still stressed by the high power, currents and temperatures. Stabilising it with higher voltage and causing even quicker degradation.

No, that's not what he's doing. The CPU was being undervolted in the first place by the motherboard (under default settings), operating below its factory VF curve.
That's why it was crashing.

 

[BoggledBeagle]

No, that's not what he's doing. The CPU was being undervolted in the first place by the motherboard (under default settings), operating below its factory VF curve.
That's why it was crashing.

All that has been happening so far was with undervolted CPUs and lax load line calibration, compared to that the voltage is higher and degradation quicker.

I am curious, what specification will Intel settle on in the end, because the "baseline settings" are needlessly constraining.

Igor writes here:

Intel plays with the name and the data: The “Intel Baseline Profile” becomes “Intel Default Settings” | igor´sLAB

The problem with the Intel desktop processors of the 13th and 14th generation, i.e. Raptor Lake-S and Raptor Lake-S-Refresh, is not getting any smaller and the bad news does not stop. At this point…

www.igorslab.de

Where is the ACDC load line?​

In the currently planned implementation, the so-called ACDC load line (sometimes also referred to as AC/DC load line) remains unclear for the time being. In Intel processors, however, this is a very important setting that influences both stability and power consumption and plays a major role in the voltage regulation of the processor, especially in connection with the power supply and power consumption under load. This setting is mainly used in BIOS or UEFI settings of Intel boards and it helps to regulate the voltage supplied to the processor when it is working under load. It influences how much the voltage drops when the load increases (voltage droop). This is important to ensure the stability and performance of the processor and to prevent excessive voltages from damaging the processor. Here again as a reminder:

• AC Loadline
  Affects the adjustment of the voltage based on the CPU load in a scenario where the processor is powered by the main power supply (AC). It helps to increase the voltage when the CPU load increases to compensate for the voltage drop.
• DC Loadline
  This setting is similar, but refers to the power supply coming from the DC source (usually the motherboard’s voltage regulator). It controls how the voltage adjusts as the load varies to optimize efficiency and stability under different operating conditions.

These settings are particularly important for overclockers who want to get the maximum performance out of their system without compromising the hardware. They allow finer control over the processor’s behavior under different load conditions, which can lead to better performance stability and potentially higher overclocking results. So it will be our job to find out exactly. If Intel changes or prescribes anything at all. However, this will only work once we have received the appropriate firmware. So far, however, this information is not available or could not be validated from at least two sources.

It does not seem that defining / changing LLC settings requirements has been considered by Intel so far.

 

[Solid State Brain]

All that has been happening so far was with undervolted CPUs and lax load line calibration, compared to that the voltage is higher and degradation quicker.

It's obvious that long-term degradation (which will happen regardless) will be higher at higher voltages.

On the other hand, operation outside the factory voltage-frequency curve is considered out-of-specification, so users shouldn't be surprised if instabilities arise if operating voltages are below that.

Here are the points of a typical VF curve for an average i9-14900K as reported by an Asus motherboard:

If load voltage doesn't reach those voltages, then the CPU is being undervolted (and out of spec).

CEP (Current Excursion Protection), which Intel recommends enabling, prevents operation under (significantly) undervolted conditions. CEP usually defaults to enabled with non-k CPUs and B-series motherboards.

As a practical example, on my 12700k I can only undervolt by about 15-20 mV from the internally-programmed voltage of 1.22V for its 4.7 GHz all-core frequency before CEP starts throttling the CPU (making it lose performance in benchmarks like Cinebench).

It does not seem that defining / changing LLC settings requirements have been considered by Intel so far.

It's not strictly necessary nor mandated, but a shallower (less droopy) LLC helps achieving lower low-load voltages, mainly due to the way the AC loadline correction works. Even Buildzoid observed in his video that a moderately high LLC with a low AC LL appeared to show lower idle/low-load voltages than a low (droopy) LLC and high AC LL to compensate that.

Then, for what it's worth, Intel mentions in their spec sheet:

 

[BoggledBeagle]

... operation outside the factory voltage-frequency curve is considered out-of-specification, so users shouldn't be surprised if instabilities arise if operating voltages are below that.

View attachment 346429

If load voltage doesn't reach those voltages, then the CPU is being undervolted (and out of spec).

Well, according to this information the MB manufacturers may have been running the CPUs "out of spec".

 

[Solid State Brain]

Yes, they have been running the CPUs out of spec for quite a while. I've written about that a few days ago, citing an example I personally observed in the past on my own hardware.

Intel Statement on Stability Issues: "Motherboard Makers to Blame"

Posting specs has nothing to do with tightening up, as @Crackong explained in the post #157 You think they would give us the user what they do on ark and NOT give the oem the values when they give they idk tell them how to use the CPUs in a new chipset? Thats a new level of denial. AMD board...

www.techpowerup.com

 

[AnonymousGuy767]

Degradation—are there credible reports or is it just really users who after testing/toying around with their CPUs in benchmarks they begin to observe instabilities in real-world loads?

Hi, yes, it's me. Seriously though, the *only* setting I had touched was XMP and maxing out the power limits and after about 6 months I got the exact Fortnite crashing stuff called out. No settings changed and I'm on a extremely high end watercooling setup.

So I'm telling you: there's a degradation problem here. Intel wants to sell a 6Ghz part that needs 1.6V to make that frequency happen, but that voltage kills the silicon. So their "default" and "baseline" profiles are likely just clamping down the power limits as an indirect way of moving left on the VF curve. No one noticed that Intel called out the "Ultimate Performance" windows power plan. They called it out because that keeps your CPU operating at the higher frequency bins (and hence...voltage) even at idle!