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

[Dr. Dro]

Clearly there is a problem if there is 6 pages of knowledgeable people bickering about what is the "Baseline". All we can do is wait for Intel's "Official" document regarding this in May. It is fun to try to figure this out ourselves though.

The baseline is already documented in the whitepaper. The Intel-recommended values are stated in section 4.4, page 98 of the data sheet.

Intel® Core™ da 13ª Geração, Intel® Core™ da 14ª Geração, processador Intel® Core™ (série 1) e (série 2), e ficha técnica do processador Intel® Xeon™® E 2400, Volume 1 de 2

Isso descreve todas as especificações divulgadas publicamente, incluindo características elétricas, mecânicas e funcionalidades de componentes, uma lista de principais recursos, uma descrição funcional e uma visão geral arquitetônica.

www.intel.com.br

For example, the i9 KS chips:

S-Processor 8+16 150 W: PL1 253 W, PL2 253 W
S-Processor 8+16 150 W, Extreme Config: PL1 320 W, PL2 320 W

Or the i5 chips:

S-Processor 6+8 125 W: PL1 125 W, PL2 181 W

The data sheet is concise and complete regarding Tau length, recommended current and wattage for all models and specifications, you just need to know how to correlate the subtype with the marketed name. S-Processor 150 W means i9 KS, S-Processor 125 W means i9 K, etc.

Again, it must be stressed that 13th and 14th Generation Core as well as the Xeon E-2400 series CPUs have the exact same denomination and stepping: "Raptor Lake-S", and they do not have any differences whatsoever between them. They have the exact same stepping and hardware revision, if you compare a i9-13900K and a i9-14900KS the sole difference between them is their clock table and silicon quality, functionally and at a technical level, they are the exact same processor unchanged.

I am glad I did not go this route and just a month ago i was playing to buy a z790 with a 13900kf for a great deal but i said no i will pass because of the heating issues and then i will need a better AIO and a better case so wasn't worth it so i am waiting on the new 15gen and maybe till 16gen will see i am very happy with my i9-10900KF 24/7 @5.1GHZ all core and Ring @4.6ghz @1.28v

There will be no further "generations" to the Core i processor line. The next will be Core Ultra series 2, and it should be radically different compared to the existing Raptor Lake chips.

 

[Tek-Check]

Clearly there is a problem if there is 6 pages of knowledgeable people bickering about what is the "Baseline"

The problem has existed for years, namely Intel refusing to tighten up the power speficifation of CPUs. It's convenient for them as this allows for flexibility in interpreting the spec by OEMs, something similar to what HDMI Forum has done with watering down the spec of HDMI 2.1...

They have been able to get away with this for the most part as soon as CPUs were stable enough and without rapid degradation over time. This time it's a bit different because people are returning top CPUs back to shops. With a permission from Intel, motherboard vendors seem to have simply overcooked it with profiles on steroids. And now they get the blame from their CPU supplier, which sounds to me like hypocrisy.

The bottom line is that a new motherboard should NEVER come with unlocked profiles on steroids as default, out-of-the-box experience. If the maximum Turno Boost is defined as 253W, that's what a default setting should always be for users to start with. Vendors can, of course, inform and educate the public that they could enable tweaked profiles with extended power range, higher voltage, etc., as an advanced option to be voluntarily enabled by PC users and not as a factory setting when you power a PC and start using it for the first time. It's nonsense what Intel has allowed OEMs to do.

 

[Solaris17]

The problem has existed for years, namely Intel refusing to tighten up the power speficifation of CPUs

This is wild. They literally post these specs on the intel ark pages. lol.

 

[user556]

Intel always had control of this. Intel is the one who decided to keep loosening the limits for each new generation. The website specs were getting more and more hazy each generation. Who knows what Intel was feeding the board makers.

Conclusion: Intel owes end users a refund for false advertising.

 

[Darmok N Jalad]

Intel always had control of this. Intel is the one who decided to keep loosening the limits for each new generation. The website specs were getting more and more hazy each generation. Who knows what Intel was feeding the board makers.

Conclusion: Intel owes end users a refund for false advertising.

Let's go all the way back to Comet Lake:

Note, when we asked Intel about why it doesn’t make these hard specifications and how we should test CPUs given that we’re somewhat enable to keep any motherboard consistent (it might change between BIOS revisions) for a pure CPU review, the response was to test a good board and a bad board. I think that on some level Intel’s engineers don’t realize how much Intel’s partners abuse the ability to set PL2 and Tau to whatever values they want.

The question was asked 4 years ago, and Intel shrugged.

 

[user556]

Let's go all the way back to Comet Lake:
The question was asked 4 years ago, and Intel shrugged.

Nice find!

 

[Crackong]

The data sheet is concise and complete regarding Tau length, recommended current and wattage for all models and specifications, you just need to know how to correlate the subtype with the marketed name. S-Processor 150 W means i9 KS, S-Processor 125 W means i9 K, etc.

Not really in a 'Exact and precise' way.

Just like what Buildzoid @Actually Hardcore Overclocking found out in this exact document.
Under 'VCCCORE DC Specifications'
Under 'Processor VCCCORE Active and Idle Mode DC Voltage and Current Specifications (S and S-Refresh Processor Line)'

The only thing being clearly specified is the Maximum value.
And in the Notes section,

Regarding 'reliability', Intel in point No.7, described 'reliability are not assured in conditions above or below Maximum/Minimum functional limits
And, While the maximum value is being specified by Intel, the minimum value is usually '--' and not clearly specified.
These reliability claims are basically useless when the so-called 'Minimum functional limits' do not exist in your specification.

Regarding the 'Recommended current' , Intel listed nothing but in point No.14, instructed the MB manufacturers to measure and set their own values, with the words 'A superior board design with a shallower AC Load Line can improve on power, performance and thermals compared to boards designed for POR impedance.'

Thus, the MB manufacturers had to figure out their own typical values, with Intel themselves encouraging 'Make more powerful VRM design then set a shallower LLC' design principle.

From Buildzoid's videos for his testing on Gigabyte and Asus 's baseline profiles,
We can see a trend of 'More LLC + lower Power limit' in these 'baseline' profiles.

Since they do solved some of the crashing problems buildzoid had.
'LLC being too low' must be one of the root cause of these problems.

However,
Since Intel does not provide minimum & typical values in these settings,
And they actively encouraged MB manufacturers to make products with shallower LLC.
Intel clearly deserves at least half of the blame.

It is not rational to put all the blame on MB manufacturers when your specification is

'Hey here is a thing , from 0-100 , figure out your own value, and we suggest you go lower'

And in reality, some of their CPUs lost the silicon lottery game, and randomly malfunctioned below 30.

 

[Dr. Dro]

Not really in a 'Exact and precise' way.

Just like what Buildzoid @Actually Hardcore Overclocking found out in this exact document.
Under 'VCCCORE DC Specifications'
Under 'Processor VCCCORE Active and Idle Mode DC Voltage and Current Specifications (S and S-Refresh Processor Line)'

The only thing being clearly specified is the Maximum value.
And in the Notes section,

Regarding 'reliability', Intel in point No.7, described 'reliability are not assured in conditions above or below Maximum/Minimum functional limits
And, While the maximum value is being specified by Intel, the minimum value is usually '--' and not clearly specified.
These reliability claims are basically useless when the so-called 'Minimum functional limits' do not exist in your specification.

Regarding the 'Recommended current' , Intel listed nothing but in point No.14, instructed the MB manufacturers to measure and set their own values, with the words 'A superior board design with a shallower AC Load Line can improve on power, performance and thermals compared to boards designed for POR impedance.'

Thus, the MB manufacturers had to figure out their own typical values, with Intel themselves encouraging 'Make more powerful VRM design then set a shallower LLC' design principle.

From Buildzoid's videos for his testing on Gigabyte and Asus 's baseline profiles,
We can see a trend of 'More LLC + lower Power limit' in these 'baseline' profiles.

Since they do solved some of the crashing problems buildzoid had.
'LLC being too low' must be one of the root cause of these problems.

However,
Since Intel does not provide minimum & typical values in these settings,
And they actively encouraged MB manufacturers to make products with shallower LLC.
Intel clearly deserves at least half of the blame.

It is not rational to put all the blame on MB manufacturers when your specification is

'Hey here is a thing , from 0-100 , figure out your own value, and we suggest you go lower'

And in reality, some of their CPUs lost the silicon lottery game, and randomly malfunctioned below 30.


View attachment 345687

I mean, these are largely disclaimers anyway, but still a nice find. I'm fortunate to have a pretty awesome board with a direct 19-phase with 105A stages, so it's particularly well behaved on my end.

 

[nguyen]

Now Intel sell extended warranty on their K/S chips, problem solved

 

[Dr. Dro]

Now Intel sell extended warranty on their K/S chips, problem solved

Until 2021 or so they actually offered a "tuning protection plan", basically an $20 insurance premium for a one-time no questions asked replacement if you burned your CPU overclocking. Supposedly discontinued due to low demand.

 

[nguyen]

Until 2021 or so they actually offered a "tuning protection plan", basically an $20 insurance premium for a one-time no questions asked replacement if you burned your CPU overclocking. Supposedly discontinued due to low demand.

I bet that plan would sell like hot cakes now

 

[Sabotaged_Enigma]

The difference is that Bulldozer sucked, and Raptor Lake doesn't.

Yes true, but I'm talking about instability here.

 

[londiste]

The interesting bit from der8auer's videos is that the real problem is undervolting. Not power limits or lack of them. CPUs can handle the power, if not they throttle etc. But what triggered the current set of problems is that boards seem to undervolt to a degree that makes CPUs unstable.

 

[Tek-Check]

This is wild. They literally post these specs on the intel ark pages. lol.

Posting specs has nothing to do with tightening up, as @Crackong explained in the post #157

 

[Zubasa]

Let's go all the way back to Comet Lake:


The question was asked 4 years ago, and Intel shrugged.

Oh it goes even further back then that. The board venders have been progressively ignoring every limit one by one and Intel turned a blind eye every time until now.
https://gamersnexus.net/guides/3389-intel-tdp-investigation-9900k-violating-turbo-duration-z390

First it was the Turbo Duration, then the PL and Voltage, and most recently the current extrusion protection on 14th gen.
GIGABYTE Releases CEP Disable Option in BIOS Updates to its Intel Z790 and B760 Motherboards | TechPowerUp
Apperently it comes from a microcode released months ago, and Intel is the only one that can release microcodes for their CPUs.
The list goes on, as long as CB R23 score goes up Intel did not care. The CPU just needed to live long enough for the reviews.

The interesting bit from der8auer's videos is that the real problem is undervolting. Not power limits or lack of them. CPUs can handle the power, if not they throttle etc. But what triggered the current set of problems is that boards seem to undervolt to a degree that makes CPUs unstable.

MSI: Nope
https://videocardz.com/newz/msi-z79...ases-intel-cpu-throttling-temperature-to-115c
Users have to go out of their way to set the "stock" values on DIY boards.
https://www.techpowerup.com/review/intel-core-14th-gen-unboxing-preview/2.html
The Asus board used was kindy provided by Intel.
Intel Core i9-14900K Review - Reaching for the Performance Crown - Test Setup | TechPowerUp

 

[Outback Bronze]

Apparently, this guy has had a fix for a while..

Me personally - Z790 Apex - 14900KS - All limits removed

 

[Assimilator]

Apparently, this guy has had a fix for a while..

Me personally - Z790 Apex - 14900KS - All limits removed

The actual fix is for Intel to stop allowing motherboard manufacturers to play fast and loose with power settings to artificially boost Intel's benchmark scores so that its CPUs don't look like the total rubbish they actually are. As opposed to stupid workarounds like these which boil down to making Intel's problem the user's problem.

Unrelated, this is why I hate most "tech" YouTubers.

 

[Daven]

Looking back at the frontpage polls from the last two years, AMD has been steadily increasing its market share of TPU readers.

Nov 2022 - AMD 60% Intel 40%
What CPU architecture do you use? | TechPowerUp Forums

Aug 2023 - AMD 70% Intel 30%
Are you using an AMD Ryzen X3D CPU with 3D V-Cache? | TechPowerUp Forums

I'm wondering if this debacle will continue to push the enthusiast DIY market towards AMD.

 

[dgianstefani]

Apparently, this guy has had a fix for a while..

Me personally - Z790 Apex - 14900KS - All limits removed

He's arrogant, but it's true that locking clocks and manually setting voltages fixes this and other issues. I recommend the same thing for Zen RAM tuning. People for example often don't know that XMP/EXPO only sets primary timings and motherboard will try to train secondaries every boot if you don't manually set them.

It's one of the few complaints I have with Zen X3D, very little you can do manually with clocks.

His attitude is a common one that people who manually overclock have towards automatic or "default" motherboard overclocking/tuning. Situations like these somewhat validate that opinion.

Looking back at the frontpage polls from the last two years, AMD has been steadily increasing its market share of TPU readers.

Nov 2022 - AMD 60% Intel 40%
What CPU architecture do you use? | TechPowerUp Forums

Aug 2023 - AMD 70% Intel 30%
Are you using an AMD Ryzen X3D CPU with 3D V-Cache? | TechPowerUp Forums

I'm wondering if this debacle will continue to push the enthusiast DIY market towards AMD.

My dude, AMD has the same issues. Remember Meltdown?

AMD just has the benefit of underdog favoritism and a bunch of people who will defend them no matter what, disregarding the fact they are a multinational corporation who should and could do better.

It's almost a meme at this point how bad the first ~year of AGESA is for a new platform.

 

[Vya Domus]

It's almost a meme at this point how bad the first ~year of AGESA is for a new platform.

It's a meme to you maybe, I haven't had any significant issues and I jumped on 7000 series pretty early on.

AMD just has the benefit of underdog favoritism and a bunch of people who will defend them no matter

AMD also has the benefit of typically not lying and that helps a lot, so even if something goes wrong they don't get as much flack for it. Maybe Intel shouldn't have that said in the past that it's actually totally within spec to have a gazillion watt power limit on their CPU, they should have kept their mouths shut and now the narrative that it was the motherboard maker's fault would have been more believable.

Unfortunately they didn't kept their mouths shut and now if you have a brain it's hard to believe they weren't at fault for this.

 

[AleXXX666]

I mean, "14th Gen" chips don't have as much as a new stepping. It's just a repackage and re-release of their existing chips to satisfy shareholders, because they didn't have Arrow Lake available on time and Meteor Lake isn't suitable to replace Raptor as a high-performance desktop processor (maxing out at around i5 level). Still, Intel did manage to improve yield to the point that the 14900K is now a mass-produced 13900KS, and the 14900KS pushed that even further, even if it's by only a few MHz or so. I didn't believe they'd be able to pull a 14900KS at all, and they indeed haven't with the "6.5 GHz" claims from early rumors, still, 6.2 with 5.9 all-core is not too bad. It's 300 MHz up from the 13900KS's average, which means they're excellent bins.



The difference is that Bulldozer sucked, and Raptor Lake doesn't.



Everyone does this. Remember how Zen 4 launched with completely broken memory training (it'd take minutes to boot), how it had a clock ceiling on the memory that wasn't because of hardware but because AGESA was flat out broken, how the Ryzen chips actually caught fire because the AGESA-level current control wasn't functional, etc.

Basically: if you want a stable platform nowadays, just don't buy latest-generation gear. "Settle" for like, a Zen 3 or Rocket Lake platform with a fully updated BIOS.



The 320 W setting is considered to be an "Extreme Power Profile" that is exclusive to the Core i9-12900KS, 13900KS and 14900KS SKUs, iirc. Otherwise you're correct.

but, 12 gen is rock solid, made pretty lot of them, however, "cheap way" - ddr4, lol, as those were "gaming" rigs, and ddr5 is "must have" in productivity tasks more...
Alder is hell faster than "rocket (lol) lake"

 

[dgianstefani]

AMD also has the benefit of typically not lying

Amazing, thanks

but, 12 gen is rock solid, made pretty lot of them, however, "cheap way" - ddr4, lol, as those were "gaming" rigs, and ddr5 is "must have" in productivity tasks more...
Alder is hell faster than "rocket (lol) lake"

What?

 

[Daven]

My dude, AMD has the same issues. Remember Meltdown?

AMD just has the benefit of underdog favoritism and a bunch of people who will defend them no matter what, disregarding the fact they are a multinational corporation who should and could do better.

It's almost a meme at this point how bad the first ~year of AGESA is for a new platform.

I guess we will get the last of the big three (AMD, Nvidia, Intel) status check with AMD earnings report this evening. Since their last quarterly report, Nvidia is up almost 100%, AMD is flat and Intel is down almost 50%. I feel like this matches how good each company’s products are faring.

 

[dgianstefani]

I guess we will get the last of the big three (AMD, Nvidia, Intel) status check with AMD earnings report this evening. Since their last quarterly report, Nvidia is up almost 100%, AMD is flat and Intel is down almost 50%. I feel like this matches how good each company’s products are faring.

IDK. Intel is in a transition period with first gen GPUs (which are surprisingly good) and revamping their foundry approach plus going disaggregated. Would not be surprised to see them coming out swinging this year.

 

[RJARRRPCGP]

I boot in about 45 seconds. Once you get to Windows it's very fast, it's the BIOS POST that takes ages.

LOL, that's the meme of my socket 775 and socket 1366 systems the BIOS takes ages, but performs well in Windows.