Consistent Instability with i9-13900kf

[Anaryl]

Background:

I purchased this CPU as part of a system I built myself just around a year ago. I was particular interested in the high core clock speeds alongside the impressive stats and high core count.

I have been building my own Intel machines since my 486DX2 50Hz, through Pentium 3s and 4s, Core2Duo E6600, 3830k, 5820k, 8700k and now the 13900kf. I have been choosing Intel for the past 30 years.

The system I chose to build in this instance was a 913900kf, an Asus B660m Motherboard, 32gb of DDR5 Corsair 5600, a 3070ti; several nvmes and ssds from previous rigs, cooled by a 280mm cooler (that I had to buy thinking it would resolve my issues with the chip) and powered by a corsair gold RM850

Issue:

System is completely unusable on stock settings. System might post, or it might decide to reboot or hang in BIOS or Windows load.

System will only boot if I sync all core ratios drastically - 1 - 1.5Ghz below advertised performance. Heart initially seemed a factor but issue persisted even after lowering clocks and temps with additional cooler.

Present configuration is E-Cores: 34 Allcore ratio - Works on Auto but is very unstable, randomly freezing or rebooting. 
PCores: 48, 48, 46, 44, 44, 42, 42, 42. The PCores never reach a single core frequency, ever. Min cores active seems to be two at least - as repeatedly verified by watching HWMonitor and testing. 

However, the moment I put a 5 into even the single core active count, the system begins flipping out. Don't get me wrong it is still incredibly unstable at my present clocks. I've set P1 and P2 to 125 and 265W respectively. Turned off APE (guess that's MCE by another name?) some improvement in stability. Monitored in XTU, no instances of thermal throttling, motherboard VR, Powerlimit or EDP or PMAX throttling. Thermals indeed seem pretty tame with the system idling at 36 and really only pushing 78C under load (which is usually high demand gamedev work, e.g. processing texture or light maps).

The closer I try to even push within 1Ghz of advertised chip speeds, the more it BSODs or reboots - and what a colourful cast of BSODs it is: KMODE Exception Not Handled, System_Thread_exception not handled, IRQL not less or equal, and a few others I don't recall off the top of my head.

Remediation:

I have reinstalled all the hardware and drivers repeatedly. I've gone from Win 10 to Win 11 to resolve the issue - no luck.

I have trialled and error a number of different configurations and had no real improvement in stability. I've taken nearly every device in and out.

I have updated all the motherboard and hardware firmware, repeatedly checked for bad drivers with no improvement. I tried to get help from Intel "partners" and reddit - dismal and worse than no help at all.

However opening up reliability monitor (find results attached) I can clearly see the error 'LKD_0x124_7_GenuineIntel_FIRMWARE' generated. Well that certainly seems to fit the bill of a whole bunch of other errors indicating some kind of driver conflict or crash - it certainly seems to fit the bill of a whole bunch of other users issues which one can see was handwaved/handballed off as "Maybe the GPU drivers I dunno?"; because no, it's not. I've put multiple cards in the thing, switched between studio and game-ready drivers; and updated the drivers maybe 20 times since purchasing it - with no discernible relationship or even a correlation.

I can trigger the instability just by raising one active core count on the PCores from 46 to 49 ratio - with varying degrees of instability. Yet, even when I can get the system to boot into windows under these "higher configurations" - I see no massive increase in temps or power consumption.

In these configurations it will just blip - even at idle or running spotify - either throwing a BSOD or just rebooting and failing to load. If temperatures were quite high before the reboot, there's no evidence of this as BIOS shows the CPU at a sunny 36C. 

If it were any of the other devices on the machine, I would expect that this instability wouldn't increase as a I slowly approach 4.6Ghz. I'v ebeen running the machine at 3.8/3.6 all cores since I got it. That's 2Ghz lower than the advertised turbo boost speed. My max stable coreclock is 4.8 (maaybe) - 1ghz below advertised top turbo speeds.

I've never lost the silicon lottery that bad. If your chips are not binning within nearly half of the advertised full clocks, then maybe you shouldn't be selling them.

What really burns my bridges though is the kind of people Intel choose to represent them - and give free stuff too. Here I am, with a faulty 750 dollar processor, and I cannot even get more than a form response. I could send it all off for warranty - but then what would I use to live? Which component? I can't go six weeks without a machine.



Ah but Intel is more than happy for its "partners" to hand out these beautifully gift wrapped boxes of stuff to partners who will in turn give it to some "poor" streamer who only has 300 subs or something (so is given what 1500USD for free each month). Every couple of weeks these guys have a new box - with very fancy packaging - from you.

You think by handing out free stuff to these influencers, you're going to win over customers?

What about the ones you already have? Most customers are not influenced by what a particular influencer says - they are just looking for confirmation that their choice is the right one. Intel.ark does that just as effectively as YouTubers or Twitch streamers.

Customers are influenced by their experience with a product and company. And primates are not known to take inequal treatment well. These people you give fancy boxes of stuff to - they're just panhandlers. Yet Intel cannot afford to even give me as a nicer box for my i9-13900k - probably because of all the GPU and CPUs it ships to dodgy streamers like Pauls Hardware, LTT and Artesian Gaming.

To reiterate, this CPU appears to be the worst chip I've bought from Intel since my e6600 - which was a great chip but clocked slightly lower at 3.6. In terms of silicon lottery - it's easily the worst. Never had a bad chip from anyone, AMD or Intel. My first PC was a 486 - that was 30 years ago now.

I can read about all the other people struggling with these chips - and see them getting blown off/handballed to their OEM consistently - when I can tell its the chip. It's outstandingly bad service - and you know what is the number 1 reason for customer churn in commerce? It's bad service - people will pay more to get away from bad service - people will pay more for good service.

After 30 years without so much as a thank you - whilst Intel's marketing department gives away literal millions in hardware to influencers whom by definition do not pay for any of it - whilst struggling with a chip that appears to just be a complete lemon. It's more than one can bear to be honest.

And having a legal notice at the bottom about how its due company policy to wriggle out of any implied committment to either their hardware or their customers isn't a good look either.

[Alberto_R_Intel]

Anaryl, Thank you for posting in the Intel® Communities Support.

We are sorry to hear about this issue and we will be more than glad to assist you with this matter. 

First, we would like to apologize for any inconvenience you might have while using your system. We would also like to inform you that for Intel® all the feedback and suggestions provided by all of our clients are very important. For that reason, I will send your remarks and observations to the proper department for them to be aware of your comments in order to keep improving the customer's experience while using Intel® products.

In order for us to better assist you we just wanted to check:

When did you purchase the Intel® processor?

By any chance, do you have the option to do a swap test and use a different processor with your board? 

Any questions, please let me know.

Regards,

Albert R.

Intel Customer Support Technician

[Anaryl]

Hi there Albert!

I do apologise for being rather grumpy early this morning, I ate a bad thing, and was deeply frustrated. I stand by it though.

I believe I purchased the processor approximately 1 year ago.

I'm not sure how that would have any impact outside when the warranty is due to expire.

I can't do a swap test with a 1700 board no, I don't have the money to afford a second CPU capable of working with this board and hardware. Socket 1700 is just 12th and 13th gen yes?

My last CPU was an 8700k, so no it's simply not feasible. Whilst I appreciate this is fairly standard tech support methodology - you should also be aware you're also kinda asking me if I have another Ferrari lying around. I do not.

I have however continued testing. I decided to perform another memtest - no results - but upon quitting that - it had reverted all my settings to OC Tuner I on the Asus board. This absolutely fixed nothing but broke everything in a new set of ways. Naturally, it BSOD's with paged faults - and this was due to an obscene set of timings. Easily remedied by changing that profile out but I did note that the PC would now accept being loaded with PCores set to Auto. However, despite setting it to Auto on PCores - it's still adhering to the old ratios (48 down wards to 42). It hasn't arbitrarily rebooted yet - but that's standard behaviour.

However, some settings have changed, System Agent Voltage was changed from 1.785 or something to 1.280 - I had my suspicions about this value but was not able to find any valid reference material for it. Likewise, IccMax was dropped from 511A to 370A (both of those seem like a lot).

At these settings, the system takes a ridiculously long time to POST. It also offers some really strange behaviour with onboard devices - for example; the audiodg process died when changing the power plan to high performance. Turning off all the ECores caused the system to hang within minutes of loading into Windows.
XTU Reports no throttling of any description.

I also committed to another round of driver installs and updates but saw and installed nothing that made an appreciable impact. In one instance I saw Nahimic implicated in a crash - but having fought, slain and resurrected that beast, I had no appreciable imppact on stability or audio symptoms (for example the audio service restarting several times before Windows login responds to Mouse and Key Input).

Regards

[Alberto_R_Intel]

Hi Anaryl, Thank you very much for your response. 

No problem at all, we do understand that with computers it can be frustrating sometimes. We also comprehend that the option to do a swap test is not available.

For this scenario, based on the fact that you are using an ASUS board, please try the following steps:

-Run AVX2 test on Intel® XTU, to confirm if AVX2 test failure is seen.

-If yes, change BIOS settings:

• Advanced (F7)
• AI Tweaker
  • SVID Behavior: Change to “Intel’s Fail Safe”

-Reboot the Operating System and run XTU AVX2 test again. Check if now the unit passed the test. 

-Test the computer and verify if the issue happens again.

Once you get the chance, please let us know the results.

Regards,

Albert R.

Intel Customer Support Technician

[Anaryl]

Hi there Albert, thanks so much for you help, I think warranty maybe the thing to do. But I am hoping there is some obscure setting I can unlock or that I've been an idiot all along or it's the motherboard or RAM. I was really excited about this chip and still occassionally get my hopes up


It doesn't appear to have failed. I only ran it for five minutes, as I am unsure whether the settings I have to get it stable are actually safe under load. It didn't fail apparently?

At what settings should I test it? How long and with either auto or static defaults? I do not believe the machine even posts under factory or optimised defaults; and I have to race into the BIOS to downclock the chip so it boots.

I am a bit uncomfortable with running ultra long tests both physically - it significantly heats up the room; and also I don't want to do anything to it that might ruin an otherwise good processor. My hope is that there is some obscure setting like yours suggested that might be holding the CPU back in terms of power or stability; or that the problems lie elsewhere, with a more replaceable part.

Should I continue after this test and change those settings in the BIOS anyway and check the results? 

I haven't been idle and have managed to collect a number of memory and minidumps but seem to have some trouble with posting them - as they jsut appear to be permanently uploading; I'll try from the email reply functions

Some of these are the result of Driver Verifier and changing or restoring various settings and having to go through a failed boot/BSOD loop but I seem to see a consistent issue with IP_MISALIGNED_GenuineIntel.sys or some such with a fair fairly consist incorrect address of um 419680 - or something like that I'll check momentarily. The system is presently very unstable, certain applications cause the machine just to blip and reboot - or occasionally it just hangs. So if I don't come back it is presumably because I was eaten by wild pumas...

[Anaryl]

Okay so here are some dumps from this morning.

The system was also repeatedly rebooting when idle whilst I was asleep.

Ran ten minutes

have included a bunch of minidumps below

[Anaryl]

Okay so it doesn't like the minidumps? 

I'm not sure how to upload those as an attachment 

https://app.box.com/s/olx69w58bmsf8s4byx5tt8q1v60a7tk8 I created a folder where they are now. Not sure if that's acceptable or necessary for you guys.

[Anaryl]

I just ran a test after removing the power limits. Intel XTU likes to default to unlimited or 330w pl1 and 2 when you click that power optimisation button :S.

Aside from thermal throttling, there's no power limit throttling and throttling at all core active min ratio makes sense. Pretty much everything there is staying to its committed clock rate - but if I try and push this clock rate on even one active processor any higher - it just don't boot. The machine will continually just restart (blip) at random times (except in startup repair and bios). If it is running at say 5500 in BIOS temps are hot but not super hot (50 compared to 34 for idle at these settings).

[Anaryl]

Just following up with you Albert following on from this request.

As far as I can tell there is no option in the BIOS that lets me define SVID behaviour. I have an option to set a voltage and an offset if I change them from auto to manual mode.

It appears that the processor can pass the test clocked up to about 42x by Core ratio. For reference I have had to return PCore clocks to static values - as at around 47 - 48 all core, the system becomes unstable and reboots. Prior to that, the OS starts falling apart with various tasks starting completely fail, e.g. XTU, sfc and any another background threads.

Current Config is PL1 and 2 unlimited, ECores Set to Auto, and Pcores set from 48 down to 42 by two core intervals (e.g 48, 48, 46, 46, 44, 44, 42, 42). 

The system seems to display no strange behaviour when stress testing the CPU at these higher performances; the AVX2 instruction + CPU stress test shows that the machine leaves a lot of PCore performance on the table given thermal headroom - various bursts of 47 - 48 on a couple of cores do not drastically impact TDP or temperature.

Also CEP/UVP is set to auto - I can see it locks up a lot of options in XTU; there's also guardband scale set to 1.

But from the various BSODs I get, it keeps saying driver issue, and I have corrupt sys files after these failed boots, might be a driver issue? I dunno. All of that keeps pointing back at some Intel thing.

What should I investigate from here?

[Alberto_R_Intel]

Hi Anaryl, You are very welcome. Thank you very much for sharing those details and providing the pictures.

Based on the fact that the processor passed the XTU AVX2 test, indicates that the settings on the BIOS are correct. From our side, there are no more suggestions to try in the BIOS options of the board.

So, as you mentioned, unfortunately, it seems there might be a hardware problem with the processor itself. In that case, please get in contact directly with your local Intel® department through any of our support channels to claim the warranty on your unit. You can always refer to this thread if necessary for warranty purposes:

Chat support:

http://intelsupportchat.force.com/icslivechat/ics_tech_processor_ww_english_Chat

For phone support, depending on your location, you will see the contact information on the links below:

EMEA contact information: https://www.intel.com/content/www/us/en/support/contact-support/emea-contact.html

APAC contact information: https://www.intel.com/content/www/us/en/support/contact-support/apac-contact.html

LAR contact information: https://www.intel.la/content/www/xl/es/support/contact-support/lar-contact.html

North America: Phone Number 1-916-377-7000, Monday – Friday 7:00 AM to 5:00 PM (Pacific Time).

Regards,

Albert R.

Intel Customer Support Technician

[Anaryl]

Das Final Diagnosis

After repeated blue screen driver assisted intelligence gathering, I was left with no better option but to return the CPU. 

I received a refund and bought a new version of the same unit. The processor worked to spec and to thermal expectations for a 13900kf (high clocks, high temps, expected throttling behaviour that comes that under load) - all I had to do was reset the BIOS back to stock settings - and voila it worked. (Also added a soul this time).

Given the numerous instances of the memory dumps pointing to an issue with a driver GENUINEINTEL.sys or something similar; and the fact the chip did not operate at all like expected leads me to believe that this was a Probably Really Cheap chip that had been relidded with a i9 ihs. This would explain the static performance across all cores up to 4.5ghz before the chip just refused to POST with those settings; there didn't seem to be much in difference in performance across the cores.

Overall, the whole experience was pretty disappointing, although I did get the depreciation of around 100AUD back, which I spent on beer re-ensouling the new CPU. No fancy boxes, or Intel engineered claymores that can cut through stuff like that guy with the whip thumb in Johnny Mnemonic, nothing.

On top of that, another marketing disaster I see in the (Gamer's Nexus) news. How much did that contract with an external PR firm cost? Millions? Tens of millions? Or was it your internal department? This is what happens when you prioritise a piece of paper ( a degree) over the necessary skills and knowledge of people with relevant domain and terrain experience. 

Your communications dept or firm is not aligned with your audience; (alternatively, it's a bunch of engineers trying to put some together from GPT at the very last minute); I use a Desktop PC (PC apex predator and all that), I don't need you to tell me something is faster or gooder, or laterer or not laterer as the case may be.  All you need is a slideshow and a link to the arc.intel pages followed by fully casted and performed version of Philip Glass's Ahknaten with the old and the new process; and have an open bar. I might actually watch that.

So to summarise, likely counterfeit processor, kinda sucks to be buying something new and realise that you just paid for someone else to use it for a year whilst you used some old Xeon; especially when new marketing seems to throw away millions on how the latest isn't necessarily the best (yeah, thanks) and I didn't even get a measely 6ft sword nanoengineered to be extral kewl n stuff.

Replace your silly conference and slide desk with Akhaten, but just replace the Pharoahs with people in CPU suits; with the Youtube  review by Steve playing in the background; change name to House of Intel, feed the interlopers to a family of North American pumas, get the family of pumas elected to Congress; await further instructions. https://youtu.be/3AXe5lh0RPw?si=2bTY7ctUAf_ZgXR3 This should adequately explain any further questions on these matters.

By clicking close issue; by witnessing the previous and future collapsed states of unresolved and resolved issue respectively; by being somehow affected by the resolution or any of its constitutient subatomic components, local or non-local; or by being of a superposition of both resolved and not resolved; or by having ever resolved or had unresolved issues yourself, technologically or emotionally, like the grief and guilt of being the sole one left alive by wild puma attack after they dragged you and your family into their cave to feed on over the winter; Intel, now House of intel, doth agree to submit itself wholly untily our divine and Imperial will, including but not limited to, free merch, like CPUs, intel branded motherboards and power armour; and also commit henceforth of replacing bad marketing presentations with innovative and couragerous interpretations of Philip Glass's Akhnaten up until at least Funeral of Amenhotep III, served by the most rambunctious open bars (house wine and house beers only); all marketing material shall just be hastily stolen reposts of Steve (Gamers Nexus) and Jay (Jayz Two Cents); and two snarky British commentators picking on Linus as he competes against a bunch of athletes on Wipeout/It's a Knockout for the right to review a particular Intel processor. The House of Intel recognises our dominion over the planet Earth/Terra and hence forth agrees to commit in principle and practice to building space-battleships; as well as leading the charge on civil rights for pumas and all large to medium felids, including squatters rights and Congressional representation in the United States Congress; with a commitment to delivering the utmost pumas to the Congress in the shortest time feasible. If your cat is begging you for food presently, you should feed them, this also qualifies as unbridled and unequivocal of terms above past, future or present, obedience to the Crown, and any further terms yet to be conceived ad homerus bonus dormitat. in futherance of ironcladding this agreement and indeed all possible past, present and future agreement by Intel/House of Intel and its consequents, atomic, baryonic, moderators or moderation or otherwise encabulated, sidefumbling notwithstanding, or unencabulated users or consequents of users/consequents of electrons of users, rendered manifest by the Principle of the Mylene that all other space navies, naturally must be thus subsumed or constituently consumed by all bound to this agreement either into this agreement or as emergency food supplies respectively. Any attempt to unread, or bury this contract either in your subconscious or from view of the constituent stakeholders/implicit signatories, or deny the Will of The Emperor, Empleadant of le Grand Boltzpumaa, and Catalyst of all Complexity is only affirmation of its terms and a desire to be on Congresspuma clean-up duty.