News Is your Intel Core i9-13900K crashing in games? Your motherboard BIOS settings may be to blame — other high-end Intel CPUs also affected

[atomicWAR]

Maybe you got confused somewhere but the solution is not clocking down from the normal clocks but from the max overclock they are getting.

Different things run different instructions, a CPU can be stable at everything else and have one single instruction that crashes the system.

Not really, they said reduce clock speed by 200mhz (no mention of oc at that time) and the next paragraph says to down clock the cpu and or undervolt it (to be fair undervolts is still considered an OC by many). So either the article is poorly written or these crashes include stock chips. Not to say they didn't address auto OC as well, they did. Plus from what I have read in comments from users here and in other forums they indeed have this issue at stock with 13900k, 14900k chips. Though to be fair some of those could be trolls.

Edit: read post below of 14900k crashing at stock.

 

[the_vede_]

Not really, they said reduce clock speed by 200mhz (no mention of oc at that time) and the next paragraph says to down clock the cpu and or undervolt it (to be fair undervolts is still considered an OC by many). So either the article is poorly written or these crashes include stock chips. Not to say they didn't address auto OC as well, they did. Plus from what I have read in comments from users here and in other forums they indeed have this issue at stock with 13900k, 14900k chips. Though to be fair those could be trolls.

I have a stock-clocked 14900K that I've kept manually set to Intel's power spec (253 watts / 307 amps) and I've experienced this crash, as well. And I am not a troll. 🙂

 

[atomicWAR]

I have a stock-clocked 14900K that I've kept manually set to Intel's power spec (253 watts / 307 amps) and I've experienced this crash, as well. And I am not a troll.

Thank you for chiming in. It helps a lot. Any luck with a fix?

 

[the_vede_]

Thank you for chiming in. It helps a lot. Any luck with a fix?

My system is no longer crashing during shader compilation in Suicide Squad after I took the bellow steps. But whether or not I truly "fixed" the issue I can't really say, since it's only been a week. For now, my PC seems stable.

Here's what I did, in order:

1. Reverted to an older NVIDIA driver via clean install, rebooted
2. Added .old to the end of the "Suicide Squad - Kill the Justice League" folder (found in AppData) so the game would automatically create an entirely new folder with a fresh shader cache
3. Deleted Suicide Squad entirely, then rebooted, then reinstalled SS and jumped back into the game, at which point I was able to make it past the shader compilation step

Since then, I've reinstalled newer NVIDIA drivers and let steam install two game patches, all of which triggered new in-game shader compilations and my system made it through them all without incident. I should also note that after I got SS running again, I also updated my motherboard's BIOS, because I noticed MSI had released a new version at the end of January that included nonspecific "updated microcode". Whether or not that's a legit piece of the puzzle here, I haven't a clue.

 

[antych]

Some other things you can try is running Windows disk cleanup and checking DirectX Shader Cache. Seems counter intuitive, but it helped me start Lords of the Fallen after I found this solution on Steam. Also running a game once as Administrator often helps, which is why I never suspected this to be a hardware CPU issue.

 

[the_vede_]

Some other things you can try is running Windows disk cleanup and checking DirectX Shader Cache. Seems counter intuitive, but it helped me start Lords of the Fallen after I found this solution on Steam. Also running a game once as Administrator often helps, which is why I never suspected this to be a hardware CPU issue.

Just for clarity, in Lords of the Fallen, were you seeing the same "out of video memory" error that many of us have been seeing during shader compilation, or was it something else? Also, you mentioned in another post that something in Oodle's statement was "very easy to reproduce" on your PC. Did you mean that you ran the AVX2 test they suggested and your system crashed?

 

[CSOCSO]

A little late to the party, but does affect the i5-13600k?

Are your games crashing left and right? Do you have out of vram errors?

 

[bit_user]

We have a situation where shader compilation in UE5 crashes in a fairly deterministic way,

I'm not sure how deterministic it really is. My guess is that there's some sloppy error-handling which basically reports any error during shader compilation as out-of-memory. That's one possibility, at least.

Another possibility is that it truly is hitting an out-of-memory error, but from years of programming a heap-based language, I can tell you such errors usually happen when an uninitialized value is used as the size of a memory allocation. It could be that some data race is causing uninitialized (or otherwise garbage) values to occur in multiple contexts, but the most common way it could cause a fatal error is simply out-of-memory? That seems a bit more far-fetched to me.

As for why this hits shader compilation, probably because that's an intense period of heavily-threaded activity - a lot more than you usually have in games.

 

[TerryLaze]

I have a stock-clocked 14900K that I've kept manually set to Intel's power spec (253 watts / 307 amps) and I've experienced this crash, as well. And I am not a troll. 🙂

You do realize that with 307 Amps you would need to run the CPU at ~0.8V to stay below 253W (245W ) right?! At 0.8V the CPU will be super unstable at anything above idle clocks.
W=a*v

 

[bit_user]

You do realize that with 307 Amps you would need to run the CPU at ~0.8V to stay below 253W (245W ) right?!

Is that current limit really like PL2, or is it instead more like PL4?

It would help if you could point us to the Intel docs which explain how the current limit is applied.

 

[antych]

Just for clarity, in Lords of the Fallen, were you seeing the same "out of video memory" error that many of us have been seeing during shader compilation, or was it something else? Also, you mentioned in another post that something in Oodle's statement was "very easy to reproduce" on your PC. Did you mean that you ran the AVX2 test they suggested and your system crashed?

That one was just a plain game crash, but I had out of memory errors in other games. By easy to reproduce I mean I can usually count on it happening in every affected game. So if I install a new UE5 game, or update drivers, I can expect it to crash on shader compilation. All stress tests pass on my PC.

edit: actually first time it crashed it might have been Oodle/memory error, but after driver update it was just crash to desktop.

 

[antych]

I'm not sure how deterministic it really is. My guess is that there's some sloppy error-handling which basically reports any error during shader compilation as out-of-memory. That's one possibility, at least.

If it always crashes at the same moment with the same error, that's deterministic behavior, rather than random instability usually associated with overclocking or faulty settings.

 

[TerryLaze]

Is that current limit really like PL2, or is it instead more like PL4?

It would help if you could point us to the Intel docs which explain how the current limit is applied.

What are you trying to get at?!
Electricity is not going to be different either way.
He set watts and ampere manually

so volts is what will be changed by the mobo when there are heavy loads.

 

[carom]

He set watts and ampere manually

so volts is what will be changed by the mobo when there are heavy loads.

The setting for ampere is called ICCMAX (from Intels datasheet: "
ICCMAX is the maximum current processor can draw, typically seen running a
virus application (stress applications specifically designed to push the SoC to
maximum Power)". So it's a maximum value. When set to auto, most mainboards will set it to 500+A.

 

[TerryLaze]

So it's a maximum value. When set to auto, most mainboards will set it to 500+A.

Yes, but they will also set watts to unlimited, or 500+ ,and that would allow the volts to be high enough for the CPU to be stable.
(Until you reach absurdity levels)

If you limit your watts below your amps then volts will always be below 1 under load and that is a big problem.

 

[the_vede_]

You do realize that with 307 Amps you would need to run the CPU at ~0.8V to stay below 253W (245W ) right?! At 0.8V the CPU will be super unstable at anything above idle clocks.
W=a*v

You do realize that I was an English major, not an electrical engineer, right?! 🙂

So, uh, no I'm not clear on what you're saying. Intel's base power spec for the 14900K is 125w PL1, 253w PL2, and 307a. I have mine slightly pushed in that I decided to set PL1 at 253w, also, which is what's set automatically when you select "Boxed Cooler" on MSI Z690 motherboards when prompted in the BIOS after installing a new CPU. I've had my system setup like this for several months, and prior to my shader compilation crash in Suicide Squad, the system was rock solid, both in games and stress tests.

So, if you wouldn't mind, can you please unpack what you're saying a bit more, and as if I was an English major, which I am. Thank you in advance. I'm sure it'll benefit others here, as well.

 

[Kirbyarm]

After owning a newly assembled PC for around 13 months, I went through two months of literal virtual hell trying to diagnose why 75% of my programs were crashing / automatically re-opening themselves progressively more frequently every day (down to a few minutes per crash). Game crashes, Steam, Discord, tabs on Opera GX, Firefox AND Google Chrome were all crashing, sometimes the entire browser would just silently crash / disappear and re-open itself.

I tried fresh installation of Windows 10, since I initially thought something software related was corrupt. BSOD during windows 10 install, failure to install almost every driver or program without 10+ tries each, and sometimes it would just go through if you spammed installation attempts enough.

Bought a brand new M.2, socketed in a different slot of the motherboard. BSOD / failure to install Windows 10 on two BRAND NEW USB thumb drives where the media was created from the official Microsoft website's creation tool download from a different, known working and perfectly stable for years computer. Identical issues trying to install drivers, redistributables, Java, anything, just like the other windows install / M.2 attempt. That rules out OS and SSD.

Bought an entire new dual channel 64 GB RAM kit, officially supported by the motherboard manufacturer, unlike the kit I've been using up until now. Tried a fresh windows install with this kit. Identical / very similar issues happened. It's clearly not the RAM.

Bought a brand new motherboard that supports the 13900K. Went Gigabyte on this one. Tried both RAM kits, both SSDs, different sockets, fresh Windows 10 installations using both brand new USB thumb drives. It still took multiple attempts to get windows installed and had a ton of issues installing drivers and other programs. So it's not the motherboard...

There's no flickering, black screens, anything graphically wrong or random BSODs or shutdowns, so I doubt the TX-Prime 1000 SeaSonic PSU was to blame, as well as the RTX 3080 Ti.

Despite everyone I talked to proclaiming "It's so exceedingly rare that you have a failing CPU or a bad CPU, there's no way it's the CPU"... it's the god damn CPU. It has to be.. right?

I've seen similar posts and problems and articles / reports of users with the 13900K also having crashes, also with the KS version as well as even the 14900K... which being an Intel fanatic all my life I've realized they've really dropped the ball here on the two things that kept me with them and trusting them the most. Compatibility and reliability.

Instead of going through hoops of RMA crap I straight up decided to make a shift to Team Red and opted to get a Ryzen 7 7850X3D as well as a decent motherboard with that socket.

EVERYthing installed perfectly fine in one attempt and after 2 weeks of testing the hardware out, I haven't had a single crash or bluescreen whatsoever, 0 problems of any kind like my computers have basically always performed prior to this nightmare. Basically re-bought an entire new PC just trying to solve such a stupid issue. This should NOT be a thing for such high end products of reputable companies that represent most of the industry they manufacture products for.

Just thought I'd share my experiences as a user who suffered far more than just 'games crashing'. I'm just happy the nightmare is over... please let it last.

 

[TerryLaze]

You do realize that I was an English major, not an electrical engineer, right?! 🙂

I don't think that I had the pleasure.

All you will ever need to know about this is in that link.
But the basics is what I already said, voltage is watts divided by ampere, and if the voltage is too low the CPU will crash.
https://www.anker.com/blogs/others/basics-of-watts-to-amps

(Also, just as an aside, the 307A rating is a maximum, as in the CPU can handle that much in an extreme without blowing up, same as the 253W rating is a max turbo rating, it's not the supposed baseline/normal setting. )

 

[the_vede_]

I don't think that I had the pleasure.

All you will ever need to know about this is in that link.
But the basics is what I already said, voltage is watts divided by ampere, and if the voltage is too low the CPU will crash.
https://www.anker.com/blogs/others/basics-of-watts-to-amps

(Also, just as an aside, the 307A rating is a maximum, as in the CPU can handle that much in an extreme without blowing up, same as the 253W rating is a max turbo rating, it's not the supposed baseline/normal setting. )

I understand that you're attempting to educate me here, and I do appreciate it. I also understand that the wattage and amperage settings are maximums. Perhaps you can explain the following:

1. Intel's recommended power settings for the 14900K are 125w/253w, 307a, and their "Extreme" setting is 253w/253w, 307a. But if I'm understanding you correctly (and I may not be), you're saying those settings will crash at anything above idle clocks? If so, then why are those Intel's recommended power settings?
2. Lots of people are using those settings, myself included. Why aren't we experiencing crashes when we use them, even with extreme workloads? Our cores are boosting just fine (assuming thermals are managed) and our benchmarks remain high. So, what's the missing piece of the puzzle here?

 

[Albert.Thomas]

How many real-world instances of that have ever been reported?

Maybe it's just a black swan event that got a lot of attention due to how evocative it is and the fact that it's a blatant hardware fault (i.e. compared with this issue, which I'm sure a lot of people at least initially assumed was a software bug).

YMMV, but I know of 3 people who were impacted by this.

 

[carom]

If you limit your watts below your amps then volts will always be below 1 under load and that is a big problem.

I'm feeling stupid for asking this, but with regards to W = A * V, what makes you think it's the voltage that has to be lowered, isn't it the other way round? E.g. running at a limit of 125W with 1.4V applied means it draws ~96A in this scenario (with 307A being just the possible maximum).

 

[bit_user]

What are you trying to get at?!
Electricity is not going to be different either way.

You took that current limit and computed voltage using the PL2 power spec, but we know that CPUs can go well beyond PL2, for up to 10 ms. This is known as PL4. For the i9-13900K, PL4 is actually 420W. If you divide 420 W / 307 A, you get 1.368 V.

So, if the current limit works like PL4 (i.e. as an absolute threshold that you cannot go beyond), then that's the power value it should be computed against.

If you limit your watts below your amps then volts will always be below 1 under load and that is a big problem.

I believe this is not accurate. I expect the CPU will not adapt its voltage to meet the current limit. Instead, it probably tries to avoid exceeding that limit by throttling clock frequencies.

If you have a good source to cite on how the current limit is applied, please share.

 

[Peksha]

You do realize that with 307 Amps you would need to run the CPU at ~0.8V to stay below 253W (245W ) right?! At 0.8V the CPU will be super unstable at anything above idle clocks.
W=a*v

See LLC power guide https://www.overclock.net/threads/a...00k-an-overclocking-and-tuning-guide.1801569/

 

[bit_user]

See LLC power guide https://www.overclock.net/threads/a...00k-an-overclocking-and-tuning-guide.1801569/

Thanks for sharing!

There's a big discussion of current limit (IccMax) in the comments.

 

[TJ Hooker]

I'm feeling stupid for asking this, but with regards to W = A * V, what makes you think it's the voltage that has to be lowered, isn't it the other way round? E.g. running at a limit of 125W with 1.4V applied means it draws ~96A in this scenario (with 307A being just the possible maximum).

Don't feel stupid, that's a very reasonable question to ask. And you are correct, there's no reason to think that power limit and current limit values would have to be active at the same time, i.e. it doesn't make any sense to assume that when the CPU is drawing 253W it must also be drawing 307A (and vice versa).

Given that you (and @the_vede_ ) are new here, you should probably know that TerryLaze is our resident pro-Intel shill. He has no evidence that what he's saying is true, and may not even care whether it is. He is just using gish gallop to try to paint this as anyone but Intel's fault.