Intel processors crashing Unreal engine games (and others)

[maddie]

And the uninformed buyers of those will have no chance of knowing that the CPU is bad unless they push it hard in specific instances.

In-person meetup sales will go something like,

Seller: So you can see Geekbench, Cinebench, 3DMark etc. all working fine. You are getting a good price. However, you know there is no warranty for used stuff so I won't be responsible if you have any issues afterwards.

Buyer: Can you show me running this and this and this?

After a few demos of stuff that the buyer wants to see, he's satisfied and seals the deal. And he's happy as could be, until a few days or weeks later when he isn't...

Nothing to do when that happens other than cursing the seller and maybe Intel and maybe himself for not researching things more. And trying to pass that crap onto some other shmuck down the line.

Those are the ones with the longest memories. Intel should be worried.

 

[DrMrLordX]

If they never went above 5.3 GHz then I really don’t know what the heck could be going on.

Here's the spot where they mention 53x multiplier:

I had a little trouble figuring out the context of this specific part of the video but it looks like:

One or more of the companies running Raptor Lake-S on W680 had problems with stability so they (when working with Wendel) settled on a few configs to mitigate that, and 53x with single DIMM configs were most stable, but still were crashy. And they're getting tray after tray of CPUs from Intel as replacements for ones that have failed to the point of being useless. So it appears as if these companies are (or at least were, as of Wendel's inquiries) taking new Raptor Lake-S CPUs, putting them in W680 motherboards, and running these settings, only for there still to be failures.

In other words maybe W680 CAN permit higher clocks under some circumstances, but for the purposes of Wendel's inquiries when testing systems with the cooperation of the game server provider companies, they settled on: 53x max multiplier, single DIMM config, disable HT, disable e-cores.

 

[Jan Olšan]

Also, so we all remember that the 1x400 and down are all still 12th gen under the hood (right?) and so should be gtg long term. Long live the 12600 in all its forms

Amusingly the 13500 is actually Alder Lake silicon. So it's probably fine.

So yeah, the 13500, as a cut down 12900, seems like a safe bet too

Well, it gets more complicated. There are variants of i5-13400 / i5-13400F and i5-14400 / i5-14400F that use the Raptor Silicon. These SKUs basically can come with either of the dies, Intel uses them to dump excess/defective dies probably.
If the issue lies in the process node, they may still be affected. But their lower frequency may be shielding them. Perhaps the speed of degradation depends on temps, current loads and clocks, all that making the Core i5s age at a much slower rate.

Not every Core i5-13400F is the same: Raptor (B0) vs. Alder (C0) lake - HWCooling.net

Core i5-13400F SRMBG and SRMBNIn the vast majority of Intel Core i5-13400F processor tests (including ours), you’ve studied the results of the variant built on the Golden Cove cores from the Alder Lake generation. But there is also an iteration from Raptor Lake (with Raptor Cove cores). However...

www.hwcooling.net

The Core i5-14400F analysis is complete. Rare stepping (C0) tested - HWCooling.net

Intel Core i5-14400F (C0 vs. B0)For completeness of Intel Core i5-14400F processor tests we add the results of its second stepping. We’ve already tested the B0 variant, now it’s time for the C0. Its availability in stores is, at least at the moment, considerably worse and the question arises...

www.hwcooling.net

 

[coercitiv]

If they never went above 5.3 GHz then I really don’t know what the heck could be going on.

In the original video it's not clear that 5.3Ghz is the "default" configuration used by the game servers. However in the stream discussed above, there is a moment where he mentions 53X being the option of the data center reps.

 

[SteveGrabowski]

I'd never heard of this setting until now and had to look it up. I know how to write a double-linked list in C using free() and malloc(). I bet most people in this forum can't do that, so I'm not a computer illiterate moron by any stretch of the imagination.

Damn I know how to write free() and malloc() and never heard of that setting lol

 

[poke01]

Well, it gets more complicated. There are variants of i5-13400 / i5-13400F and i5-14400 / i5-14400F that use the Raptor Silicon. These SKUs basically can come with either of the dies, Intel uses them to dump excess/defective dies probably.
If the issue lies in the process node, they may still be affected. But their lower frequency may be shielding them. Perhaps the speed of degradation depends on temps, current loads and clocks, all that making the Core i5s age at a much slower rate.

Not every Core i5-13400F is the same: Raptor (B0) vs. Alder (C0) lake - HWCooling.net

Core i5-13400F SRMBG and SRMBNIn the vast majority of Intel Core i5-13400F processor tests (including ours), you’ve studied the results of the variant built on the Golden Cove cores from the Alder Lake generation. But there is also an iteration from Raptor Lake (with Raptor Cove cores). However...

www.hwcooling.net

The Core i5-14400F analysis is complete. Rare stepping (C0) tested - HWCooling.net

Intel Core i5-14400F (C0 vs. B0)For completeness of Intel Core i5-14400F processor tests we add the results of its second stepping. We’ve already tested the B0 variant, now it’s time for the C0. Its availability in stores is, at least at the moment, considerably worse and the question arises...

www.hwcooling.net

so what your saying is the 12400 is still goat xx400 CPU.

 

[Hitman928]

Here's the spot where they mention 53x multiplier:

I had a little trouble figuring out the context of this specific part of the video but it looks like:

One or more of the companies running Raptor Lake-S on W680 had problems with stability so they (when working with Wendel) settled on a few configs to mitigate that, and 53x with single DIMM configs were most stable, but still were crashy. And they're getting tray after tray of CPUs from Intel as replacements for ones that have failed to the point of being useless. So it appears as if these companies are (or at least were, as of Wendel's inquiries) taking new Raptor Lake-S CPUs, putting them in W680 motherboards, and running these settings, only for there still to be failures.

In other words maybe W680 CAN permit higher clocks under some circumstances, but for the purposes of Wendel's inquiries when testing systems with the cooperation of the game server provider companies, they settled on: 53x max multiplier, single DIMM config, disable HT, disable e-cores.

In the original video it's not clear that 5.3Ghz is the "default" configuration used by the game servers. However in the stream discussed above, there is a moment where he mentions 53X being the option of the data center reps.

It’s not clear to me that this wasn’t him coming in after the fact and switching their boxes to a max multiplier of 53x to help them mitigate the issue. I see no reason a 14900k or similar couldn’t achieve its max boost in a w680 board, it just won’t be able to sustain high boost clocks on many cores due to the boards better sticking to power and current limits.

 

[DrMrLordX]

It’s not clear to me that this wasn’t him coming in after the fact and switching their boxes to a max multiplier of 53x to help them mitigate the issue. I see no reason a 14900k or similar couldn’t achieve its max boost in a w680 board, it just won’t be able to sustain high boost clocks on many cores due to the boards better sticking to power and current limits.

He didn't specify, but if you look earlier in the video at the support contract costs, they do specify disabling e-cores and disabling HT as a standard practice for their Raptor Lake game servers. It looks like they went through a lot of CPUs to figure out "optimal" settings for their users, and it's logical to conclude that they handed those settings to Wendel for his tests as an option rather than vice versa. That still doesn't change the fact that Wendel was probably working with a relatively new or new Raptor Lake in a W680 motherboard while running y-cruncher for stability, indicating that it hadn't seen much (if any) time above 5.3 GHz.

 

[Hitman928]

He didn't specify, but if you look earlier in the video at the support contract costs, they do specify disabling e-cores and disabling HT as a standard practice for their Raptor Lake game servers. It looks like they went through a lot of CPUs to figure out "optimal" settings for their users, and it's logical to conclude that they handed those settings to Wendel for his tests as an option rather than vice versa. That still doesn't change the fact that Wendel was probably working with a relatively new or new Raptor Lake in a W680 motherboard while running y-cruncher for stability, indicating that it hadn't seen much (if any) time above 5.3 GHz.

Yeah, it’d be nice to get his exact testing procedures that home users could try. Could give us some more insights.

 

[zir_blazer]

One of the devs that made the Coreboot port for the MSI PRO Z690-A DDR4 also reported symthomps of degradation on his 14900K that has run almost fully stock since its purchase, let alone some XMP related tests. This is perhaps even more serious, because I was letting fly that CPU degradation could be caused by Motherboard vendors using overclocking defaults so any Processor in use before Motherboard vendors released the Intel default profiles could be considered exposed to degradation conditions, but this one worked @ stock since deployment and has no such excuse.

 

[NTMBK]

So did Intel finally push their 10nm process too far? That thing has been problematic ever since Cannonlake (which they couldn't even get working). They've been tweaking and pushing it for 5 years, but it's always been a dog.

 

[John Carmack]

As expected, the culprit is ICC_max set to above spec - it should be 307 A for the 14900K.

https://www.reddit.com/r/overclocking/comments/1axepvu

ICC_max is more important than the PL1/PL2 values.

Much ado about nothing.

I don't know about you but within the past 5 days I've had enough nothingburgers to last me until Windows on ARM becomes an actual thing.

Does anybody remember when Intel floated the idea of phasing out socketed CPU's and making all of their future desktop releases in BGA form? How about the LGA1151/1200/1700 versus AM4 flame wars?

There was a persistent argument that same-socket CPU upgrades was a useless gimmick feature because nobody changes a CPU without changing the motherboard too. The point was that it was unfair to rip Chipzilla for their constant socket churn.

If Chipzilla and those people had their way, not only would we be sending a million faulty CPU's to the landfill shortly but also a million perfectly functioning overbuilt motherboards with them.

 

[coercitiv]

So did Intel finally push their 10nm process too far?

Don't know about that, but in case it turns out to be true, we can take solace in a cinematic ending: node starts like a dumpster fire, ends like a supernova.

 

[igor_kavinski]

symthomps of degradation

I don't understand people like that. He's gonna continue using his degraded CPU by limiting it instead of asking for refund or replacement

 

[DrMrLordX]

ends like a supernova.

An Intel supernova in the sky?

 

[DAPUNISHER]

I don't understand people like that. He's gonna continue using his degraded CPU by limiting it instead of asking for refund or replacement

Rene will also line right up to buy 15th gen; Stockholm Syndrome.

I can understand not being able to go 2-3 weeks without the system. But why not buy another one, use it until your replacement arrives, and then return it? Because we are well past any of the normal ethical considerations about buying and returning a CPU, given the present circumstances.

There was a persistent argument that same-socket CPU upgrades was a useless gimmick feature because nobody changes a CPU without changing the motherboard too. The point was that it was unfair to rip Chipzilla for their constant socket churn.

Could not agree more where retail is concerned. It's a bad talking point based on outdated behaviors. The reason many changed platforms after 4-5yrs, was that there was no viable alternative. Retail sales the last few years clearly reflect that shoppers prefer a superior upgrade path. Seeing as the present Intel socket has gone tick tock boom! We'll see how things shake out.

I'd like to see a 25-40 game bench of the 14900K using the settings from the discussion @zir_blazer was involved in -

I quickly compiled a ROM with turbo multiplier x53 on all P cores, disabled TVB, disabled IA CEP and the problems were gone.

 

[igor_kavinski]

I can understand not being able to go 2-3 weeks without the system. But why not buy another one, use it until your replacement arrives, and then return it? Because we are well past any of the normal ethical considerations about buying and returning a CPU, given the present circumstances.

Precisely!

 

[eek2121]

I am curious as to what the v/f curve looks like on the 13900k and 14900k compared to other chips that don’t appear to have the problem.

 

[Mopetar]

From a previous post that had a pie chart of crash reports the 13/14900K (and variants of them) dominated the charts, but there were some 13/14700K mixed in, but at a much lower rate. The 13/14600K wasn't on it or too small for it to show up.

That should make for the most interesting point of comparison.

 

[eek2121]

Rene will also line right up to buy 15th gen; Stockholm Syndrome.

I can understand not being able to go 2-3 weeks without the system. But why not buy another one, use it until your replacement arrives, and then return it? Because we are well past any of the normal ethical considerations about buying and returning a CPU, given the present circumstances.

Could not agree more where retail is concerned. It's a bad talking point based on outdated behaviors. The reason many changed platforms after 4-5yrs, was that there was no viable alternative. Retail sales the last few years clearly reflect that shoppers prefer a superior upgrade path. Seeing as the present Intel socket has gone tick tock boom! We'll see how things shake out.

I'd like to see a 25-40 game bench of the 14900K using the settings from the discussion @zir_blazer was involved in -

I’ve done in socket upgrades before, and will be doing so again. I would like to think many out there would do it at least once if the socket has 4-5 years left on it. Technically on AM5 you could go from something like an 1800X to a 5800X3D.

I also imagine it makes buying used and/or having a strict budget a bit easier. You can buy a decent motherboard, a serviceable CPU, and decent GPU now, then upgrade as you have funds. Shoot I did that in the socket 7 days and prior. It was particularly handy when everyone used Intel’s socket for compatibility.

Just looking at ebay (subject to change), You can score a decent x470 board or even x370 (beware the QVL) for under $100, and a 5800X3D for $250. Add in a 3070-3080 for $300 and you have a decent gaming machine for under $750 once you include the case/PSU and stuff.

When the socket changes every 2 years, it makes supply/demand a much bigger issue.

I am not really a vendor specific guy, but socket longevity is one of the factors I use when buying into a platform. Another is stability, but now Intel is failing here as well. 🤦

 

[gorobei]

buildzoid chimes in.

ac loadline setting isnt being enforced (CEP) and vdroop isnt reported correctly resulting in undervolting so low that it is making the cpu unstable. while not the cause, it does contribute.

he is speculating that the ring is killing itself when the cpu tries to crank up voltage for the 6ghz on the 2 p cores. the i9s are failing sooner because they are running higher voltages, but the i7s may just be damaging themselves slower but are still degrading.

the warframe dev's pie chart shows the main determinant by cpu model is voltage, with the only outlier being the ks models which have a different customer target.

 

[blckgrffn]

This all smells like a class action lawsuit waiting to happen. I got like $28 for buying several FX CPUs back in the day, maybe raptor lake i7 & i9 owners can look forward to this too.

 

[Jan Olšan]

It's going to be quite unfair to various other WORKING chips that got hit by class actions, if there isn't one for this, IMHO.

 

[DAPUNISHER]

A lot of affected users in the comments section of the Buildzoid vid. More instances of multiple RMAs by a single user.

Intel right now -

 

[AdamK47]

I remember people sticking with Intel 20+ years ago because the perception was Intel = stable and AMD = unstable. Where are these people today?