Owners of 13th/14th Gen Raptor Lake CPUs - Media Reports of serious stability issues

[FlyingScot]

This information is being offered with the intention of keeping owners of these CPUs informed, i.e. knowledge is power. Perhaps if enough people become aware of this situation (and the information provided is indeed accurate) then maybe Intel will offer affected users alternative remedies other than the usual warranty replacement process.

Owners of 13th/14th gen i9’s are reporting what looks like rapid degradation over a period of months that leads to sudden instability. Users are then forced to increase voltage and/or drop boost frequencies to stabilize their systems. In some cases, this problem resurfaces and requires the same approach to be repeated. Another major cause of concern raised in the video, and in user comments, is the stability of the Integrated Memory Control (IMC) and the relationship to DDR5 frequencies above 4200. Random instability due to IMC issues, and perhaps degradation of the IMC, cannot be ruled out. This is clearly a story that is still developing. Owners of i9 CPUs, and indeed all Raptor Lake CPUs, should stay tuned. Follow up videos on this story are promised by Steve at Gamers Nexus.

What’s most alarming about this situation is the fact that even users who took every preventative safety measure, i.e. not overclocking, low PL1/PL2, good cooling, even undervolting, etc., are still reporting instability after just a few months of usage, presumably from what could be degradation. And, judging from user comments, these issues may not just be limited to the i9's.

I do not own one of these chips, but I have been following this story with great interest.

UPDATE 1: For folks who want a quick way to see where things stand at the moment, I suggest you start reading this thread from here. In the first post by forum member, CiTay, you will find several relevant links of importance. I will endeavour to keep adding to this shortcut list as events develop.

UPDATE 2: By now, many of you will be aware that Intel and the motherboard manufacturers have collaborated to release new BIOS code to address the Raptor Lake instability and degradation issues. Should you choose to update to the latest BIOS (with microcode 0x129) and then experience thermal and/or performance issues, you may find this new guide helpful in addressing those concerns.

UPDATE 3: [Sep 26, 2024] Intel and the motherboard manufacturers have collaborated to release new BIOS code (0x12B) to address the Raptor Lake instability and degradation issues. For more information, jump to the following post in this thread.

ADDITION 1: Don't forget to stop by our very own online Raptor Lake Survey, where you can see how others have configured their Raptor Lake systems to balance performance and stability, and to reduce the risk of degradation. You can check it out here.

UPDATE 4: [May 1, 2025]: Intel releases 0x12F microcode, which includes a fix for voltage management issues for PCs that are left powered on for days on end, with long periods of idle. See more information here.

 

[Arctucas]

<SNIP> Are you by any chance running on an older version? Also, would you be willing to try disabling CEP to see if anything changes?<SNIP>

A30 (AMI BIOS 7D86vA3 2024-04-30 11.13 MB). This BIOS works for me, I see no reason to change.

I will give it a try.

I only have one issue with this motherboard; Thunderbolt has a problem when Hyperthreading is enabled.

 

[MigraineFilms]

Ive tinkered with these settings on my own, and these are my findings:

For information, this is my build:
MSI MAG Z790 Tomahawk Wifi MAX
i9 14900k
Deepcool LT720
G-Skill Trident Z5 DDR5 6400
RTX4080 Super

My Cinebench 24 score was about 2100, with the TDP reaching nearly 400 watts and my CPU absolutely cooking itself. I lowered the PL1 and PL2 to 253-300 and reached about the same score.
With the new MSI BIOS update, the scores plummeted. It was due to AI-CEP being enabled with a Load Lite of 10. Increasing that to 16 made it stable, but again cooking my CPU.
I decreased the voltages with -0,110, dropped the Load Lite to a stable 12, PL1-PL2 at 200-253 and 310 Amps.

I have yet to test the AC/DC settings. Intel Default Load Lite got me a bit too toasty

My current Cinebench 24 score is 1951, so its about 10% slower with the intel recommended settings, but the temperatures average around 66 degrees, with peaks to 85. For some reason, the Cinebench R23 ran for 10 minutes straight, without throttling, and when I took a screenshot, it started to throttle at 85 degrees.
Intel recommended me to RMA the processor due to thermal throttling.

I will wait until after the microcode patch before I do that, because this just seems to be that weird voltage spike it has during low effort tasks.

 

[FlyingScot]

A30 (AMI BIOS 7D86vA3 2024-04-30 11.13 MB). This BIOS works for me, I see no reason to change.

Just as I suspected. You are on the "pre-Intel advisory" BIOS. And, judging from the "release notes", it should behave like all previous MSI BIOS releases (e.g. 10th gen, etc.), which is quite possibly why your traditional method of undervolting works so well (albeit with the one wrinkle of fixed+offset). The next two BIOS updates for your motherboard start messing with CPU Lite Load and Load Line Calibration. And, again, without telling the user what they are doing behind the scenes. People on newer BIOS code should seriously consider backtracking in my opinion, especially if they are comfortable with undervolting and overclocking.

BTW, I found this strange comment in the release notes for your current BIOS version: CPU uCode version 0x123 was updated for CPU performance optimization by disabling the CEP function for the 14th Gen CPU (stepping B0).

Now why would MSI/Intel think it necessary to disable CEP based upon your stepping? What the heck is going on???? Is this possibly why some people have problems with CEP enabled and others do not? This whole mystery goes deeper than the average Scooby-Doo plot!

 

[FlyingScot]

Ive tinkered with these settings on my own, and these are my findings:

For information, this is my build:
MSI MAG Z790 Tomahawk Wifi MAX
i9 14900k
Deepcool LT720
G-Skill Trident Z5 DDR5 6400
RTX4080 Super

My Cinebench 24 score was about 2100, with the TDP reaching nearly 400 watts and my CPU absolutely cooking itself. I lowered the PL1 and PL2 to 253-300 and reached about the same score.
With the new MSI BIOS update, the scores plummeted. It was due to AI-CEP being enabled with a Load Lite of 10. Increasing that to 16 made it stable, but again cooking my CPU.
I decreased the voltages with -0,110, dropped the Load Lite to a stable 12, PL1-PL2 at 200-253 and 310 Amps.

I have yet to test the AC/DC settings. Intel Default Load Lite got me a bit too toasty

My current Cinebench 24 score is 1951, so its about 10% slower with the intel recommended settings, but the temperatures average around 66 degrees, with peaks to 85. For some reason, the Cinebench R23 ran for 10 minutes straight, without throttling, and when I took a screenshot, it started to throttle at 85 degrees.
Intel recommended me to RMA the processor due to thermal throttling.

I will wait until after the microcode patch before I do that, because this just seems to be that weird voltage spike it has during low effort tasks.

I know I’m probably driving everybody up the wall with all my posts of late, but I didn’t want to ignore your post. Thank you for sharing your results; they’re exactly what I would suspect. See my note above. There are definitely combinations of user settings and BIOS versions that work best, and I just think one has to be careful not to mix and match them accidentally. I think you can take different paths to reach the same destination, which is my whole undervolting table thingy, but the trick is obviously picking one method at a time based upon your comfort zone/experience.

It looks to me like MSI / Intel are trying to automate the process so the Raptor Lake user has a more plug-and-play experience - while trying to solve the instability issues, of course. However, in my humble opinion, the latest BIOS releases neither help the novice nor the super-user. And they especially don’t help the guy in the middle who likes to tinker. My best advice (given all the empirical evidence I have crunched to-date) is to either stay on (or rollback to) a pre-Intel advisory BIOS and use the traditional method of undervolting, combined with a manual multicore overclock (to eliminate single core boosting), or just wait for the latest Intel BIOS and experiment like crazy, i.e. with and without using CEP, and CPU Lite Load=Intel Defaults, etc. In theory, I guess there could be a benefit to the new eTVB bug fix and the upcoming voltage bug fixes. But again, how hard are you going to have to fight against these new auto-mechanisms that drive down performance or drive up TDP? Hey, perhaps I’ll be proven wrong and the new August code will finally deliver acceptable results out of the box. We’d all be happy campers then!

Keep us posted with any of your further tinkering results.

 

[crimson8814f802d4]

I know I’m probably driving everybody up the wall with all my posts of late, but I didn’t want to ignore your post. Thank you for sharing your results; they’re exactly what I would suspect. See my note above. There are definitely combinations of user settings and BIOS versions that work best, and I just think one has to be careful not to mix and match them accidentally. I think you can take different paths to reach the same destination, which is my whole undervolting table thingy, but the trick is obviously picking one method at a time based upon your comfort zone/experience.

It looks to me like MSI / Intel are trying to automate the process so the Raptor Lake user has a more plug-and-play experience - while trying to solve the instability issues, of course. However, in my humble opinion, the latest BIOS releases neither help the novice nor the super-user. And they especially don’t help the guy in the middle who likes to tinker. My best advice (given all the empirical evidence I have crunched to-date) is to either stay on (or rollback to) a pre-Intel advisory BIOS and use the traditional method of undervolting, combined with a manual multicore overclock (to eliminate single core boosting), or just wait for the latest Intel BIOS and experiment like crazy, i.e. with and without using CEP, and CPU Lite Load=Intel Defaults, etc. In theory, I guess there could be a benefit to the new eTVB bug fix and the upcoming voltage bug fixes. But again, how hard are you going to have to fight against these new auto-mechanisms that drive down performance or drive up TDP? Hey, perhaps I’ll be proven wrong and the new August code will finally deliver acceptable results out of the box. We’d all be happy campers then!

Keep us posted with any of your further tinkering results.

Not sure I agree with that, mate. Intel have advised that people get the latest bios for their systems because these include microcode changes that aim to alleviate some of the instability issues people have been facing.
I would suggest latest bios + set conservative PL values (below 253w), ensure that the core current is kept on or under intel's spec, and then use LiteLoad modes to find a low / stable setting for your particular system - this must be tested.

Also, be sure to disable CEP so it doesn't screw with the LiteLoad options and half your perf.

Hoping the August bios puts this to bed.
Cheers

 

[FlyingScot]

Not sure I agree with that, mate. Intel have advised that people get the latest bios for their systems because these include microcode changes that aim to alleviate some of the instability issues people have been facing.
I would suggest latest bios + set conservative PL values (below 253w), ensure that the core current is kept on or under intel's spec, and then use LiteLoad modes to find a low / stable setting for your particular system - this must be tested.

Also, be sure to disable CEP so it doesn't screw with the LiteLoad options and half your perf.

Hoping the August bios puts this to bed.
Cheers

Yeah, I do see the potential benefit to the microcode bug fixes. But then again, MSI/Intel's previous efforts actually drove up the voltage rather than lowering it. Hopefully, that will change. Intel cannot have their cake and eat it. They either need to lower Vcore to save the chips that have not suffered major degradation yet, or keep voltages elevated to stabilize those chips that have already become unstable. They simply cannot have it both ways, at least not without some fine footwork in the code. Plus, from what I have read, there are scores of i7 users who are very happy with their setup (with a mixture of old overclocking/undervolting techniques) and who have not updated their BIOS in eons. But you make a good point. Perhaps the i9 users should tilt towards staying on this merry-go-round until it stops. Those are the most affected CPUs...we think.

 

[MigraineFilms]

I specifically bought the K model because it was just as expensive in the Netherlands as the 13700k, and I upgraded from an 6700k. I wanted to tinker a bit with overclocking and undervolting, I got way more than I bargained for, but instead of overclocking and being scared of breaking it, I need to work to undervolt it to not break at stock.

Right now, Intel is quite desperately trying to fix the bug, one that occured by pushing every last bit of performance out of the chip and thus pushing it way beyond its comfort zone, but it wouldnt be releasable as new product if it didnt have at least a few percent of perfromance increase of the last generation.

What I want to know is: is my chip damaged. I can undervolt it, and it seems to be running quite stable and cool with an occasional spike in thermal throttling, god knows where that comes frome, but so far its stable, is this a sign of a healthy chip?

 

[FlyingScot]

I specifically bought the K model because it was just as expensive in the Netherlands as the 13700k, and I upgraded from an 6700k. I wanted to tinker a bit with overclocking and undervolting, I got way more than I bargained for, but instead of overclocking and being scared of breaking it, I need to work to undervolt it to not break at stock.

Right now, Intel is quite desperately trying to fix the bug, one that occured by pushing every last bit of performance out of the chip and thus pushing it way beyond its comfort zone, but it wouldnt be releasable as new product if it didnt have at least a few percent of perfromance increase of the last generation.

What I want to know is: is my chip damaged. I can undervolt it, and it seems to be running quite stable and cool with an occasional spike in thermal throttling, god knows where that comes frome, but so far its stable, is this a sign of a healthy chip?

No one can really answer that question because it looks like even Intel doesn't have a comprehensive understanding of what is going on. The spikes in thermal throttling could be due to this latest leaker who reports an issue with CPU Cores transitioning from idle to low-load state and back. But maybe not. Keep in mind that it only takes one core to thermal throttle to set that flag. And PL1/PL2 limits will not really help you if just one or two cores are under load. Of course, that brings us right back to voltage spikes. All I can say is that I tend to fall back on what we know worked well for Skylake Sandybridge onwards, which was to aim for a max of 1.35v and definitely no more than 1.4v for the newer chips. I think that would be a good way to protect the Cores and the Ring as best as we can. How long have you had the chip in it's current BIOS configuration? That should tell you something, at least. If it's a year or more then maybe that's an encouraging sign.

 

[MigraineFilms]

Ive had the chip working fine since november, with power limits and liteload on 10, keeping it under 1,4 volt. I just set the voltage manually to lower liteload with CEP enabled yesterday due to the new BIOS crippling performance with "recommended" settings. The only issues I had, with benchmarking and video processing, was the weird lag of half a second in the audio while gaming. I think that might be those low-load state switches. I could ofcourse run a few hours of prime95 to see if the temperatures throttle or if it crashes.

 

[FlyingScot]

Ive had the chip working fine since november, with power limits and liteload on 10, keeping it under 1,4 volt.

That sounds like a good base to work with.

The only issues I had, with benchmarking and video processing, was the weird lag of half a second in the audio while gaming.

You know, I'm hearing quite a bit about latency issues with Raptor Lake, just like you describe. In some cases, Youtubers actually went back to 10th and 11th gens. What a crazy world we live in. Let's hope that gets resolved along with the other issues as it certain could be related to the cores going in and out of sleep state. That's the leading theory.

I could ofcourse run a few hours of prime95 to see if the temperatures throttle or if it crashes.

Hey, don't go longer than 30mins. It's better to use a basket of different apps than run an app that one user on this forum called "Medieval torture!"
Plus, if you're not suffering from game crashes, etc. or WHEA errors, then I won't even stress test until this whole thing settles down and we get some actionable guidance from Intel.

 

[MigraineFilms]

That sounds like a good base to work with.

You know, I'm hearing quite a bit about latency issues with Raptor Lake, just like you describe. In some cases, Youtubers actually went back to 10th and 11th gens. What a crazy world we live in. Let's hope that gets resolved along with the other issues as it certain could be related to the cores going in and out of sleep state. That's the leading theory.

Hey, don't go longer than 30mins. It's better to use a basket of different apps than run an app that one user on this forum called "Medieval torture!"
Plus, if you're not suffering from game crashes, etc. or WHEA errors, then I won't even stress test until this whole thing settles down and we get some actionable guidance from Intel.

That might be the best advice someone has given me. There is a loot of doom and gloom, always people saying something is broken and I need to be careful before I accidentally nuke my whole neighborhood. The thing works, the latency issue is, according to LatencyMon, the netoskrnl.exe which suddenly causes my computer to lag. Weirdly enough, it just happens during games like elden ring, deeprock galactic, helldivers and baldurs gate 3.

edit:
I got another question; why is the vcore 1,3 when idle (it varies between 0,9 and 1,3) and 1,2 when under load? Do I need to set the vcore for the E-cores individually?

 

[FlyingScot]

The thing works, the latency issue is, according to LatencyMon, the netoskrnl.exe which suddenly causes my computer to lag.

Interesting....

...games like elden ring, deeprock galactic, helldivers and baldurs gate 3.

Baldurs Gate 3 is very CPU intensive if I remember correctly. If you are playing all those games without crashes, that's a good sign. Although, I know that UE5 games are the worst.

I got another question; why is the vcore 1,3 when idle (it varies between 0,9 and 1,3) and 1,2 when under load? Do I need to set the vcore for the E-cores individually?

Oh boy. You just opened up a can of worms for me. It's hard to say. 1.2v under significant load might easily just be due to Vdroop, which is both normal and healthy. In theory, it keeps the TDP and temps down, although the current also plays a part. Vcore in a semi-idle state will be erratic because (I think) it's driven by the highest requesting core. I would take a look at what kind of TDP you are driving when Vcore spikes like that. I bet it's quite low, meaning that only one or a handful of cores are consuming that higher voltage. The other cores might receive the 1.3v, but not use it as they are either asleep or not doing much of anything. That's my best guess.

 

[MigraineFilms]

The CPU rises to 40 watts during the spike, maybe 50. Lowest was 20.

Thanks for all this information, I'm learning a huge deal about this all, and I truly enjoy it. Its a shame this processor is the cause of me having to learn it, but I'm having great fun with it.
I'll just use the computer for a week or two and see how stable it actually is, for now, 8 hours of Snes9X, all good! Lord knows how cpu intensive those SNES games are!

I barely play EU5 games, the first one was a demo of an RTS I wanted to play, which had the oodle compiler crash, I dont know how to test it without actual EU5 games though.

Edit: I forgot, been playing Alan Wake 2 for weeks now, all good. 1,3 volt, 80 watts, 66 degrees!

 

[Arctucas]

Tried disabling CEP, got overclocking failed at boot.

 

[FlyingScot]

Tried disabling CEP, got overclocking failed at boot.

What do you think that means? Perhaps too much undervolt, as CiTay theorized a few posts back, i.e. the "training wheels"?

EDIT: At least thanks to your willingness to be my test subject, we now know that CEP is working in the background even when you use a manual undervolt. So I wonder what it is about people's chips that either like it or don't? Is it that stepping thing? Hmmm. Your R23 scores are fine, right?

 

[FlyingScot]

The CPU rises to 40 watts during the spike, maybe 50. Lowest was 20.

Thanks for all this information, I'm learning a huge deal about this all, and I truly enjoy it. Its a shame this processor is the cause of me having to learn it, but I'm having great fun with it.
I'll just use the computer for a week or two and see how stable it actually is, for now, 8 hours of Snes9X, all good! Lord knows how cpu intensive those SNES games are!

I barely play EU5 games, the first one was a demo of an RTS I wanted to play, which had the oodle compiler crash, I dont know how to test it without actual EU5 games though.

That's great to hear! Learning can certainly be fun once you see how your inputs affect the outputs. Let's just be glad that car companies don't expect us to set the engine timing and fuel-to-air mixture before we take it for a Sunday drive!!

 

[FlyingScot]

Edit: I forgot, been playing Alan Wake 2 for weeks now, all good. 1,3 volt, 80 watts, 66 degrees!

Those are nice numbers, and typical for what even an overclocked 10th gen would do. Hey, I was a big fan of the original Alan Wake...incredible forest settings and mood. Perhaps one day I will play #2. I hear it's pretty good.

 

[sascha117715db02f2]

Intel have advised that people get the latest bios for their systems because these include microcode changes that aim to alleviate some of the instability issues people have been facing.

I'd love to be able to do that, but MSI seem to be concentrating on z790 boards for the moment. I'm on a Z690 Tomahawk DDR4 (running an i7-14700KF) and the latest BIOS version I could find is from April 2024 ("1H"). As I was on "1F" (Nov 2023), I did upgrade, but - naturally - after the flash my CPU is still running 0123 microcode, not 0125.

That said: I haven't had any issues with this CPU, running high-ish PL1/2 and ICCmax (but not nearly as high as the unlimited settings all my BIOS versions have defaulted to). It's been a while since I've tinkered with my settings, but the only other relevant changes I've made (and now simply duplicated after I flashed to "1H") is setting CPU Lite Load to 4. IIRC, with the default LL-setting of 9, my power consumption went a bit crazy without really giving me any meaningful performance uplift.

 

[mardon85]

Any pertinent information is always welcome.

What I have managed to do with Adaptive + Offset, set VCore to 1.33V and -0.12V offset, Lite Load to 1, LLC to 3.

Also, both CEP enabled in BIOS. All TVB disabled.

View attachment 190748

Just wanted to say thanks for sharing. I managed to achieve a nice OC/UV with a slightly tweaked varient of these settings.

 

[Arctucas]

What do you think that means? Perhaps too much undervolt, as CiTay theorized a few posts back, i.e. the "training wheels"?

EDIT: At least thanks to your willingness to be my test subject, we now know that CEP is working in the background even when you use a manual undervolt. So I wonder what it is about people's chips that either like it or don't? Is it that stepping thing? Hmmm. Your R23 scores are fine, right?

Not sure what to think about that. Probably an unexposed setting.

Like I said, I spent weeks trying most every setting and combination of settings, probably reset the BIOS at least a dozen times, actually had a couple of corrupted Windows installs that I needed to restore with a backup.

CB R23 scores: ~36K with HT off, ~41K with HT on.