Possible fix for Intel CPUs crashing

[zzetta]

Hi there.

I am posting this because I had a similar cpu instability issue with my new 14700kf with default bios settings. After many tweaks, tries i have found (possibly) a fix for newer intel cpus unstability at default/stock bios settings.

After reading articles about new intel cpus not being stable on stock settings i went ahead and tested my new 14700kf on my z690 aorus elite ddr4 board. Using stock settings, i started Prime95 small fft AVX2 torture test and everything was fine, for about 20 minutes, after which my pc bluescreened. I was mortified. Why would a new intel cpu bluescreen on stock settings??? This was the beginning of my journey in finding a fix.  First i tried to enable intel power limits which is not applied at stock settings. My motherboard would default at 4095w PL2 and 280w PL1. After setting the INTEL power limits, the crashes dissapeared, but i still wasn't happy with the fact that my new cpu was not stable with unlocked power limits. I mean, my old 9900k was overclocked with unlocked power limits, eating up to 250w and after 4 years it did not degrade at all. So I reenabled auto power limits and i started switching options in bios. After 2 days of turning on and off settings in bios, I have found something called "IA CEP" which was on auto. After i disabled that setting, all of the instability with Prime95, Cinebench R23,R24 simply dissapeared. Not only that, but now I can run with auto power limits(4095w/280w) and I don't have any crashes no matter how hard the cpu test is. Also the load voltage seems to be more stable on loads, cpu temperatures are a minimum of 5c lower in most of the tests/games, and i have more fps and more consistent frametimes on games like cs2.

So what I can say is that this IA CEP setting needs to be addressed. In my case it was not the higher power limits as many people stated, or bad motherboard loadlines, it was this Current Excursion Protection that affected the stability. Reading online, this setting is implemented by Intel in the cpus, not motherboard vendors, thats why I am posting here, maybe it will help somehow.

Have a nice day.

[ACarmona_Intel]

Hello Zzetta,

Thank you for posting in our communities.

We sincerely appreciate the information that you have shared or the workaround that you have discovered with the crashing issue with Intel processors.

This would be beneficial to other customers and give us some further hints while we look into the problem.

Thank you, and have a great day ahead!

Best regards,

Carmona A.

Intel Customer Support Technician

[zzetta]

I'm here to give you a quick update, sadly a bad update on the situation. After a few days of working fine, today i started experiencing crashes again on Cinebench and even game shader compilation. I reloaded BIOS defaults and enabled Intel POR limits but now even with those limits my system is crashing. Unfortunate.

[zzetta]

Back again with good news. I reset the BIOS and disabled BCLK Adaptive Voltage and now I am stable again even with fully unlocked power limits. I will test it this week, but it looks promising. IA CEP is back on default settings, AUTO. I attached pictures with my limits and BCLK Adaptive voltage disabled. I really want other people to try this setting and report back.

Option disabled:

Power limits unlocked: 

[KrissyG]

i noticed the 12V is quite low on the BIOS photo, i get that voltage value when teh CPU runs at 200W.
Usually the 12V stays at 12,28V when CPU is on idle. I use a 14 years old PSU corsair GS800, works still best.

What power supply do you use? and did you actually connect both 8pin EPS to the motherboard?

[zzetta]

It's not related to the psu and that voltage on 12v is perfect. I use a RMx 1000w from corsair.

[KrissyG]

there is a voltage drop on the cables betwenn the motherboard and teh PSU.
On 50% power it will drop significantly under 12V.

I recall EVGA SR2 and a 1000W PSU from OCZ, it could not power the board bcoz the voltage was not stable enough.

[zzetta]

There is no issue with my voltage. It stays at 12v no matter the load.

[KrissyG]

if you say so.

[ACarmona_Intel]

Hello Zzetta,

Please check your email inbox, as we have sent you a message via email.

Thank you, and have a great day ahead!

Best regards,

Carmona A.

Intel Customer Support Technician

[zzetta]

Hello,

Thank you, I did reply to the email.

[zzetta]

After more thorough testing, I concluded my CPU is NOT stable at default clocks and voltages. I did reply to the support email in hopes maybe I could've found the solution, but sadly I did not. No matter what I try to switch in BIOS, it will eventually fail. Unfortunate.

[zzetta]

Ok, I'm coming back today with another strategy. I decided to downgrade my BIOS, to the version before they upgraded the intel microcode: "Follow Intel to upgrade Microcode CPUID ver. 0x11D" which is F26 for my motherboard. The latest one I had before the downgrade was F29. F26 is the second BIOS after they added support for 14th gen cpus. Following the same suite of tests, it doesn't crash in anything(Prime95 AVX, Blend, Cinebench R23, 24), AT LEAST FOR NOW. I'll have to test it hard for the next days to have some solid conclussion. 

I have noticed some differences in how the cpu is working with the older BIOS:

1. There is a slight VCORE increase compared to the new BIOS. On the same type of loads I can clearly see that VCORE has an increase of +0.020-0.025 on auto compared to the new BIOS.

2. Given that VCORE is higher on the same loads, the temperature is higher and the scores a just a bit lower compared to the latest BIOS.

3. It doesn't crash with fully unlocked power limits. 

I'll come back with the results in the following days.

[zzetta]

Another interesting thing i noticed with this old bios. It seems that now there is no internal CPU IA OFFSET VOLTAGE. It is 0v.

F26 internal voltage offset picture:

With F29 there IS an IA voltage offset(even if it's not dropped down, i did drop it down in the past and it was on IA line that voltage offset):

So my question is:

Can this microcode update actually be buggy and cause these instability issues becasue it is adding this IA internal cpu loadline offset??

Given that most motherboard vendors have this crashing issue I imagine they all patched this microcode update and did not mess with their own mb loadlines.

I am on a 4 hour marathon Prime95 small fft AVX2 torture test with FULLY unlocked power limits with this old bios and nothing crashed. Let's hope it won't crash. I will leave it run overnight. If it won't crash then I'm pretty sure it has to something at least to do with this.

I'll keep you updated.

[zzetta]

I'm coming back with amazing news: After 14 hours of non stop Prime95 small fft AVX2 torture test there were ZERO crashes or BSODS. You have attached the results file to see that nothing crashed. 

Monitoring the test I can confirm that the supplied voltage is a bit higher, varies from +0.010v to +0.030v on the VCORE. 

The CPU pulled a constant 300w the whole time and sat around 100c. 

This tells me that the CPU is actually working and the problem might actually be related to that IA CORES Voltage offset. 

[zzetta]

Coming back with another set of good news. I have tested Cinebench R23, Cinebench R24 with 2 x30 minutes on each stability test and I'm happy to say nothing crashed with fully unlocked power limits and no overclock. The scores are down by about 300 points on R23 and R24 doesn't suffer at all. 

I also went ahead and checked if the motherboard has modified loadlines compared to the newest bios(F29) and indeed it has IA AC higher, photo below(F26

F29  BIOS  has AC 0.400 / 0.900 loadlines!!!

So,

-> on F26 we have NO IA Cores Offset and higher AC loadline - 0.600 - FULLY STABLE IN ANYTHING with maxed power limits

-> on F29 we have IA Cores Offset (for me it was 0.039v) and lower AC loadline - 0.400 - BSOD with "clock watchdog timeout" and various crashes.

All tests were made with BIOS defaults except:

- Enabled XMP Profile - Memory kit on QVL

- FULLY UNLOCKED POWER LIMITS

With F26 we have higher voltages on auto but there is 0 instability. Notice highlited voltage lines in the picture: Idle voltages are higher, passing 1.5v but on loads they are just a bit higher compared to F29:

 

Basically F26 fixed every instability issue. I'm glad my CPU is not broken.

[zzetta]

Coming back with another set of good news. I have tested Cinebench R23, Cinebench R24 with 2 x30 minutes on each stability test and I'm happy to say nothing crashed with fully unlocked power limits and no overclock. The scores are down by about 300 points on R23 and R24 doesn't suffer at all. 

I also went ahead and checked if the motherboard has modified loadlines compared to the newest bios(F29) and indeed it has IA AC higher, photo below(F26

F29  BIOS  has AC 0.400 / 0.900 loadlines!!!

So...

-> on F26 we have NO IA Cores Offset and higher AC loadline - 0.600 - FULLY STABLE IN ANYTHING with maxed power limits

-> on F29 we have IA Cores Offset (for me it was 0.039v) and lower AC loadline - 0.400 - BSOD with "clock watchdog timeout" and various crashes.

All tests were made with BIOS defaults except:

- Enabled XMP Profile - Memory kit on QVL

- FULLY UNLOCKED POWER LIMITS

With F26 we have higher voltages on auto but there is 0 instability. Notice highlited voltage lines in the picture: Idle voltages are higher, passing 1.5v but on loads they are just a bit higher compared to F29:

 

Basically F26 fixed every instability issue. I'm glad my CPU is not broken.

[zzetta]

I'm trying to post my findings but the thread has a bug or something. 

The key idea is that switching back to F26 fixed my CPU instability in everything.

[zzetta · Accepted solution]

After the 14 hour AVX test, i went and tested Cinebench R23 and R24 and they both passed 2 30 minutes stability test each.

Looking to see what changed, I managed to find these differences between F29 and F26:

- On F26 AC Loadline is higher, set at 0.600, photo below:

So on F26 we have NO IA Offset voltage and higher AC loadline - 0.600 - Higher Vcore but 0 instability in any test with all power limits removed.

- On F29 we have an IA Offset of 0.039v but lower AC loadline - 0.400 - Lower auto Vcore but crashes in heavy workloads with "Clock Watchdog Timeout" BSOD.

Bios settings were auto on both versions except these settings being changed:

1. Enabled xmp - Kit on QVL.

2. Unlocked/Maxed out power limits.

There are now higher voltages in idle, passing 1.5v, but on loads, it's between +0.010v and 0.030v higher compared to F29, photo below:

Conclusion:

CPU is not unstable from factory, but gimped either by a mistake with that IA CORES Offset, or motherboard vendors lowering the AC loadline to get better temps and higher scores, sacrificing stability.

[zzetta]

Again i did a batch of hard tests containing Prime95 AVX, Cinebench, OCCT AVX, and games that had the crashes at sharder compilation. I'm happy to say I have 0 issues. 

[zzetta]

The CPU is fully stable and can handle any amounts of current and watts i throw at it:

Notice the green and orange lines. It spiked to 300 amps and almost 400w and it did not sweat at all.

With the previous BIOS (F29) such a spike would cause an instant BSOD.

F26 fixed everything.

Also,

It seems that my posts that did not work when I wanted to post them got posted now. Buggy forum.