6850K @4.5GHz Random Crashes

[3cl1ps3]

6850K @4.5GHz
VCore 1.338 (adaptive, all cores synced)
Cache @35, cache voltage 1.10
CPU Input voltage, 1.80
VCCSA, 1.10 maximum (via offset)
Load Line Calibration: level 7
RAM, 32GB - 2x16GB, Corsair manual mode 3200MHz
MoBo: Asus Rampage V Edition 10
Graphics card: 1080Ti
PSU: Corsair 1200i
CPU Cooling: Scythe Mugen 5 PCGH edition (air) modded, quite a lot of case fans, almost all of them controlled by the motherboard. All phases of the CPU are connected and the CPU in BIOS is set not to be power limited.

The PC would randomly crash when gaming or under other heavier load, especially when room temperature gets higher than 26-27 degrees. Even at these times the CPU package temperature would rarely reach 78-80 degrees, most of the time it would stay up to 75, core temperatures would be a few degrees lower and most of the time in their 60s. If the room temperature is in the 21-24 degrees range, there would rarely be crashes whatsoever and the Vcore could go down to 1.334 without any issues.

I can raise the VCore to 1.34 or more but when hot, this doesn't solve the problem. It looks like the issue is somehow related to some setting in the BIOS where the motherboard will take control upon reaching certain temperature stage (to prevent damage). Like there are a few settings which refer to the type of cooling the CPU is using (air vs liquid) and I believe my setting there is Auto. I may be wrong, however find it pretty weird that there are no problems when room temperature is just 4-5 degrees lower and I start getting "system thread exception" and "clock watchdog" crashes when it gets hotter. But CPU package and CPU core temperatures never reach 80 degrees, even if they approach it it is only in moments of random peaks. I also monitor all voltages that are all set manually and never exceed the above values.

Anyone got similar issues with Broadwell-E? What could be the culprit? RAM was extensively tested with Memtest off a flash drive, never had any errors. Not certain about the graphics card which is a factory overclocked Strix 1080TI edition, but it never goes (even the VRM) above 75 and looks stable. I believe the PSU is OK too. The motherboard is running the latest BIOS.

Thanks everyone.

 

[darket]

Firstly your making the mistake of trying to overclock everything at once.

Don't run xmp , don't overclock the cache (broadwell-e doesn't really benefit in any real way from cache oc's anyway) just deal with the cpu overclock first.

Secondly with broadwell-e any frequency requiring a core voltage of 1.3v+ is at the limits of what your going to realistically achieve whilst maintaining stability and thermals.

By the looks of the volts and heat that chip was destined to overclock to 4.3-4.4 , so i'd set a static vcore (don't mess with adaptive yet) of around 1.25v and set the all core ratio (mce to auto) to 44 (avx offset 2) and make sure fully manual mode is set for the voltage control)

Make sure svid support is disabled for the cpu and memory

Input voltage to around 1.9v

Set LLC to around level 4

If thats stable then try to reduce the vcore a little at a time then you can try reducing the input voltage a little if you want.

Then and only then set the memory to 3200 , personally i would set the timings manually and set the voltage manually to 1.35v

Set the SA voltage to some where between 0.85-1.1 v again static voltage !

With the Cache , don't bother, it makes ZERO real world difference in anything outside very specific benchmarks.

Just remember heat isn't causing the instability , its just showing the cracks in the unstable overclock.

 

[darket]

RAM, 32GB - 2x16GB, Corsair manual mode 3200MHz
All phases of the CPU are connected and the CPU in BIOS is set not to be power limited.
.

You don't need to fiddle with ANY of the power limits , you wont trip them in any usage scenario even heavy benchmarking , those settings are for extreme overclocking , so just leave everything on auto that i didn't listed above (apart from the svid support and cpu spread spectrum)

And why on earth are you running only 2 sticks of ram in a quad channel board ?

 

[3cl1ps3]

And why on earth are you running only 2 sticks of ram in a quad channel board ?

The rig was pretty expensive at the time of purchase. I am also having a 1.2TB u.2 Intel drive that was like £800 only. m2 drives. Xonar Essence STX II etc. When initially I bought the RAM, Corsair RGB it was also over the top so in the end I decided to stay where I am and go for an entirely new rig when Intel goes for 10 nm, or AMD, but for now I will keep this rig as my main one.

Then and only then set the memory to 3200 , personally i would set the timings manually and set the voltage manually to 1.35v

The RAM runs at 15-16-16-34 1T from default/xmp 16-18-18-36 2T. Voltage is 1.35. No problems with it at 3200MHz, I tested it extensively but couldn't go lower than that. I tried and it's a downpour of errors.

Fully manual mode works fine at 4.5GHz but keeps the VCore constant. The VCCSA is static, I can lower it a little bit more but in adaptive VCore mode it can be done only by an offset, thus it is quite a lot of play and test to get the final result at OS level.

4.4GHz manual is unachievable on my chip below 1.299 (call it 1.30). More or less 0.4+ is the step every 100MHz over 4.2. Isn't LLC 4 too little, even if I increase the input voltage? 4.5GHz even in manual mode would not boot under LLC 7 and 4.4GHz wouldn't boot under LLC 6.

Are you all manual 24/7 on your 6950X and at what VCore/input voltage and frequency? What I want to know is why the system is fully stable at temperatures below 75 and start crashing just 5 degrees over it. I know this is a consequence, not a reason but there must be something that triggers it. I believe derba8er had a video about broadwell-E where he adjusted some BIOS settings about temperature thresholds on his 6950X and Rampage V E10. I need to watch it again.

 

[darket]

The rig was pretty expensive at the time of purchase. I am also having a 1.2TB u.2 Intel drive that was like £800 only. m2 drives. Xonar Essence STX II etc. When initially I bought the RAM, Corsair RGB it was also over the top so in the end I decided to stay where I am and go for an entirely new rig when Intel goes for 10 nm, or AMD, but for now I will keep this rig as my main one.

Are you all manual 24/7 on your 6950X and at what VCore/input voltage and frequency? What I want to know is why the system is fully stable at temperatures below 75 and start crashing just 5 degrees over it. I know this is a consequence, not a reason but there must be something that triggers it. I believe derba8er had a video about broadwell-E where he adjusted some BIOS settings about temperature thresholds on his 6950X and Rampage V E10. I need to watch it again.

Firstly derba8er's videos wont help you do a 24/7 stable overclock , his overclock videos are just quick and dirty to get an idea for the potential of a chip before he Ln2's the crap out of it.

Are you all manual 24/7 on your 6950X and at what VCore/input voltage and frequency? What I want to know is why the system is fully stable at temperatures below 75 and start crashing just 5 degrees over it. I know this is a consequence, not a reason but there must be something that triggers it. I believe derba8er had a video about broadwell-E where he adjusted some BIOS settings about temperature thresholds on his 6950X and Rampage V E10. I need to watch it again.

I've already addressed this above , its very simple your overclock is unstable, AND your air cooler can't manage the thermals, you need to reign in your expectations of this cpu. Broadwell-E was very poor in terms of overclocking potential , pretty much anything over 4.2ghz is luck , VERY few chips can manage a real stable 4.4 let alone 4.5ghz and anything above is golden sample territory.

And just because your memory isn't producing errors does't mean its not putting stress on the cpu overclock.(Same for cache)

My advice would be to follow my above guide , just concentrate on the cpu, start at around 4.2 and see what volts you need (static) then work from there. As soon as you need a large bump in voltage (say .4 .5 etc) to get the the next ratio means you have reached the limit of that chip and without a custom loop cooling it all you haven't got a prayer of maintaining any sort of stability.

And i'll say one more time , Just because your within the thermal limits of the cpu does NOT mean that temps arn't having an effect on a dodgy overclock.

 

[darket]

Are you all manual 24/7 on your 6950X and at what VCore/input voltage and frequency? What I want to know is why the system is fully stable at temperatures below 75 and start crashing just 5 degrees over it. I know this is a consequence, not a reason but there must be something that triggers it. I believe derba8er had a video about broadwell-E where he adjusted some BIOS settings about temperature thresholds on his 6950X and Rampage V E10. I need to watch it again.

My 6950x at 4.4 needed a fixed Vcore of 1.3v for stability, temps under heavy stress tests (Rb,Prime,Occt etc) never exceeded 72 degrees, gaming the cpu was always around 50-55 degrees , i could run adaptive voltage at 4.3 with a target of 1.275v , Input was 1.85v with a Sa voltage of 0.85v , but all this took me MANY hours of tweaking and testing