Prime 95 fail

[Tan3l6]

Lately I've gotten crashes and black screens, so I tried to get to bottom of it by running Prime95.

And it works fine (Blend mode) for about half an hour. After that fails always either thread 3 or 4 or both. So my CPU might be defective? I've tried several RAM sticks and lowered OC to default speeds but yet  3. or 4. thread fail.

I'm guessing theres nothing else to do than buy a second CPU to test?

My CPU specs are in my profile, but motherboard is GA-Z270X-Gaming K5 and CPU 6700K. CPU delidded and at 4,6GHz (1,3v)

Also, Small FFTs runs fine. Maybe memory controller faulty or overheating? Though temperatures of CPU under load are below 80C.

Haven't run Memtest86 much longer than 30 minutes, as usually memtest reacts faster to RAM errors than Prime95. And memtest86 does not find any errors.

 

 

---

But maybe VRM MOS gets too hot?

[Vishera]

Never use blend mode,it's useless,

Small FFTs is what i recommend,if it works well with small FFTs then there is no reason to worry.

[Tan3l6]

2 minutes ago, Vishera said:

Never use blend mode,it's useless,

Small FFTs is what i recommend,if it works well with small FFTs then there is no reason to worry.

OK, but I'm trying to find the black screen / freeze cause, blend mode seemes to be the only test that gives error, nothing to go on otherwise.

Also sometimes screen froze and a terrible shriek at maximum volume looped, that was scary.

[Vishera]

Just now, Tan3l6 said:

OK, but I'm trying to find the black screen / freeze cause, blend mode seemes to be the only test that gives error, nothing to go on otherwise.

Try testing the GPU with GPUPI 3.3.3,

Tell us if you get any errors while testing.

 

GPUPI 3.3.3 download page:

https://www.overclockers.at/news/gpupi-international-support-thread

 

Press Calculate then OK to start.

[Tan3l6]

5 minutes ago, Vishera said:

Try testing the GPU with GPUPI 3.3.3,

Tell us if you get any errors while testing.

 

GPUPI 3.3.3 download page:

https://www.overclockers.at/news/gpupi-international-support-thread

 

Press Calculate then OK to start.

GPUPI does not crash. 1B 

Also I'm starting to suspect the VRM temp, as it gets hotter very slowly and Prime 95 errors never appear earlier than about half an hour.

[Vishera]

Just now, Tan3l6 said:

GPUPI does not crash. 1B 

 

You shouldn't fiddle with settings like that...

The default is 32B and this is what should be used,1B splits the load between GPU and CPU and as a result is less demanding.

[Tan3l6]

30 minutes ago, Vishera said:

You shouldn't fiddle with settings like that...

The default is 32B and this is what should be used,1B splits the load between GPU and CPU and as a result is less demanding.

OK, running only GPUPI to 32B places introduces no problems. But...

before testing only GPU, I ran Prime95 Blend and GPUPI at the same time and got a green screen about NVLDDMKM.SYS.

 

So, not sure what to think...

[Vishera]

Just now, Tan3l6 said:

OK, running only GPUPI to 32B places introduces no problems. But...

before testing only GPU, I ran Prime95 Blend and GPUPI at the same time and got a green screen about NVLDDMKM.SYS.

 

So, not sure what to think...

NVLDDMKM.SYS is the Nvidia driver,

Could be a problem with the graphics driver,but it could be a problem with the hardware too.

I hope for you it's only a driver issue,

Try uninstalling the Nvidia driver with DDU then install the newest driver.

[Tan3l6]

Well it seems maybe something in CPU has failed, as always the failing threads in Prime95 are the same 3 and 4. Just like one core with its hyperthreading.

But I hope it's something repairable too.

[Zando_]

55 minutes ago, Vishera said:

Never use blend mode,it's useless,

Small FFTs is what i recommend,if it works well with small FFTs then there is no reason to worry.

How so? Small FFT is for 10 minute worst-case temp testing runs and a quick basic stability test. Blend is for actual hour+ run stability testing, it hits all parts of the CPU, power delivery, IMC, and RAM itself AFAIK. 

[Vishera]

Just now, Zando Bob said:

How so? Small FFT is for 10 minute worst-case temp testing runs and a quick basic stability test. Blend is for actual hour+ run stability testing, it hits all parts of the CPU, power delivery, IMC, and RAM itself AFAIK. 

You are probably right

[Zando_]

1 hour ago, Tan3l6 said:

Lately I've gotten crashes and black screens, so I tried to get to bottom of it by running Prime95.

Sounds like an unstable OC.

1 hour ago, Tan3l6 said:

And it works fine (Blend mode) for about half an hour. After that fails always either thread 3 or 4 or both. So my CPU might be defective?

Most likely not defective:

1 hour ago, Tan3l6 said:

CPU delidded and at 4,6GHz (1,3v)

Delidding can damage the CPU if you do it wrong and don't mask certain stuff (assuming you used liquid metal as replacement TIM), but if your CPU has been fine for a while then it's not faulty. 6700K is a 4Ghz, max 4.2Ghz boost CPU. 4.6 at 1.3v is an OC, have you not considered the fact your OC could be unstable? 

Memory is likely fine, if anything is up with that, windows and then memtest will usually crash it. Failing threads will be down to your CPU, again like I said above, blend is a different load than small FFT. Small FFT is max power consumption in the interest of creating the most possible heat, so you can account for that if you come under a very heavy all-core AVX workload or something in the future, and make sure your CPU doesn't slap thermal limits and either throttle or BSOD. Blend does a lot of other stuff, in fact it actually tells you what each test does in Prime95 itself, blend runs all of them basically on a loop, it hits all parts of your CPU and RAM, power delivery and IMC. ASUS RealBench is another good option, I use it for stability testing on a lot of my rigs as well, it also hits all parts of the CPU.

 

You can often be stable in one workload and then crash in others. Or run one test for 3 hours straight and then fail at 3 hours and one minute. If you get crashing and such in Windows and general use, that's when you take a look-see at your OC settings and start making sure you do proper stability testing. Once you fine tune it and it passes an hour run, leave it running RealBench or Prime95 blend overnight and see if it survives till morning (I've had OCs survive an hour run of RealBench but crash in Destiny 2, then when I ran RealBench overnight it found the instability and stopped the test). A lot of OCs sadly need more voltage than people think, most just don't do multiple hour stability testing and haven't hit a workload where it consistently crashes yet. 

[Vishera]

10 minutes ago, Zando Bob said:

Delidding can damage the CPU if you do it wrong and don't mask certain stuff (assuming you used liquid metal as replacement TIM), but if your CPU has been fine for a while then it's not faulty. 6700K is a 4Ghz, max 4.2Ghz boost CPU. 4.6 at 1.3v is an OC, have you not considered the fact your OC could be unstable? 

I didn't notice the CPU was overclocked,good catch.

[Tan3l6]

22 minutes ago, Zando Bob said:

6700K is a 4Ghz, max 4.2Ghz boost CPU. 4.6 at 1.3v is an OC, have you not considered the fact your OC could be unstable? 

Yep, but mostly small FFTs gives an error earlier than Blend mode if it's an unstable OC, in my experience. Now I lowered the clock by 100Mhz and am testing if it's stable.

22 minutes ago, Zando Bob said:

If you get crashing and such in Windows and general use, that's when you take a look-see at your OC settings and start making sure you do proper stability testing.

Well recently i played Outer Worlds, a somewhat taxing game for hardware with no problems. Many hours straight with no problems. And no problems in most other games.But one UE4 game made it obvious though that the system is not stable. Fantasy Blacksmith that is.

 

Will test for stability, but if downclock does not help maybe the VRM MOS temp is at its max possible limit?

Also the machine was stable for a long time.

 

But the delidding was done not in the best fashion, I didn't use any sort of polish to shield the top layer of cpu components under IHS, only liquid metal and a "loose" lid. Still, if the problem was some kind of a short circuit the problem would be instant in my opinion.

[Vishera]

1 minute ago, Tan3l6 said:

Will test for stability, but if downclock does not help maybe the VRM MOS temp is at its max possible limit?

 

If it doesn't help try with stock clock speeds and voltages,

If it works on stocks speeds than the OC was not stable.

[Zando_]

1 minute ago, Tan3l6 said:

Yep, but mostly small FFTs gives an error earlier than Blend mode if it's an unstable OC, in my experience. Now I lowered the clock by 100Mhz and am testing if it's stable.

Like I mentioned above, it's good for a quick burst stability test, quickly finds any glaring issues. Still not as reliable as Blend, but quicker when you're getting an OC up and running, then you take the time to tweak once you've hit your basic clock/voltage limit. 

2 minutes ago, Tan3l6 said:

Well recently i played Outer worlds, a somewhat taxing game for hardware with no problems. Many hours straight with no problems. And no problems in most other games.But one UE4 game made it obvious though that the system is not stable. Fantasy Blacksmith that is.

Yep. I can run Battlefield 4 and Battlefront 2015 just fine on my X58 rig with an X5670 running at 4.3Ghz and SLI GTX 780s running 1300-1320Mhz or at stock, but in Destiny 2 it randomly crashes (had to increase voltages on the CPU IIRC, still need to do an overnight stability test). You can do everything fine and then have one random game that crashes, or one specific workload. Just how it is with overclocking, that's when you see it's not as stable as you thought it was. First you feel kinda disappointed, and then you decide what you'll do about it. You have two options: lower clocks, raise voltages. Raising voltages increases temps, Intel max safe for a 6700K is 1.35v AFAIK (if you wanna play it safe), most 14nm boyes other than the HEDT chips handle up to 1.4v (if you're brave enough) just fine though (I ran my 8600K at 1.38v for a 5Ghz OC, 1.42v or so for 5.1Ghz IIRC, it didn't scale well past 5Ghz). Lowering clocks drops performance, but a 100 to maybe 200Mhz difference usually isn't noticeable in games. 

7 minutes ago, Tan3l6 said:

Will test for stability, but if downclock does not help maybe the VRM MOS temp is at its max possible limit?

Seems to have solid VRMs and you have chunky aluminum heatsink boyes on them:

 

IIRC most VRMs are rated to well, well in excess of 100C. Unless you have no airflow whatsoever in your case, you'd need to reeeeeeally be pushing the board to overheat the VRM. Reviewers pushed 7700Ks to 5Ghz just fine on this board, a 6700K at that voltage and clock is ez street. 

 

[Tan3l6]

Yep, preliminary tests indicate that probably one core was not able to OC stable 4,6Ghz, now at 4,5 all seems stable. Will leave over night.

Also found out thet putting CPU and GPU to max load - Prime95 and GPUPI the temps go through the roof: 86C. Bad ventilation, I know.

 

Why I didn't suspect OC - I had that 4,6Ghz over a year with no problems in any games or applications.

Preliminary Thanks @Zando Bob and @Vishera!

[Zando_]

1 minute ago, Tan3l6 said:

Yep, preliminary tests indicate that probably one core was not able to OC stable 4,6Ghz, now at 4,5 all seems stable. Will leave over night.

Also found out thet putting CPU and GPU to max load - Prime95 and GPUPI the temps go through the roof: 86C

That's hot but still below safe max, what cooler/case/setup are you running? 

2 minutes ago, Tan3l6 said:

Why I didn't suspect OC - I had that 4,6Ghz over a year with no problems in any games or applications.

Lol yeah, it's a common mistake. Really easy to think its stable, but then one specific app or game crashes it all. 

[Tan3l6]

11 minutes ago, Zando Bob said:

That's hot but still below safe max, what cooler/case/setup are you running? 

Now the CPU cooler is Snowman M-T6, Case is Modecom Harry 3.

I just saw that the only case ventilator is not ramping up and rotates at ~960 rpm.

With open sidepanel  the cpu temp is around 75C max.

[Vishera]

1 minute ago, Tan3l6 said:

Now the CPU cooler is Snowman M-T6, Case is Modecom Harry 3.

I just saw that the only case ventilator is not ramping up and rotates at ~960 rpm.

With open sidepanel  the cpu temp is around 73C

The cooler seems kinda fine,but that case is one of the worst case scenarios of airflow.

[Tan3l6]

1 minute ago, Vishera said:

The cooler seems kinda fine,but that case is one of the worst case scenarios of airflow.

Bought the case as it's affordable and can use longer videocards than most cases which have HDD racks.

Not a good case to use for cooling though, agreed.

[Zando_]

12 minutes ago, Tan3l6 said:

Now the CPU cooler is Snowman M-T6, Case is Modecom Harry 3.

I just saw that the only case ventilator is not ramping up and rotates at ~960 rpm.

With open sidepanel  the cpu temp is around 75C max.

Cooler is solid, just airflow in that case is yikes. Tried removing the front panel? Would probably be quieter and perform better than removing the side panel (especially if you have front fans. If you're down to mod the case, putting two fans in the front with a pair of cheap dust filters, then either drilling holes or just cutting them out so the fans can breathe through that front panel, would help a lot with temps. 

[Vishera]

1 minute ago, Tan3l6 said:

Bought the case as it's affordable and can use longer videocards than most cases which have HDD racks.

Not a good case to use for cooling though, agreed.

There are cases on the market that provide maximum airflow with a lot of space for wide GPUs,

and with space for 2 or 4 storage drives.

 

The Cooler Master MB511 is a cheap option,and there is a RGB version too.

[Tan3l6]

10 hours ago, Zando Bob said:

Tried removing the front panel?

Not yet, but good idea, I haven't had such extreme loads to get 86C. So haven't payed much attention to the case, after a check, the front panel indeed is removable, and 120mm fans installable.

---

No crashes for some time (not 2 hours as i previously miscalculated)

---

 

No crashes for a night of Prime95 Blend test. All is resolved.

[Tan3l6]

What the hell is this, now 4,5Ghz is unstable, testing on 4,4Ghz 

 

Is my CPU dying?