Overclock unstable at idle -.-

[karlpiuemeno]

HI, I'm currently having a strange behaviour with my overclock.
On full load, my 6600k is rock stable at 4,7 GHz (tested with aida) , 1,36vcore (adaptive mode), LLC 4, optimized phase control... but sometimes it crashes when doing low load things... such as a file copy or just browsing...
I don't undestand why... any hint?
my configuration:
PSU: Seasonic G 650
MoBO: Asus Z170-A
CPU: Core i5 6600K
Fan: CoolerMaster Hyper 212 evo
Ram: Corsair LPX 16 GB 3000 @ 3200

[BoredErica]

Set ram to stock or under stock and see if that fixes it. Also, Aida doesn't make an overclock "rock stable". CPU may still be unstable, but seemingly random crashes could be ram.

 

2 cents

[karlpiuemeno]

1 minute ago, Dark_wizzie said:

Set ram to stock or under stock and see if that fixes it. Also, Aida doesn't make an overclock "rock stable". CPU may still be unstable, but seemingly random crashes could be ram.

 

2 cents

Already done... Ram doesn't change the behaviour..

[BoredErica]

Well if it's the CPU you know what to do, set to stock to see if it's stable. If you're sure it's stable, then your overclock isn't stable. Test with something tougher than Aida like x264 test (not x264 bench), or maybe Realbench, 5-12 hours. GL. 

[GM_Peka]

Try a 6+ hour test with Prime95 Blend (AVX2 isn't a problem with skylake)

 

I personally don't think Asus Realbench is good enough, from my own personal experience, I did a 12 hour run of Realbench without error, then I crashed while doing something random, I then ran Prime95 for 5 minutes and then error so I don't think Realbench is very tough for Skylake.

 

It took me 1.280volts to get 4.4 stable in the end, LLC 5, and I set the CPU VCCIO and System agent voltage to 1.5 volts for good measure, cause some chips can't handle 3000mhz RAM with stock Memory controller volts, although I'm not sure that's a problem for you since you still crashed without XMP.

 

Oh, and just an FYI, when running anything that uses AVX2 with adaptive voltage, your CPU will be supplied with +0.012 volts, because AVX2 is tougher generally.

 

[karlpiuemeno]

17 minutes ago, Pvt. 8Ball said:

Try a 6+ hour test with Prime95 Blend (AVX2 isn't a problem with skylake)

 

I personally don't think Asus Realbench is good enough, from my own personal experience, I did a 12 hour run of Realbench without error, then I crashed while doing something random, I then ran Prime95 for 5 minutes and then error so I don't think Realbench is very tough for Skylake.

 

It took me 1.280volts to get 4.4 stable in the end, LLC 5, and I set the CPU VCCIO and System agent voltage to 1.5 volts for good measure, cause some chips can't handle 3000mhz RAM with stock Memory controller volts, although I'm not sure that's a problem for you since you still crashed without XMP.

 

I'm with XMP !
I used to work with my system overclocked @ 4,6 GHz, 1,32 vcore ... and it was really stable under all kinds of load (I make video editing)... so I wanted more and tried the 4,7 scenario... but the instability is on low load  
I'll investigate more on the memory controller side, maybe I'm stressing it too much

[karlpiuemeno]

1 hour ago, Pvt. 8Ball said:

Try a 6+ hour test with Prime95 Blend (AVX2 isn't a problem with skylake)

 

I personally don't think Asus Realbench is good enough, from my own personal experience, I did a 12 hour run of Realbench without error, then I crashed while doing something random, I then ran Prime95 for 5 minutes and then error so I don't think Realbench is very tough for Skylake.

 

It took me 1.280volts to get 4.4 stable in the end, LLC 5, and I set the CPU VCCIO and System agent voltage to 1.5 volts for good measure, cause some chips can't handle 3000mhz RAM with stock Memory controller volts, although I'm not sure that's a problem for you since you still crashed without XMP.

 

Oh, and just an FYI, when running anything that uses AVX2 with adaptive voltage, your CPU will be supplied with +0.012 volts, because AVX2 is tougher generally.

 

Well... it looks like 4,6 GHz wasn't really 100% stable: I had no crashes, Aida and x264 were ok, I worked for hours with premiere and after effects, etc.. but now Prime95 blend is revealing errors after few minutes !
I restarted from the beginning: disabled xmp, ram at stock, etc, and it looks like it needs more vcore!
After some tries, now I'm at 1,360 (but cpu-z says 1.369) and Prime95 is passing the tests

So ok... prime95 is the right test for me. Thankyou!

[GM_Peka]

12 minutes ago, karlpiuemeno said:

Well... it looks like 4,6 GHz wasn't really 100% stable: I had no crashes, Aida and x264 were ok, I worked for hours with premiere and after effects, etc.. but now Prime95 blend is revealing errors after few minutes !
I restarted from the beginning: disabled xmp, ram at stock, etc, and it looks like it needs more vcore!
After some tries, now I'm at 1,360 (but cpu-z says 1.369) and Prime95 is passing the tests

So ok... prime95 is the right test for me. Thankyou!

Yeah, Prime95 is tough as fuck, but atleast you know your system is stable.

[karlpiuemeno]

Ok, now I'm really confused: I was able to find the sweet spot for my overclock, after lots of reboots and prime95 tests
4,5 GHz , 1,30 Vcore (adaptive mode) , LLC 5,  XMP with ram @3000 
with that configuration, no errors or glitches in prime95 for more than an hour of workload with the magic 8 k calculations... cpu overloaded as hell...
but today it freezed 2 times on idle:
1) it was idling with my facebook feed, the display was off for power saving, and I wasn't able to wake up the screen
2) I sent to sleep the whole pc , the display went off, but the pc itself was on.

It's a multiplier overclock, so teorically it doesn't need more vcore on lower clocks (I would use offset mode if I did a bclk overclock) ... 
Any hint?
Epu and other asus stuff are disabled... it's saving power only with intel built in stuff...

I found in another forum a user with the same issue... and he says it's a platform issue, and he fixed by giving 0.06 v more to ram... (he is stock, no xmp)...
I'm trying... let's see...

[runit3]

2 hours ago, karlpiuemeno said:

Ok, now I'm really confused: I was able to find the sweet spot for my overclock, after lots of reboots and prime95 tests
4,5 GHz , 1,30 Vcore (adaptive mode) , LLC 5,  XMP with ram @3000 
with that configuration, no errors or glitches in prime95 for more than an hour of workload with the magic 8 k calculations... cpu overloaded as hell...
but today it freezed 2 times on idle:
1) it was idling with my facebook feed, the display was off for power saving, and I wasn't able to wake up the screen
2) I sent to sleep the whole pc , the display went off, but the pc itself was on.

It's a multiplier overclock, so teorically it doesn't need more vcore on lower clocks (I would use offset mode if I did a bclk overclock) ... 
Any hint?
Epu and other asus stuff are disabled... it's saving power only with intel built in stuff...

I found in another forum a user with the same issue... and he says it's a platform issue, and he fixed by giving 0.06 v more to ram... (he is stock, no xmp)...
I'm trying... let's see...

Does it happen when the voltage is in manual mode? Could be dropping out too low while in adaptive. Also, never run synth benchmarks while your voltage is in any other mode besides manual.

[karlpiuemeno]

42 minutes ago, runit3 said:

Does it happen when the voltage is in manual mode? Could be dropping out too low while in adaptive. Also, never run synth benchmarks while your voltage is in any other mode besides manual.

It looks like I ignored the fact that Windows SUCKS.
I was obsessed by my overclock, but it looks like to be more a storage controller issue:
the first sleep after a reboot is ok ----> than I wake and I begin to have problems (apps won't start or close, etc) ---> I look at task manager: 25% (so, ONE core at full) used by the process "System and compressed memory"...
---> it won't go to sleep.
SO...
Googling suggests it's an Intel Rapid Storage issue...
OK_ let's investigate the software side... maybe my overclock and adaptive mode were absolutely OK.