[CPU Question] Does this mean im stable ?

[Swerus]

Hello guys!
My first post on this forum, so i hope someone could help me out.
I recently overclocked my CPU and i got a couple of questions. I first clocked my i5-6600k from 3.9ghz to 4.5 and thought i was stable with 1.265v so i started push to 4.6ghz.

I clocked to 4.6ghz with the same vcore of 1.265 and ran some AIDA64 test, within minutes i got the blue crash screen, so i started pushing the volts by 0.005.
It was a slow climb, but i ended up on 1.305/1.310v and things seem to be stable after a 12 hours AIDA64 testing (I tested the cpu + cache) and not the RAM yet.

As i said, it was stable at 1.310v, but while running stress test the CPU reached 60-64*c in temperature quite fast, and it stayed there.
SO I put another 0.005v to total of 1.315v and started testing. The weird thing is that the system ran much smoother and with MUCH lower temperatures.
It took AIDA about 10 hours to reach the 64*c temperature, but with just 0.005v less it reached fast.

My first question: Does that mean i found my stable Vcore if it behaves this way ?
Second question: Could someone explain the noticeable jumps in temperature just because of a 0.005v push ?
Third question: As it feels that things are stable, should i try and add another 0.005v to see if it gets even better ? or do you think that i found the right spot ?

My last 2 AIDA64 tests ran stable. 1st test 15 hours and the 2nd test was on overnight for 14 hours.
Another question: *AIDA64 vs Prime95. I heard alot of people run AIDA for testing, others run Prime95.
I had 15 hours stable run with aida, but i dont know that much of Prime95, i got it downloaded and started it only 1 time.
I started the test with Prime95 but my worker#2 says that it is not active ( i got 4 workers in total ) could someone explain?


Thanks in advance!!!

 

[Organik]

Very nice temps under stress test. I would say confidently your set, Your OC is good to go, now go ahead and enjoy it. If you go to 1.400v I think you can really do some OCing damage. I wouldn't after seeing your tests just be happy and end the OC. But if you want more then go slow, in increments. good luck

 

[Swerus]

4.6Ghz and 1.315v, i alredy run 4 test with 13-15 hours each. The last one is also run with system memory stress.
My memory is at 2133, ive got a Corsair Vengeance LPX DDR-4 (2133) C13 kit. Is there anything i could do about it for better preformance or should i just leave it as it is ?

Another thing is that AIDA64 runs flawless without any errors. BUT, as soon as i start Prime95, my Worker#2 stops the second i run the test, the other 3 workers keep testing without any problems. Any suggestions on what it could be ?

Thanks for the reply.

 

[EarthDog]

Id leave the memory alone personally. The effort versus gains generally isnt worth the time.

As far as why p95 fails and aida doesnt.... they are different tests and one is more difficult to pass then the other. If the machine is stable for your uses, then you are stable.

As far as your original questions...

1. That beavior doesnt define stability. It defines nothing. In fact, that is incredibly odd...
2. Cant explain it... see #1
3. No. I wouldnt chasing ghosts with how your system responded. Stable is stable with these tests AFAIK.

 

[Swerus]

Id leave the memory alone personally. The effort versus gains generally isnt worth the time.

As far as why p95 fails and aida doesnt.... they are different tests and one is more difficult to pass then the other. If the machine is stable for your uses, then you are stable.

As far as your original questions...

1. That beavior doesnt define stability. It defines nothing. In fact, that is incredibly odd...
2. Cant explain it... see #1
3. No. I wouldnt chasing ghosts with how your system responded. Stable is stable with these tests AFAIK.

I heard there is X diferent versions of Prime95, and the latest one are to hardcore for personal use PC's, is that correct ?

But what does it accually means that my Worker#2 stops immediately when i start the "small-test", does that my 2nd core isnt following the rest or is underpowered ? How much and how could this affect my PC use in a long run ?

I didnt touch the memory because as you said, i saw somewere that the gains isnt worth the effort... But, im at 4.6Ghz, should i still leave the RAM voltage to auto or 1.200

 

[EarthDog]

1. There are different versions, yes. The latest isnt for 'hard core' users but the latest doesnt stress test harder due to the instruction set (avx/fma3) it uses. As i said, if you are stable for YOUR uses, then its fine. If you want something closer to p95 while using aida, testing with ONLY FPU checked off and see what happens.

2. It means that core isnt stable. It shouldnt affect your pc in any way unless you use those instruction sets heavily.

 

[Swerus]

1. There are different versions, yes. The latest isnt for 'hard core' users but the latest doesnt stress test harder due to the instruction set (avx/fma3) it uses. As i said, if you are stable for YOUR uses, then its fine. If you want something closer to p95 while using aida, testing with ONLY FPU checked off and see what happens.

2. It means that core isnt stable. It shouldnt affect your pc in any way unless you use those instruction sets heavily.

I heard that the unstable core could be fixed with a little bump on vcore ?
Do you mean that i should run with FPU checked only or with everything checked except the FPU?

If you mean with FPU only on, it runs fine i guess. 61-59-59-60(c) temperature.
Here is the screen, you mean like this right ? https://prnt.sc/ki0elw

 

[EarthDog]

It can be fixed with a bump in vcore likely, sure, but if it is already stable for your uses, why would you do that?

Yes, I mean FPU only (please host images here at the site like you did originally and not third party - those will go away. )

 

[trents]

The latest version of Prime95 is one of the most demanding stress tests on the planet. It makes heavy use of AVX 2 instruction sets while stressing your CPU. Many other stress tests use only AVX 1 instructions or no AVX at all. AVX (Advanced Vector Extension) is typically found in specialized applications such as those that handle very precise scientific calculations. AVX is not commonly found in everyday applications. But if you can pass a stress test which deploys AVX it's kind of insurance that you truly are stable and will not have crashes when running normal applications. AIDA 64 does use some AVX but I'm not sure if it uses AVX 1 or AVX 2 or how much of either one it uses during stress tests.

Many or most of us will make use of a bios parameter called "AVX offset" which lowers the CPU multiplier by a set amount when it encounters AVX instruction sets. For instance, if your CPU multiplier is set for 46x an AVX offset of 1x would cause it to throttle down to 45x. An AVX offset of 2x would cause it to throttle down to 44x. The same amount of throttling will occur regardless of whether AVX 1 or AVX 2 is encountered. My suggestion is you try an AVX offset of 1x and retest with Prime95.

It's always best to use a variety of stress test tools. Another good one is Realbench (test for 4 hr.) which uses AVX 1 and is very "real world". It uses demanding video rendering tasks to stress.

Still another is OCCT (test for 3 hr.) which uses AVX but I'm not sure which version of AVX. It has some neat built in voltage and temp monitoring tools and is pretty much as strenuous as newer versions of Prime95.

I might also suggest using an older version of Prime 95 without AVX 2 (v. 27.7) or no AVX at all (26.6). Prime95 stress tests should be run for 2 hr. to be confident of stability. What you want is an overclock that will pass both non AVX stress tests at the highest level of overclock possible on safe voltages but also pass AVX stress tests with a small AVX offset.

If you want a stress test that is "quick and dirty" try IntelBurnTest v.2.54 on "Very High" settings which will drive temps to their max very quickly. Set it to Very High and 10-20 rounds. Not sure but I think it may use AVX 1.

I cannot give you a definite answer as to why the overclock runs cooler at the slightly higher voltage except to say that modern processors handle power in a very dynamic way and at 1.315 you may have allowed it to slip into power management level that is more favorable. The other possibility is that it started thermal throttling but your AIDA64 graph doesn't seem to indicate that.

Finally, what is your LLC (Load Line Calibration) level set to? Playing with that can help find the best balance between running the lowest possible stable voltage when running easy computing tasks while supplementing the voltage under load to keep it stable during demanding tasks.

Hope this helps.

Oh, yes. Please create a "Signature" or "Sig" that lists your system components. You will find the tool under "Settings" down the left column. Settings is found at the top of the page right corner.

 

[Swerus]

It can be fixed with a bump in vcore likely, sure, but if it is already stable for your uses, why would you do that?

Yes, I mean FPU only (please host images here at the site like you did originally and not third party - those will go away. )

They dont go higher than 61/62-60-60-60~

 

[EarthDog]

You ran it for 2.5 minutes bud....... give it some time (15-30 mins) and see what happens.

I'm not going to get into the stress testing thing as far as what is what or best or even other options that will only serve to confuse. My best advice here is to stick with whatever is working and don't worry about if it doesn't-pass-this-but-passes-that-am-I-stable type of mental mind bend. Using a bunch of different apps and versions really just confuses most people.

 

[Swerus]

The latest version of Prime95 is one of the most demanding stress tests on the planet. It makes heavy use of AVX 2 instruction sets while stressing your CPU. Many other stress tests use only AVX 1 instructions or no AVX at all. AVX (Advanced Vector Extension) is typically found in specialized applications such as those that handle very precise scientific calculations. AVX is not commonly found in everyday applications. But if you can pass a stress test which deploys AVX it's kind of insurance that you truly are stable and will not have crashes when running normal applications. AIDA 64 does use some AVX but I'm not sure if it uses AVX 1 or AVX 2 or how much of either one it uses during stress tests.

Many or most of us will make use of a bios parameter called "AVX offset" which lowers the CPU multiplier by a set amount when it encounters AVX instruction sets. For instance, if your CPU multiplier is set for 46x an AVX offset of 1x would cause it to throttle down to 45x. An AVX offset of 2x would cause it to throttle down to 44x. The same amount of throttling will occur regardless of whether AVX 1 or AVX 2 is encountered. My suggestion is you try an AVX offset of 1x and retest with Prime95.

It's always best to use a variety of stress test tools. Another good one is Realbench (test for 4 hr.) which uses AVX 1 and is very "real world". It uses demanding video rendering tasks to stress.

Still another is OCCT (test for 3 hr.) which uses AVX but I'm not sure which version of AVX. It has some neat built in voltage and temp monitoring tools and is pretty much as strenuous as newer versions of Prime95.

I might also suggest using an older version of Prime 95 without AVX 2 (v. 27.7) or no AVX at all (26.6). Prime95 stress tests should be run for 2 hr. to be confident of stability. What you want is an overclock that will pass both non AVX stress tests at the highest level of overclock possible on safe voltages but also pass AVX stress tests with a small AVX offset.

If you want a stress test that is "quick and dirty" try IntelBurnTest v.2.54 on "Very High" settings which will drive temps to their max very quickly. Set it to Very High and 10-20 rounds. Not sure but I think it may use AVX 1.

I cannot give you a definite answer as to why the overclock runs cooler at the slightly higher voltage except to say that modern processors handle power in a very dynamic way and at 1.315 you may have allowed it to slip into power management level that is more favorable. The other possibility is that it started thermal throttling but your AIDA64 graph doesn't seem to indicate that.

Finally, what is your LLC (Load Line Calibration) level set to? Playing with that can help find the best balance between running the lowest possible stable voltage when running easy computing tasks while supplementing the voltage under load to keep it stable during demanding tasks.

Hope this helps.

Oh, yes. Please create a "Signature" or "Sig" that lists your system components. You will find the tool under "Settings" down the left column. Settings is found at the top of the page right corner.

Thanks alot for the detailed information. I will try and see how it turns out with the AVX setting, ill try the other stress tests first..
I have all power saving settings off.
If the AIDA64 graph doesnt notice the thermal throt, Real Temp should ? I got it running on the side to follow the temperature.
My LLC level is set to 5 ( i think i got 8 diferent levels to pick from ) Im not that good with LLC levels so i put it on 5 because i saw that level 5 was most recommended, is it right or should i switch it ?
I also noticed in CPU-Z that even tho i set my vCore to 1.315, cpu-z caps it at 1.312 and never goes above that, but goes down to 1.265-95 and jumps up.

- - - Auto-Merged Double Post - - -

You ran it for 2.5 minutes bud....... give it some time (15-30 mins) and see what happens.

I'm not going to get into the stress testing thing as far as what is what or best or even other options that will only serve to confuse. My best advice here is to stick with whatever is working and don't worry about if it doesn't-pass-this-but-passes-that-am-I-stable type of mental mind bend. Using a bunch of different apps and versions really just confuses most people.

Yea i know I took the pct just to see that i understood you correct.

Could it be that my ram is running slower than the core speed in any way ? idk what im talking about haha.
Check this, and my ram volt is at 1.2000

 

[EarthDog]

Well, its your time... I don't chase stress tests ghosts and glad I learned not to years ago.

If the AIDA64 graph doesnt notice the thermal throt, Real Temp should ?

They both will. Realtemp will log it though. That processor is good to 90C and throttles at 100C.

My LLC level is set to 5 ( i think i got 8 diferent levels to pick from ) Im not that good with LLC levels so i put it on 5 because i saw that level 5 was most recommended, is it right or should i switch it ?

This depends on what your goals are with LLC. LLC is there to mitigate the difference between what was set in the bios and load voltage in windows. LLC prevents that voltage from sagging or drooping (its called vdroop). What I do and reccomend others do with it is to set it so the voltage under load matches what you set it to. This way when in the BIOS if you set to 1.3V, you want your load voltage to match.

I also noticed in CPU-Z that even tho i set my vCore to 1.315, cpu-z caps it at 1.312 and never goes above that, but goes down to 1.265-95 and jumps up.

CPU-z is software and isn't always accurate (software in general)...but its the best we have. When you say 'goes down and jumps up' is this under load when you see it or just randomly jumping around?

Could it be that my ram is running slower than the core speed in any way ? idk what im talking about haha.
Check this, and my ram volt is at 1.2000

No, just as the brakes on your car have nothing to do with the horn.

Joke aside, they are unrelated. Your ram is running its correct speed.. CPUz reports the actual rate not the DOUBLE DATA RATE (DDR RAM we all have). If you do some quick math, you will notice that 1066 is half of 2133... or 2133 is DOUBLE 1066.

 

[trents]

Now that Sig. You can refer to one of ours if you want guidance as to what is helpful for us to see in your component list.

The original design of LLC was to offset the voltage drop we see when moving from what we declare it to be in bios to what the net amount is under load. Some like to use it to supplement the bios vcore amount, to give a little extra. So I would say experiment with LLC until you find the level that keeps your vcore at or slightly above what you have set it to in bios throughout a stress test cycle.

When you say you "have all power savings off" what specifically do you refer to in bios? That's probably not necessary or advisable as it once was several years ago.

 

[Swerus]

The latest version of Prime95 is one of the most demanding stress tests on the planet. It makes heavy use of AVX 2 instruction sets while stressing your CPU. Many other stress tests use only AVX 1 instructions or no AVX at all. AVX (Advanced Vector Extension) is typically found in specialized applications such as those that handle very precise scientific calculations. AVX is not commonly found in everyday applications. But if you can pass a stress test which deploys AVX it's kind of insurance that you truly are stable and will not have crashes when running normal applications. AIDA 64 does use some AVX but I'm not sure if it uses AVX 1 or AVX 2 or how much of either one it uses during stress tests.

Many or most of us will make use of a bios parameter called "AVX offset" which lowers the CPU multiplier by a set amount when it encounters AVX instruction sets. For instance, if your CPU multiplier is set for 46x an AVX offset of 1x would cause it to throttle down to 45x. An AVX offset of 2x would cause it to throttle down to 44x. The same amount of throttling will occur regardless of whether AVX 1 or AVX 2 is encountered. My suggestion is you try an AVX offset of 1x and retest with Prime95.

It's always best to use a variety of stress test tools. Another good one is Realbench (test for 4 hr.) which uses AVX 1 and is very "real world". It uses demanding video rendering tasks to stress.

Still another is OCCT (test for 3 hr.) which uses AVX but I'm not sure which version of AVX. It has some neat built in voltage and temp monitoring tools and is pretty much as strenuous as newer versions of Prime95.

I might also suggest using an older version of Prime 95 without AVX 2 (v. 27.7) or no AVX at all (26.6). Prime95 stress tests should be run for 2 hr. to be confident of stability. What you want is an overclock that will pass both non AVX stress tests at the highest level of overclock possible on safe voltages but also pass AVX stress tests with a small AVX offset.

If you want a stress test that is "quick and dirty" try IntelBurnTest v.2.54 on "Very High" settings which will drive temps to their max very quickly. Set it to Very High and 10-20 rounds. Not sure but I think it may use AVX 1.

I cannot give you a definite answer as to why the overclock runs cooler at the slightly higher voltage except to say that modern processors handle power in a very dynamic way and at 1.315 you may have allowed it to slip into power management level that is more favorable. The other possibility is that it started thermal throttling but your AIDA64 graph doesn't seem to indicate that.

Finally, what is your LLC (Load Line Calibration) level set to? Playing with that can help find the best balance between running the lowest possible stable voltage when running easy computing tasks while supplementing the voltage under load to keep it stable during demanding tasks.

Hope this helps.

Oh, yes. Please create a "Signature" or "Sig" that lists your system components. You will find the tool under "Settings" down the left column. Settings is found at the top of the page right corner.

I just checked my bios and i cant find an AVX option, i dont think my motherboard has it.
Im running a ASUS z170-p motherboard ( Not the pro version )

I tried to raise the vcore to 1.320 but it still is limited to 1.312. Is it because of my LLC lvl 5 ? or what could it be ?
Im at 4.6ghz with 1.320v right now BUT i got a question, i havent got into the CPU VCCIO Voltage and CPU System Agent Voltage settings yet, how important is it that they follow the vCore settings ? If so, how much should i raise them to ? At the moment they both are on auto and i think the VCCIO is at 0.931 or something like that.

 

[EarthDog]

It may not have it...

LLC does not limit voltage.

Vccio and system agent do not need to be associated with vcore.

 

[Swerus]

It may not have it...

LLC does not limit voltage.

Vccio and system agent do not need to be associated with vcore.

Hello, thanks for the reply. The first 1-2 days the OC felt stable. ytoday i ran cinebench and aida afterwards. While running i experienced small freezes / studders.
I went into my bios and took some pictures of what i think is most important, could you please take a look and see how it looks like and if you would suggest me to change any of the following settings.
Thanks again for your help.

 

[EarthDog]

Sometimes it does stutter a bit in AIDA in my experience... I do not think there is a problem. What about in NORMAL use? Again, chasing stress tests is time consuming with little returns in most cases. If its stable for YOUR USES, then its stable.

 

[Swerus]

Sometimes it does stutter a bit in AIDA in my experience... I do not think there is a problem. What about in NORMAL use? Again, chasing stress tests is time consuming with little returns in most cases. If its stable for YOUR USES, then its stable.

Well, since i started with the OC right after i installed windows on my new nVMe SSD, i honestly had no time to "regular use" my pc. Past 5-6 days ive spent reading and finding the right information for overclocking. Only game i installed is running fine i think

Did you see the pictures above ? Does everything look fine ?
Thanks