• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Drago's First Intel Overclocking Adventure

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

DragoXT

Member
Joined
Jun 14, 2005
Thanks to EarthDog, MattNo5ss, and others for the help with the XMP memory issue. I have that sorted and things are good. It has been busy but i am itching to see what my chip can OC to.


System Specs:
i7 6700K @ stock
GA-Z170X-UD5 w/F3 bios
2x8gb Corsair Vengeance LPX 3000Mhz
GB G1 Gaming GTX 970
Sandisk Extreme Pro 480gb SSD
Seasonic X750
Cryorig R1 Ultimate Cpu Cooler
Phanteks Enthoo Evolv ATX Case
Win 10 Pro


Please review my OCing Data Spreadsheet from Google Sheets for info as i have questions.

My goal right now is to get this chip stock speeds with manual settings. So far i have worked on getting my voltages manually set. The more things i can leave auto without messing up my chance at a good OC i am for.

Do you know what the default vcore for Skylake is? With this GigaByte board auto has the default voltage acording to the BIOS at 1.284 - 1.3, and EasyTune shows 1.284 - 1.302 and voltage points on mobo 1.322. These are at various loads. When i manually set my voltage and left LLC at auto omg holy vdroop batman. 1.280v went down past 1.19v. Since that voltage seemed stable i ran 1.2v with LLC at high. This yields a full load vcore swing according to software down to 1.188v. I really need to take apart my case and get the voltage point readings and see what they really are when i have that kind of time.

I know on my AMD systems i turned off cool and quiet to OC. Do i need to do the same thing to the Cstates for Intel? I have left that stuff to auto and it appears that manually setting voltage stopped the go to lower speeds and voltages at idle. I have windows power options at high performance right now. With vcore auto the speed wouldnt change but the vcore would drop under 1v. My ambient temps are 22c and my cpu cores were idling at 21-22c with speedstep. Now my idle temps are 25c to 26c without speedstep. Not bad but i wouldnt mind having the ability to use the low power at idle, if there is a way to do this and OC let me know.

As always i appreciate the help and thanks for the advice.
 
Ok i have been tooling away with my Overclocking Quest. OMG Intel you are giving me a headache with this chip. I need some advice on VCCIO and VCCSA settings. Stock settings are low, bumping them sometimes seems to help but other times it does not. VCCIO is supposed to be IMC voltage right? If so normally running your RAM overclocked will require extra volts to the IMC to keep things stable. I dont want to go nuts upping voltage unecessarily, just some advice on when to push the volts up for both settings.

Traditionally you run several tests that will stress the CPU for stability but i always like to stress my RAM a little bit to. If my cpu tests pass but my RAM tests dont then i bump the IMC voltage. This worked great on AMD, P95 and OCCT for cpu and wprime and SuperPI for RAM.

My issue now is that i get OCCT and WPrime done but P95 blend fails on worker 7 or 8 or both. No matter VCCIO or VCCSA i run seem to make a diff. I keep bumping vcore but it doesnt seem to make it stable. Seems like the amount of voltage required to get to the next step is quite high. Can someone let me know if this is normal for Intel?
 
Many will recommend AIDA64 stability test for the newer Intel or even an older p95. The new P95 uses a pretty heavy load and results in what some say is unnecessary heat /torture/current draw for the CPU. It's up to you in the end but you'll likely get a higher OC leaving P95 out of the question. If you're running high speed ram SA and IO can be raised a bit to help with stability but usually default/auto is good
 
What version/revision of Prime95 are you running? Because most of the newer versions aren't really recommended for newer AMD and Intel setups as they sometimes overpower the CPU with too much current or too much voltage, resulting in temps that are too high and sometimes damage to the CPU.

I personally run Prime95 v25.11 when I'm testing for stability, mainly just because that was the version I found when I first got into overclocking and benchmarking. And I've not been bothered enough to look for a newer version.

Edit: According to some searching I just did it's recommended to not go beyond Prim95 v26.6 for stress testing.
 
Last edited:
What version/revision of Prime95 are you running? Because most of the newer versions aren't really recommended for newer AMD and Intel setups as they sometimes overpower the CPU with too much current or too much voltage, resulting in temps that are too high and sometimes damage to the CPU.

I personally run Prime95 v25.11 when I'm testing for stability, mainly just because that was the version I found when I first got into overclocking and benchmarking. And I've not been bothered enough to look for a newer version.

Edit: According to some searching I just did it's recommended to not go beyond Prim95 v26.6 for stress testing.

Note that the reason why newer Prime95s generate so much heat is because they use the new AVX instruction set. AVX allows programs to work on 8 single precision floating point numbers at the same time (or 4 double precision), compared to 4 SP/2 DP for the SSE family of instructions.

Use of AVX is required to get the full floating point performance from Sandybridge and newer CPUs.

It also generates more heat because the CPU is doing a lot more work (if you only use SSE, half of the floating point units will be idle all the time).

Now whether you can say a CPU is stable if it's only stable when half of the floating point units are idle... well obviously you can say whatever you want, but if it's not AVX-stable, your overclock is not stable enough for programs to use all of your CPU's features.

I don't know how common AVX is in real life at the moment, but it won't be long before many numerically-intensive programs start using AVX.

P95 isn't doing anything unusual. It's just one of the first numerically intensive programs (others include encoding, etc) to use AVX. It isn't the only one either. Many programs in scientific computing already use AVX.

IMHO, stress testing is about using the most stressful program there is for testing, because if your CPU is stable even with that program, you can be more confident that any real program you use will run stable as well. I don't think pre-AVX P95 is a valid stress testing program anymore. It's like using a 2-threaded version for testing because a 4-threaded version makes your CPU overheat.
 
I am running P95 28.7. Seems really weird that i can get my system "stable" and suddenly have issues with P95 dropping a worker a few seconds into the run. I have had to go back to 4.3 and try to change settings for it as it drops workers in P95. I miss the old days of P95 and the red indicator that a worker stopped. I double checked all my overclock settings and 4.0-4.2 are perfect in 20min of P95 Blend and SmallFFT.

My goal with any stress testing software is to not over do it, but make it so that my games will not crash on me when i OC. The heat output of P95 is not that bad. The temps i am putting in my spreadsheet are the hottest core which is 3c above the lowest two, and 2c above the other core in temps.

Does anyone have any advice on when to bump core voltage, VCCIO and VCCSA when it comes to different stress testing programs failing in different ways?
 
Note that the reason why newer Prime95s generate so much heat is because they use the new AVX instruction set. AVX allows programs to work on 8 single precision floating point numbers at the same time (or 4 double precision), compared to 4 SP/2 DP for the SSE family of instructions.

Use of AVX is required to get the full floating point performance from Sandybridge and newer CPUs.

It also generates more heat because the CPU is doing a lot more work (if you only use SSE, half of the floating point units will be idle all the time).

Now whether you can say a CPU is stable if it's only stable when half of the floating point units are idle... well obviously you can say whatever you want, but if it's not AVX-stable, your overclock is not stable enough for programs to use all of your CPU's features.

I don't know how common AVX is in real life at the moment, but it won't be long before many numerically-intensive programs start using AVX.

P95 isn't doing anything unusual. It's just one of the first numerically intensive programs (others include encoding, etc) to use AVX. It isn't the only one either. Many programs in scientific computing already use AVX.

IMHO, stress testing is about using the most stressful program there is for testing, because if your CPU is stable even with that program, you can be more confident that any real program you use will run stable as well. I don't think pre-AVX P95 is a valid stress testing program anymore. It's like using a 2-threaded version for testing because a 4-threaded version makes your CPU overheat.

Not that I totally disagree, many feel that a lot depends on what you're using your PC for. For gaming and light tasks P95 w/avx2 isn't necessary. There isn't a game available that will tax the CPU like P95, the authors for p95 intentionally maxed it's output for specific reasons. P95 isn't necessarily a stress test software but a distributed computing software dedicated to finding prime numbers with actual cash prizes for those who are successful.
IMO if someone wants to use P95 that's fine, I do on my systems. I just know that most don't and in certain communities (haswell-E) is highly frowned upon. They'll claim that you can damage your CPU using it. Which in part may be true if you have an excessive OC. So, these users will use a combination of software such as Aida64, Realbench, HCI memtest etc..

I am running P95 28.7. Seems really weird that i can get my system "stable" and suddenly have issues with P95 dropping a worker a few seconds into the run. I have had to go back to 4.3 and try to change settings for it as it drops workers in P95. I miss the old days of P95 and the red indicator that a worker stopped. I double checked all my overclock settings and 4.0-4.2 are perfect in 20min of P95 Blend and SmallFFT.

My goal with any stress testing software is to not over do it, but make it so that my games will not crash on me when i OC. The heat output of P95 is not that bad. The temps i am putting in my spreadsheet are the hottest core which is 3c above the lowest two, and 2c above the other core in temps.

Does anyone have any advice on when to bump core voltage, VCCIO and VCCSA when it comes to different stress testing programs failing in different ways?

I believe it's not likely that those voltages are the issue with your testing, P95 doesn't max out the memories . Most likely it's just more core voltage needed.
If you run HCI memtest and are having failure then it could be ram or cache related. That's when you'll want to raise the voltages mentioned.
 
I believe it's not likely that those voltages are the issue with your testing, P95 doesn't max out the memories . Most likely it's just more core voltage needed.
If you run HCI memtest and are having failure then it could be ram or cache related. That's when you'll want to raise the voltages mentioned.

Thanks for the info, it has been a while for me OCing and my tools are different now. I wish there was a SuperPi that was multicore as that would show an unstable OC fast if the RAM or IMC was not stable. It is such a pain to have 8 instances of the software running and affinity's set on all of them.

I did get my 4.3ghz stable at 1.245v. I will see if i can back it down on the VCCIO, cause if i dont need it i dont need it, no reason to add more heat.
 
Ocing is much easier than it used to be. OCING was once an art-form now its easy to do in copmparision. But nice one have fun in ocing
 
Not that I totally disagree, many feel that a lot depends on what you're using your PC for. For gaming and light tasks P95 w/avx2 isn't necessary. There isn't a game available that will tax the CPU like P95, the authors for p95 intentionally maxed it's output for specific reasons. P95 isn't necessarily a stress test software but a distributed computing software dedicated to finding prime numbers with actual cash prizes for those who are successful.

Certainly. I guess the natural question to that is, why bother overclocking the CPU if it's only going to be used for gaming and light tasks?

Things like encoding and video processing will probably start using AVX soon, if they haven't already. It would seem that for people who are actually overclocking because they want to run CPU-intensive programs, including AVX in stress testing is probably a good idea.

If it causes the CPU to overheat, that just means the overclock is too high or cooling is not good enough. The fact that less stressful programs don't cause the CPU to overheat just means, well, they are less stressful programs. It's not like P95 is doing something extraordinary to overheat the CPU. Like you said, it's not designed to be a stress test. It just tries to compute PI, and does it in the most efficient way possible.
 
Back