6700k overclock, seemed too easy, I'm suspicious now

K

kandor

Gawd
Joined
Aug 24, 2005
Messages
632
Hi,

I just built some systems with asus z170 sabertooth mark 1, i7 6700k and 64gb of G.skill 2800 ddr4. Cooled with an H100i GTX. I planned on a mild overclock to 4.5.
Based on reading hear and elsewhere it sounded like getting the ram at 2800 was going to be unlikely. Other people / reviews showed pretty high cpu voltage and temps required to overclock up around 4.7.
Settings I used are below. I'm using adaptive mode without really knowing why, which probably isn't smart. Except it seems to working very well.

So far I have only stress tested with XTU, cinebench and handbrake. So, a couple of questions: I'm not clear how to test with handbrake - I just found a dvd iso and converted to MP4, took about 3 hours. Cinebench gave great scores as well.
But, I'd like to stress it harder. I read that prime95 or aida64 aren't as good for testing skylake, but I have no way to know this. Maybe I don't even need 1.32 volts? What's a better way to ensure a stable overclock here? Stability is the most important thing for me. Also, temps on the 6700k are max at 60 on coretemp. Which seems lower than others report. i.e. I'm suspicious why everything was so easy.

I used the following settings:
Ai overclock tuner = manual
BCLK freq = 100 mhz
Asus multicore enhancement = disabled
CPU core ratio = sync all cores
Core ratio limit = 45
CPU core/cache voltage = adaptive mode
Offset sign = +
Total adaptive mode CPU core voltage = 1.332
DRAM voltage = 1.35
DRAM speed = 2800
CAS 15
RAS 15
ACT 35
 
Last edited:
May have just got lucky and won the silicon lottery. I see some people doing x264 encodes, prime95 (of course) and some gaming benches to stress the IPC some. Can try the IBT as well if you'd like.
 
Getting Skylake to behave at 4.5GHz doesn't take much to achieve at all. Your settings are in line with what I've seen at those clocks. It takes a lot more to go to 4.7GHz and beyond. Getting the RAM to clock high is in part the motherboard and in part the CPU's IMC. The Sabertooth Z170 Mark 1 sucks at clocking memory in my experience. 2800MHz is about the highest clock you are likely to see. I can get 3600MHz on my test CPU up to 4.7GHz. So far at 4.8GHz it won't do that. 2800MHz is the most I've seen there.
 
Ok, cool - thanks for the info. The G.skill is rated 2800, so I wasn't going to try it faster. But I had been informed earlier not to expect 64GB to go beyond 2133, so I was happy that it works. But these are new builds to me, so I'm not totally trusting it all yet.
 
kandor said:
Ok, cool - thanks for the info. The G.skill is rated 2800, so I wasn't going to try it faster. But I had been informed earlier not to expect 64GB to go beyond 2133, so I was happy that it works. But these are new builds to me, so I'm not totally trusting it all yet.
Click to expand...

I think people are saying crap like that based on their experiences with X99 and Haswell-E. Skylake and Z170 are much better about getting higher memory clocks.
 
I also find the 6700k easy to overclock and get stability compared to x99.
Love the fact that cache volts are now tied with core clocks. Makes it easier to overclock cache.
 
Hi Dan_D, Thanks for the info, it was from the thread below that I posted before I ordered parts:

So what in your opinion is the best way to stress test skylake for stability. I know it's only 4.5, but I need 100% stability or I'll just go back to stock clocks. Up until skylake I exclusively relied on prime95, but now I'm reading in several places that prime and aida don't properly test skylake.

I'm new to handbrake and didn't find a guide on how to best use it for stress testing.

Thanks

kandor

http://hardforum.com/showpost.php?p=1041967479&postcount=2
 
definitively Asus RealBench its actually the best software out there for stability test. be sure to do at least 5 or 6 consecutive Benchmark runs as it work stressing different parts of the CPU (and/or GPU if wanted) exploiting possible instabilities faster..
 
Realbench 1 hour seems to be the sweetspot for 24/7 stability.
The guys on OCN Haswell-e thread swears by it.
 
So, realbench stress test for an hour. Or is it better to keep running the benchmark.

At 4.5ghz my machine score 91,000, which is ok I guess. Not too familiar with realbench as of yet.
 
*edit three questions left:

I've left the CPU System Agent voltage and the CPU VCCIO voltage on auto, even though I have 4x16 (64gb) ram. I've read that I should hard code these but since I'm not sure on values I've left them on auto. Any advice?

I'm using adaptive as shown below, but I've just finished reading lots of posts from 2 months ago about people having problems with adaptive. I've seen no voltages above my total adaptive mode cpu core voltage, even when using p95 etc. But I want to be sure.

Secondly. What about LLC, I've left all those settings stock and again everything is fine. So is it necessary to play with these settings for a mild 4.5 overclock?

CPU core/cache voltage = adaptive mode
Offset sign = +
Total adaptive mode CPU core voltage = 1.332
 
Last edited:
kandor said:
*edit three questions left:

I've left the CPU System Agent voltage and the CPU VCCIO voltage on auto, even though I have 4x16 (64gb) ram. I've read that I should hard code these but since I'm not sure on values I've left them on auto. Any advice?

I'm using adaptive as shown below, but I've just finished reading lots of posts from 2 months ago about people having problems with adaptive. I've seen no voltages above my total adaptive mode cpu core voltage, even when using p95 etc. But I want to be sure.

Secondly. What about LLC, I've left all those settings stock and again everything is fine. So is it necessary to play with these settings for a mild 4.5 overclock?

CPU core/cache voltage = adaptive mode
Offset sign = +
Total adaptive mode CPU core voltage = 1.332
Click to expand...

As long as it's stable, auto should be fine. These haven't been around long enough for any long term effects to be noted in 64GB systems. Especially as uncommon as they are. Many people are transposing their X99 experiences and understanding onto Skylake / Z170 which isn't necessarily valid. The DDR4 controller in Skylake is far better than the one in Haswell-E. The latter also has four channels as opposed to two. This makes stability more difficult to achieve. You have double the modules, slots, power phases, and trace paths. It's not surprising that Haswell-E (an older design at that) doesn't clock RAM as well as Skylake does. 1.332v is fine. You can take these chips to 1.42v and so far they can sustain that.

ASUS has some of the best automatic settings out there. These only improve with BIOS updates in most cases. I wouldn't worry about it. Adaptive voltage is fine so long as your temperatures are within ranges your cooling system can reasonably dissipate in year round conditions. Load-line calibration is something that we used to have to mess with back in the day but the last two or three motherboard / chipset generations have generally been fine with this set to automatic. There are a few configurations where you might not be able to get away with that. If and when you encounter instability you can tackle that problem then. Just know that this doesn't need to be adjusted too often anymore.

If your system is passing stress tests at 4.5GHz and the temperatures aren't hitting limits which would induce throttling then your golden.
 
Last edited:
Thanks for the info Dan,

Every thing has been perfectly stable so far at 4.5. No glitches (although occasionly I've had trouble with getting the keyboard recognized in the upper usb2 ports).

Max temp was upper 70's in p95 w/ avx2. Normal stress temps are in the 50's and 60's. I've tested cinebench, XTU, realbench, p95 and handbrake (although I don't know how to use handbrake).

Idle temps are funny, 16c which is cooler than ambient. Cooler is H100i GTX.

I appreciate all the info here, it's been a rush to finish and it's a brand new chipset to me.
 
I overclocked my 6700k to 4.5 and my G skill 32gb memory (4 x 8) is at 3200 (default XMP). I'm using an Asus Deluxe board. I had so many RMAs for various components in this build that when I got this all stable at 4 hours of Real Bench I never touched it again. Not worth my time to try to get more.
 
Realbench doesn't seem very stressful. I think I'm used to prime95. It's hard to let go of. But I keep reading that it's not a very good indicator for skylake.
 
Since you asked:
Cinebench 11.5 or 15
and Hyper-Pi 32m on all cores twice...on high priority.
Prime95 was known to overheat and kill DC chips;I wouldn't recommend it.
 
Cool, how long for hyper-pi? Is it an overnight thing?

*edit, ok I see - hyper-pi is pretty quick. Takes about 9.5 min to run. Sometimes a core finishes instantly though, does it report if there is an error, similar to p95?
 
Last edited:
Hi,

I have a question on this overclock. I used 'Manual' in the bios. BCLK to 100, then I set the dram voltage (1.35) and basic timings: 15,15,35 manually.

The bios lists the full XMP settings on the SPD page. But I didn't enter all of those settings manually, some of the codes don't match whats in the dram timing page in any case.

So, should I use XMP instead? Otherwise, it seems like I have a frankenstein situation with half Jdec and half manully entered XMP.

The XMP settings says it automatically sets the BCLK and that's why I kept it at manual.

Thanks!
 
kandor said:
Hi,

I have a question on this overclock. I used 'Manual' in the bios. BCLK to 100, then I set the dram voltage (1.35) and basic timings: 15,15,35 manually.

The bios lists the full XMP settings on the SPD page. But I didn't enter all of those settings manually, some of the codes don't match whats in the dram timing page in any case.

So, should I use XMP instead? Otherwise, it seems like I have a frankenstein situation with half Jdec and half manully entered XMP.

The XMP settings says it automatically sets the BCLK and that's why I kept it at manual.

Thanks!
Click to expand...

Are you trying to OC the memory or the cpu? If doing the cpu I'd leave the memory at relaxed timings to exclude it as a variable. Or you could try your current method and just stress test the memory for a bit to make sure it's 100% stable.
 
Hi, the ram is g.skill 2800 ddr4. I'm also running a 45x mulitplier on the cpu (6700k).

Leaving it on manual, and hard coding the ram voltage, and the 3 basic timings works fine. Also, (I've test for an hour on realbench), setting it to XMP also runs fine.

The funny thing is, the XMP profile parameters that are listed on the spd page list a lot of variables, like tRRD_S, _L, tRFC1, C2, C4 etc.

Yet, when you set the XMP profile and save the bios, when it lists what's been changed, it shows only tCL, tRCD and and tRAS.
 
kandor said:
Hi,

I have a question on this overclock. I used 'Manual' in the bios. BCLK to 100, then I set the dram voltage (1.35) and basic timings: 15,15,35 manually.

The bios lists the full XMP settings on the SPD page. But I didn't enter all of those settings manually, some of the codes don't match whats in the dram timing page in any case.

So, should I use XMP instead? Otherwise, it seems like I have a frankenstein situation with half Jdec and half manully entered XMP.

The XMP settings says it automatically sets the BCLK and that's why I kept it at manual.

Thanks!
Click to expand...

XMP timing detection is fucked on most Z170 and X99 motherboards. I think I've run into one motherboard this generation that hasn't detected the timings incorrectly. In some cases the timings are close enough that they'll still work at lower speeds. You can enable XMP and hopefully the voltages will detect correctly. After that you can set the timings manually and everything should work fine.

In the past motherboard manufacturers have detected memory timings and often set the actual timings 1 faster. The manufacturers will also sometimes run the base clock and thus memory clocks slightly over Intel's specifications. They do this in order to win benchmarks against other motherboards. This is the only reason I can see for such rampantly incorrect memory timing detection. You always want to look at your settings carefully and ensure that they are in fact what you expect them to be.
 
Thanks Dan,

I'll keep mine at manual then - but to be clear, I'm manually settings only 5 things:

Voltage = 1.35
Speed=2800
TCL=15
tRCD=15
tRAS=35
 
just subscribing, may need this in the future OC'd
picked the same CPU and cooler...waiting on memory purchase
good luck sir
 
Update: I had a couple random instances where the bios would come up 'overclocking failed' press F1 to enter setup. Even though nothing had actually failed. And simply entering and exiting the settings allowed it to boot. Not sure what causes this.
 
kandor said:
Update: I had a couple random instances where the bios would come up 'overclocking failed' press F1 to enter setup. Even though nothing had actually failed. And simply entering and exiting the settings allowed it to boot. Not sure what causes this.
Click to expand...

Are you using any Fast/Quick Boot features? You can try disabling those. Also does this happen on a reset or always on a power on from the system being off? If it is the later, try turning off the system, turn off the power supply, click the power button on the case, turn the power supply back on, then hit the power button again. Does it fail each time after doing that?

And the fact that this is now failing should validate your suspicions :p
 
Last edited:
I'll be putting together a Sabertooth and 6700K build after the holidays, so this thread is very interesting to me. I'll be quite satisfied if I can run the CPU at 4.5GHz and memory at 2800Mhz. I am hoping someone can comment on how much power that configuration consumes.
 
Ocellaris said:
Are you using any Fast/Quick Boot features? You can try disabling those. Also does this happen on a reset or always on a power on from the system being off? If it is the later, try turning off the system, turn off the power supply, click the power button on the case, turn the power supply back on, then hit the power button again. Does it fail each time after doing that?

And the fact that this is now failing should validate your suspicions :p
Click to expand...

I thought I had responded to the above, but must have never submitted. So...

It's only happened on a power on from system being off. I'll have to test your procedure when I can as these systems are in a different location now.

Thanks,
 
I got my gaming rig skylake system (6700k) over clocked to 4.5 with memory at 3200. It was easy and while I could aim for higher I haven't messed with it since the system is 100% stable
 
evilsofa said:
I'll be putting together a Sabertooth and 6700K build after the holidays, so this thread is very interesting to me. I'll be quite satisfied if I can run the CPU at 4.5GHz and memory at 2800Mhz. I am hoping someone can comment on how much power that configuration consumes.
Click to expand...

I get that people want to know about the power consumption, but the difference between a GIGABYTE Z170X Gaming 7 and an MSI Z170A SLI wouldn't be enough to create a noticeable difference in your electric bill. (At least I would hope not.) If we are talking about a 50 watt difference then sure, over the life of the system I could see that but I doubt it's that big a gap.
 
I built a 6700K few weeks ago and got 4.7ghz @ 1.37V and it passed everything I have thrown at it. It was an easy OC.. Could not get 4.8 stable @ 1.42V so I backed it back down. I didn't want him on the edge of stability. I took a 6600k to 4.5ghz and all I did was bump voltage to 1.32V. Anything lower on the voltage it would fail OCCT. Passed 8 hours of OCCT and IBT with no issues. Been playing on it a week with no problems.
 
You must log in or register to reply here.
Back
Top