PDA

View Full Version : Kaby Lake overclocking guide



Raja@ASUS
01-31-2017, 10:19 PM
http://edgeup.asus.com/2017/01/31/kaby-lake-overclocking-guide/

enjoy.

Arne Saknussemm
01-31-2017, 11:13 PM
Nice guide....as always!

Menthol
02-01-2017, 12:09 AM
Raja,
Very good guide

bn880
02-01-2017, 12:48 AM
Thanks, quite useful. :cool:

Raja@ASUS
02-01-2017, 06:33 AM
Its quite basic, but has some useful info. Glad you liked.

jab383
02-01-2017, 05:38 PM
Thanks for another gem, Raja.

bn880
02-03-2017, 09:36 PM
Its quite basic, but has some useful info. Glad you liked.

Yep. Actually I have a question, not sure if it belongs here: How would you go about overclocking in such a way as to allow the super low voltages at say 1GHz (like 0.7V) but yet get up to say 1.32 or 1.34V at 4.9GHz? (My CPU seems to be happy at 1.3V to 1.32V at 4.9 but I want to make full use of undervolting it at low freq.)

Is that even possible? I've tried just going "Auto" and it bumps me up to 1.36V and doesnt come down (though I set LLC to 1)

Raja@ASUS
02-04-2017, 08:13 AM
Yep. Actually I have a question, not sure if it belongs here: How would you go about overclocking in such a way as to allow the super low voltages at say 1GHz (like 0.7V) but yet get up to say 1.32 or 1.34V at 4.9GHz? (My CPU seems to be happy at 1.3V to 1.32V at 4.9 but I want to make full use of undervolting it at low freq.)

Is that even possible? I've tried just going "Auto" and it bumps me up to 1.36V and doesnt come down (though I set LLC to 1)

Do the following:

1) Use Adaptive Vcore.

2) Set the C-States like this:

62345


3) Make sure the Windows Power Plan is set to Balanced.

bn880
02-04-2017, 11:15 PM
Thanks a lot Raja, that seems to have done the trick quite well. (Hard to believe at times, 0.3V and 4.9G on a core) whaaaaat :)

Do the following:

1) Use Adaptive Vcore.

2) Set the C-States like this:

62345


3) Make sure the Windows Power Plan is set to Balanced.

robtherobot
02-18-2017, 05:01 PM
Excellent guide! Thank you! :)

MrAgapiGC
02-22-2017, 07:57 AM
I just receive my hero ix big question these could apply to 6th gen CPU? Or only for new one. The oc data looks the same without some data live avx*

Silent Scone@ASUS
02-22-2017, 08:48 AM
I just receive my hero ix big question these could apply to 6th gen CPU? Or only for new one. The oc data looks the same without some data live avx*


http://edgeup.asus.com/2015/08/05/all-things-asus-z170-skylake-platform-overviews-build-guides-overclocking-more/

https://drive.google.com/file/d/0Bz2VRRbLPrZnMXBnOXRWeVlHcHM/view

MrAgapiGC
02-24-2017, 07:26 PM
Thanks Raja amazing link.

Question. there are is a way to check if it is posible to adapt the manual to skylake on the z270. *We just put ours here to 4,7 at 1,392 volts it is amazing how clean an more stable Is ere with z170. if someone knows a good reading for z270 running 6700K please send it so I can check my data


Carlos

*

http://edgeup.asus.com/2017/01/31/kaby-lake-overclocking-guide/

enjoy.

MrAgapiGC
02-24-2017, 08:44 PM
7Thanks Silent Scone for the links. I was thinking that having a z270 board the OC where diferent. I sit downs with a friend of mine yesterday and we got 4.7 at 1.392 with few tweeks here and there. Very cstates off and performance on windows. it is stable. but maybe other ideas will be nice. Still learning. [


http://edgeup.asus.com/2015/08/05/all-things-asus-z170-skylake-platform-overviews-build-guides-overclocking-more/

https://drive.google.com/file/d/0Bz2VRRbLPrZnMXBnOXRWeVlHcHM/view

[ATTACH=CONFIG]62709

KuraiShidosha
02-26-2017, 06:47 PM
Does memory XMP affect CPU voltage demands? At base memory speeds my 4.8Ghz 1.250vcore is perfectly fine and stable. Can bench for hours without fault. As soon as I set my memory to XMP which is 3200Mhz, I get crashes in Realbench after only a couple minutes.

I've tried raising DRAM voltage, especially because for some weird reason the board isn't providing 1.35v like the memory asks for, it's only getting 1.344v instead. Raising that to 1.36 didn't help at all.

Then I tried raising VCCIO and SA to 1.2v up from ~1.15v. Also didn't help at all. Strangely power consumption dropped tremendously when I had the above voltages applied when compared to Auto and they were lower. Still, stability was much worse.

The only thing I can think of is raising CPU voltage. But is this normal? Does memory overclock really require a stable CPU voltage to be raised? I'm trying to keep voltage down as much as possible to avoid degradation. I'm pretty sure I've already seen some, or maybe it was just that initial burn in phase, because when I first got the chip 4.8Ghz was stable for Prime95 small FFT at 1.20v. After a few days I had to keep raising it to 1.25v to get it stable. This whole time XMP has been active but now I'm not stable at even 1.250v. That's when I decided to set the memory to standard and see what happens and yep the chip is perfectly fine at 1.250v again.

What should I do and/or be experiencing here?

Silent Scone@ASUS
02-26-2017, 08:26 PM
Does memory XMP affect CPU voltage demands? At base memory speeds my 4.8Ghz 1.250vcore is perfectly fine and stable. Can bench for hours without fault. As soon as I set my memory to XMP which is 3200Mhz, I get crashes in Realbench after only a couple minutes.

I've tried raising DRAM voltage, especially because for some weird reason the board isn't providing 1.35v like the memory asks for, it's only getting 1.344v instead. Raising that to 1.36 didn't help at all.

Then I tried raising VCCIO and SA to 1.2v up from ~1.15v. Also didn't help at all. Strangely power consumption dropped tremendously when I had the above voltages applied when compared to Auto and they were lower. Still, stability was much worse.

The only thing I can think of is raising CPU voltage. But is this normal? Does memory overclock really require a stable CPU voltage to be raised? I'm trying to keep voltage down as much as possible to avoid degradation. I'm pretty sure I've already seen some, or maybe it was just that initial burn in phase, because when I first got the chip 4.8Ghz was stable for Prime95 small FFT at 1.20v. After a few days I had to keep raising it to 1.25v to get it stable. This whole time XMP has been active but now I'm not stable at even 1.250v. That's when I decided to set the memory to standard and see what happens and yep the chip is perfectly fine at 1.250v again.

What should I do and/or be experiencing here?

What memory kit?

Auto ruling scales with frequency and is designed to work on a worst-case basis for poorer samples. So reducing this voltage manually would likely result in a reduction in power consumption (however you might be measuring it).

You can try Maximus Tweak Modes 1 / 2 from under the DRAM Timing page and test memory stability accordingly to see if these help. A slight bump in vcore may be required depending on how conditional the overclock stability is. If raising vcore helps pass Prime, then this is the solution.

Personally I would test with Realbench for 2 to 4 hours followed by HCI Memtest or Google Stress App for memory stability.

KuraiShidosha
02-26-2017, 08:59 PM
I bumped my vcore up to 1.275v. It's running stable now, but I never thought I'd need to push CPU voltage up because I increase DRAM frequency. That seems odd to me. I know on my last build, a 3770k and 1600Mhz DDR3, I never noticed any drop in stability when using XMP.

I just hope I don't discover instability further down the road and need more and more Vcore to maintain. Very frustrating.

Raja@ASUS
02-26-2017, 09:16 PM
Vcore on this platform supplies both the CPU cores and Uncore - so, memory overclocking can impact the required Core/Cache voltage for a given core/uncore frequency.

ycodryn
02-27-2017, 01:08 AM
Wish that my 7700k could go with 1.250 on variable turbo voltage. I need to use 1.285v (on aida64 I see 1.344v) with level 5 LLC to pass tests on 4.8 GHz :(

Raja@ASUS
02-27-2017, 01:38 AM
There is plenty of variance between samples. A CPU like that may benefit somewhat from de-lidding/re-lidding.

lufiron
02-27-2017, 02:04 AM
I got my delidded 7700k currently running 5ghz @ 1.35v with a 3200 c14 gskill kit. *Max temp Ive seen when benchmarking is 67c. *Using air cooling with a Noctua NH-D15S.

Delidded using the Rockit 88 and CLU.

MrAgapiGC
03-05-2017, 08:27 PM
guys! i know that there some differences in 6700k and 7700k. i am not on the mark but more on 24/7. i am using the hero ix and is very stable for 4.7 at 1.392. states all off prefect off etc. old school oc. a friend of mine help me with that. Question. can i use the guide to OC my 6700k in hero ix. I am using h110iV2 with link since on my bios i read pump speed on quiet and link i can change to performance. so far running 0701 bios ok. also i check that i plug my fans to the cpu fan and opt i read only 1 that is normal?
62934

johnathonm
03-10-2017, 04:59 PM
Does anyone have a cached copy of this guide? It won't load for me.

Arne Saknussemm
03-10-2017, 05:08 PM
The link opens fine for me on Chrome

http://edgeup.asus.com/2017/01/31/kaby-lake-overclocking-guide/?_ga=1.263696652.1894420463.1489165646

You using strange browser or strange settings?

Raja@ASUS
03-14-2017, 02:06 PM
I think the server was a bit slow for a few days last week. Seems to be back to normal now.

Nate152
04-05-2017, 03:07 AM
Hi Raja

I see there is a setting called the ASUS Thermal Control Tool you mention in your awesome guide that can extend overclocking by up to 300MHz. Now I know this doesn't mean an automatic 300MHz but if I could get another 100MHz or 200MHz it would be worth trying it. As it's a new feature with z270 I have no idea how to use it or what to set.

Would you have any suggestions ?

I.G.SYME@BTINTERNET.COM
04-05-2017, 06:49 AM
Guys, have briefly skimmed down this thread and I too have been down these paths in an attempt to get to 5ghz on a 7700K all to no avail. Yes, you can fiddle with ratios and core voltages and also play around with DIP5 to your hearts content BUT it was not possible to get a stable 5ghz setting in my case. Realbench would always crash at some point. So, I thought to myself "why not try the easy way and play round with the tuning wizard in the BIOS (F11 I think it is). Did that and voila! A stable 4.95ghz and also 3200 ram bumped up automatically to 3300 and all stable in Realbench testing. I also used the excellent guide on this site to overclocking the GPU and followed instructions to bump up the GPU to 1350mhz as well. All stable as hell and this bumped me up to Number 1 spot in Passmark ver 9.0 for the same spec of hardware. I am not a gamer and have no graphics card, I just like to overclock and have a fast machine. I employ a custom watercooled loop with a 480 rad and a 360 rad so cooling is not a problem. Just my tuppenceworth.

Raja@ASUS
04-08-2017, 09:15 PM
Hi Raja

I see there is a setting called the ASUS Thermal Control Tool you mention in your awesome guide that can extend overclocking by up to 300MHz. Now I know this doesn't mean an automatic 300MHz but if I could get another 100MHz or 200MHz it would be worth trying it. As it's a new feature with z270 I have no idea how to use it or what to set.

Would you have any suggestions ?


Info on how to use the feature is in the OC guide...

Matrixero
04-11-2017, 03:22 PM
Hello, is the LLC left in Auto? thanks

Raja@ASUS
04-13-2017, 05:12 AM
Hello, is the LLC left in Auto? thanks

You can leave it on Auto if you wish. If using a different value, you'll need to match the load voltage.

Matrixero
04-14-2017, 02:09 AM
You can leave it on Auto if you wish. If using a different value, you'll need to match the load voltage.

Ok thanks, I'll leave it in automatic.

Killer344
05-24-2017, 10:01 PM
Hey, quick question, I got DRAM stable at 4133mhz 1.405v with:

VCCSA: 1.288v |
---------------| Both during high load, idle/set voltages are a bit lower.
VCCIO: 1.264v |


From the guide I'd assume less than 1.3v for VCCSA/VCCIO should be considered fairly safe for 24/7. The only reference is this sentence: "If possible, do not venture too far from 1.35V as a maximum."

I also guess 1.4v should be pretty safe for the IMC as well since ram manufacturers are shipping them rated like that, correct?

Nate152
05-24-2017, 10:37 PM
Hi killer344

Welocme to the ROG forum !

Dram voltage at 1.405 is fine, SA and IO voltages are fine too, you can run it there 24/7.

Nice going getting your high speed ram stable. :)

Killer344
05-24-2017, 11:27 PM
yeah, getting the right VCCSA/VCCIO values was a total *****, any slight increase/decrease from those values leads to instability.

I got a SL binned 7700k 5.1ghz at 1.37v (1.39v during avx loads) running as well, I can go toe to toe against 6850k/1600x at these speeds lol...

I can't thank enough the Apex's recovery system from failed settings, I didn't need to clear the CMOS even once.


The only odd thing worth mentioning is I needed to rise the CPU Core PLL to 1.075v to reach CPU stability, and everyone says everywhere vcore is all that matters.... mine must be the exception :rolleyes:

dirtyred
06-01-2017, 12:20 PM
I'm having issues with Adaptive voltage mode. I apply 50 multiplier and 1.32 Vcore with LLC 5 the system seems to be stable (didn't stress test it for long time). But if I use an AVX offset then it instantly crashes. It seems that the BIOS is not guessing the correct voltage needed. This problem seems to exist with lower multiplier as well and the only solution I've found is to raise the Vcore but that results in unnecessary high voltage under non-AVX load.

I've read that VRM switching frequency can help with stability when the voltage requirements suddenly change. My motherboard supports up to 500 Hz but the BIOS is warning me that active cooling is recommended. The default value is 250 Hz and I bumped it up to 400 Hz. That seems to help with less vdroop but I'm afraid to go higher. How safe is to go up to the maximum 500 Hz? I'm using Phanteks PH-TC14PE air cooling with good case airflow, is that enough to cool the VRM's?

Is there any benefit and risk setting the CPU Power Duty Control from T.Probe to Extreme and the CPU Power Phase Control from Auto to Extreme or Power Phase Response (Fast or Ultra Fast)?

Also CPU PLL Voltage according to Intel's specs is 1 V but according to BIOS the default is 0.9 V. Why is there such a difference?

PS.: My CPU is not delidded and it seems that the maximum Vcore is around 1.32 - 1.34 V to keep the CPU from overheating. But even at 1.32 Vcore my CPU reaches 85 C running RealBench so anything that can help to reduce the Vcore / CPU temps while maintaining stability would be appreciated.

Raja@ASUS
06-22-2017, 11:22 AM
I'm having issues with Adaptive voltage mode. I apply 50 multiplier and 1.32 Vcore with LLC 5 the system seems to be stable (didn't stress test it for long time). But if I use an AVX offset then it instantly crashes. It seems that the BIOS is not guessing the correct voltage needed. This problem seems to exist with lower multiplier as well and the only solution I've found is to raise the Vcore but that results in unnecessary high voltage under non-AVX load.

I've read that VRM switching frequency can help with stability when the voltage requirements suddenly change. My motherboard supports up to 500 Hz but the BIOS is warning me that active cooling is recommended. The default value is 250 Hz and I bumped it up to 400 Hz. That seems to help with less vdroop but I'm afraid to go higher. How safe is to go up to the maximum 500 Hz? I'm using Phanteks PH-TC14PE air cooling with good case airflow, is that enough to cool the VRM's?

Is there any benefit and risk setting the CPU Power Duty Control from T.Probe to Extreme and the CPU Power Phase Control from Auto to Extreme or Power Phase Response (Fast or Ultra Fast)?

Also CPU PLL Voltage according to Intel's specs is 1 V but according to BIOS the default is 0.9 V. Why is there such a difference?

PS.: My CPU is not delidded and it seems that the maximum Vcore is around 1.32 - 1.34 V to keep the CPU from overheating. But even at 1.32 Vcore my CPU reaches 85 C running RealBench so anything that can help to reduce the Vcore / CPU temps while maintaining stability would be appreciated.

AVX stability can require significant Vcore. If the AVX offset feature is resulting in the wrong level of voltage for the target ratio, your other option is to configure the temperature control features and apply separate voltages for the light and heavy-load overclocks. Details on doing this are in the OC guide.

ycodryn
09-06-2017, 11:15 AM
After staying with 4.8ghz I tried 5.0 just for fun. All freq tested with realbench 2.5.4, 1h. Here is some info:

3.5ghz > 45,46,45,44c / 0.992v / 34.569w > cache 4.5ghz (this is limited in windows 10 to stay at 3.5ghz for common tasks only, same settings in bios like 4.8)
4.8ghz > 65,65,63,62c / 1.280v / 82.106w > cache 4.5ghz (1.220v in bios, llc4)
5.0ghz > 73,71,69,68c / 1.360v / 92.886w > cache 4.0ghz (1.305v in bios, llc5)
5.0ghz > 74,72,71,69c / 1.360v / 94.410w > cache 4.3ghz (1.305v in bios, llc5)
5.0ghz > 74,72,72,71c / 1.360v / 94.979w > cache 4.5ghz (1.305v in bios, llc5)

What do you think? CPU is delided and has H115i h2o cooler, room temp 24-25c.

Chino
09-06-2017, 01:48 PM
5.0ghz > 74,72,72,71c / 1.360v / 94.979w > cache 4.5ghz (1.305v in bios, llc5)

ycodryn
09-06-2017, 05:11 PM
Are ok those temps for a delided cpu? Or should I try again that operation?

Chino
09-06-2017, 05:34 PM
Very good temperatures in my book.

jjxakergm
09-12-2017, 10:50 PM
How to bypass a capricious CPU IMC (7600k) using a motherboard Maximus IX Apex (bios 1010)? Increasing the voltage on DIMM and SA/IO does not offer, I myself know it.
We tuned overclocking only the memory in this way (complete stability):
Ai Overclock Tuner [Manual]
BCLK Frequency [100.0000]
DRAM Frequency [DDR4-4133MHz]
DRAM Voltage [1.3500]
CPU VCCIO Voltage [1.2500]
CPU System Agent Voltage [1.3000]
DMI Voltage [1.35000]
Maximus Tweak [Mode 1]
DRAM CAS# Latency [18]
DRAM RAS# to CAS# Delay [18]
DRAM RAS# ACT Time [38]
DRAM Command Rate [2N]
MRC Fast Boot [Enabled]
MCH Full Check [Disabled]
DRAM Current Capability [130%]
DRAM Power Phase Control [Extreme]

But if I choose a multiplier for the frequency 4266Mhz then I get an Q-Code 26. experiments with voltage and main timings do not give anything...
Then we try to use memory profiles, select 2x8GB New Samsung Single-Side/"Load Raja's 2x8GB Samsung B-die 1.40V 4133" we get an Q-Code 49 (and even reducing the frequency to 3866Mhz will not help, there all available profiles 2x8GB New Samsung Single-Side do not work). but as we can see earlier, there were no problems with this frequency. it turns out that the profile is not working for our CPU instance...
The question is how to circumvent these whims of the bios?
Why the bios automatics can not cope with the calculation of extended timings for system start?
Is it possible that problem Q-Code 26 is associated with an option DRAM CLK Period? very capricious thing, and use uncomfortable, even unclear what value is assigned in automatic mode... for each memory frequency has its own meaning.
The problem is not only with the frequency, but also with the use of low timings, for example 3600Mhz CL14-14-14-34-2 on APEX there is no starts, with on the board of another brand all ok. so it turns out that, why change the processor if on other board it works good with memory.

Naturally, the use of another CPU instance solves all the issues, but the current instance has problems only with Asus boards. so I would like to hear a professional opinion about current problems.