PDA

View Full Version : Overclocking Tips & System stability.



varkey
11-14-2017, 01:17 PM
First time building a system with unlocked CPU. Have few questions for overclocking.

System : 8600K, MaximusXHero, KrakenX62, TridentZRGB_3200_CL14. Running at high performance in windows power setting for testing.
Stability Tests : AIDA64(CPU,FPU,Cache,iGPU)-2Hr, HandBrake(x265)-30mins, RealBench-30mins
Temperature Readings after AIDA64(CPU,FPU,Cache,iGPU)-1Hr

Common Settings: MCE disabled, AC/DC LL 0.01

Stable OverClocks
@ 4.8Ghz -Adaptive 1.27V, LLC Auto, MCE disabled, AVX offset 0, RAM@2133; Readings: Vcore Idle-1.248 Load-1.180, Package Avg Temp 64
@ 4.9Ghz -Adaptive 1.33V, LLC Auto, MCE disabled, AVX offset 0, RAM@2133; Readings: Vcore Idle-1.312 Load-1.248, Package Avg Temp 70
@ 5.0Ghz -Adaptive 1.36V, LLC 4 , MCE disabled, AVX offset 0, RAM@2133; Readings: Vcore Idle-1.344 Load-1.328, Package Avg Temp 77

Question 1: How much Vdroop is healthy for 8600K's long life?
For 4.9Ghz I could get 0.064 vdroop. For 5 GHz I limited vdroop to 0.016 in order to avoid idle vCore > 1.35.
Question 2: What is the max VCore I should use for 8600K?
Vcore is max at Idle. How about load change voltage overshoots for 5GHz settings, am I crossing 1.4V?
Question 3: Any other settings I can try for 5GHz?

Tried XMP with below settings
@ 5.0Ghz -Adaptive 1.38V, LLC 4 , MCE disabled, AVX offset 0, XMP@3200;
Readings: Vcore Idle-1.360 Load-1.328, Package Avg Temp 78, System Agent 1.32V, VCCIO 1.14V
Question 4: But could not get stability. What voltages I should try to SA and VCCIO?

Prime 95 will fail in 15 mins for all the above settings in FFT 1344K.
It was only stable without XMP at higher voltage.
Question 5: Should I consider P95 for stability test?

Raja@ASUS
11-14-2017, 01:47 PM
A few things:

1) Droop isn't harmful in any way. It's there to ensure the magnitude and duration of overshoot stays within specifications. With less vdroop, you will have more overshoot. How much depends on the capabilities of the circuit. The level of overshoot will also vary according the LLC level used. Overshoot is more harmful because it occurs when the processor is drawing significant current. If the applied voltage is 1.36V, I would expect some short-term transitions over 1.40V with higher levels of LLC (it may not with your CPU, though, as this is all current dependant).

2) Use a monitoring tool to check processor power consumption. Typically, I find sticking under 2X the stock power consumption is wise. The higher the power consumption, the more chance there is of degradation. Remember that voltage is just part of the equation and that you also need to factor in the current.

3) The Vcore rail supplies both the CPU cores and Uncore (AKA Cache in the ASUS UEFI). If the Cache frequency is set high, it will impact how much Vcore the CPU needs for a given core frequency.

4) Outside the SA and IO rails, given what I've cited in number 3, you may need more Vcore for a given memory and cache frequency. I'd also confirm you can get the XMP settings stable at 4.9GHz CPU core frequency instead of trying only at 5GHz. That will give you more of a hint as to how much margin you have and where the instability lies. For DDR4-3200, a good CPU should need no more than 1.25V for SA, and say, 1.20V for IO. Too high a voltage can be just as unstable as not enough. Bear that in mind.

5) Some people swear by Prime, and some don't because it subjects the CPU to massive amounts of current. Depends what you're going to use the system for. There is no one single test that catches every type of system instability (even if you run the hottest test). Sometimes it takes a complex data pattern to trip up a system and sometimes it doesn't. Personally, I use a combination of tests (Realbench, some Prime, and HCI Memtest).

-Raja

varkey
11-15-2017, 12:13 PM
Thanks Raja for taking time to respond to my questions.
I am working on your recommendations and will post the results shortly.

varkey
11-17-2017, 12:46 PM
Got the system stable with below settings for AIDA64 -45m, HandBrake(x265) -30m, realbench -30m
@ 5.0Ghz -Adaptive 1.4V, LLC 3 , MCE disabled, AVX offset 0, XMP@3200; Cache x45, SystemAgent Auto, VCCIO 1.10V; Readings: Vcore Idle-1.376 Load-1.328, SAgent 1.14, VCCIO 1.112V, Package Temp Avg 77 Peak 79, Package Power 126W

For Prime95 to run for 15mins in FFT 1334K, had to set AVX offset 2.

Package Power reading in AIDA64/HWinfo64 in stock configuration @4.2GHz without MCE at Load is 85W.
Q1)So Peak package power under 150W is good?

I will be using the system primarily for photo editing, some video editing and gaming.
Q2) How long should I run Prime 95 and what kind of setting I should use in torture test for stability testing.
Will be trying LinX and MemTest later.

I was planning to increase cache to x47, but while testing in windows power saving mode got the Video_TDR_failure error with BSOD. This error is happening only while windows power management is set to either power saver or balanced profiles. Performance mode works fine.
I dont have any external graphics card installed, currently using itegrated GPU of 8600K. I tried setting the intel graphics card power saving setting to performace mode for all the power profiles. But still getting the BSOD. I will have to see if down clocking helps as I only tested in performance power pofile all along.

Raja@ASUS
11-17-2017, 01:38 PM
1) The small FFTs will pull a lot more power than 1334K. You'll be surprised how much more.

2) I usually run a stress test for at least 4-8 hours.

3) If using onboard graphics, you sometimes need to tune the iGPU voltage.

varkey
11-19-2017, 03:26 AM
Tried the iGPU voltages Offset 0.02, 0.04, 0.06, 0.08 and Manual 1.00V, 1.25V.
Still not able to stop the BSOD in power saving mode.
What other voltages should I try? Do I need to tune iGPU LLC too?

Raja@ASUS
11-19-2017, 05:12 AM
Can't say for sure. You may need to reduce the overclock and see where the issue starts to manifest. I'm not a person that uses the iGPU. I always run my systems with graphics cards.

flacer
11-20-2017, 03:42 PM
Vocre @ 1,4 is too high for 6/8 hours stress test of x265 handbrake ...

Try to lower, for a 11h 29m test with 88% CPU Thread (4.800 Ghz) I use 1,240+0,020 with medium power of 108W of Package ... medium Package temp 65

varkey
11-23-2017, 05:55 AM
1) GPU crash was due to change in Core/Cache ration of x45, lowered to auto @x40 and the system is stable for power saving modes. Any voltages to play around to get a better Core/Cache ratio?

2) Prime95 Small FFT was drawing 170W for CPU @ 1.296V and 4.8GHz. Just imagining how much voltage and power it needs for 5GHz. Its crazy, would not try again.

varkey
11-24-2017, 05:28 AM
Vocre @ 1,4 is too high for 6/8 hours stress test of x265 handbrake ...

Try to lower, for a 11h 29m test with 88% CPU Thread (4.800 Ghz) I use 1,240+0,020 with medium power of 108W of Package ... medium Package temp 65

VCore is now set at 1.32V for Adaptive mode and AVX offset 1. The system draws max 1.296V for LLC3 when idle @ 5GHz. When under AVX load @4.9GHz for AIDA64/HandBrake VCore is only 1.234V and power draw at package level is 105W. Did a stress test for 3 Hrs with AIDA64, HandBrake & RealBench.
I am planning to keep this setting for daily use as CPU Temp stays just below 70C for the above stress tests. This way I will have a head room of 10C for summer days.