Newbie finding best settings for 8086k, Noctua NH-U14s and Prime Z390A

[granolamonster]

Also, any advice on fans for my case. At the moment I've got one low front input fan and one high rear exhaust. The case has plenty of space for additional fans at the front, bottom and side. Is it worth me adding a third fan and if so where and in which direction?

- - - Auto-Merged Double Post - - -

The voltage, yes.

To add to this, now that you know X.xx volts is stable, you can then use an offset if you choose. For example. If 1.15V is stable at your clockspeed. Reboot with the voltage on auto and see what the LOAD voltage is. Then use an offset to reach your known voltage value. If your CPU needs 1.25V on auto, use a -100mv offset to reach 1.15V under load.

Sorry, still trying to get my head around this.....so.....I find manually that 1.15v is stable at 46x. I then set the voltage to auto and run a stress test and find that with auto vCore settings (and presumably SVID/Auto enabled) the cpu wants 1.3v so I would use a -150mv offset to get back to 1.15v when it is put under load in the future?

 

[EarthDog]

Also, any advice on fans for my case. At the moment I've got one low front input fan and one high rear exhaust. The case has plenty of space for additional fans at the front, bottom and side. Is it worth me adding a third fan and if so where and in which direction?

- - - Auto-Merged Double Post - - -



Sorry, still trying to get my head around this.....so.....I find manually that 1.15v is stable at 46x. I then set the voltage to auto and run a stress test and find that with auto vCore settings (and presumably SVID/Auto enabled) the cpu wants 1.3v so I would use a -150mv offset to get back to 1.15v when it is put under load in the future?

Front/sides = intake, top/rear = exhaust in most cases. I'd add one more intake and one more exhaust.

You are correct with your math.

 

[granolamonster]

I'm still not quite getting this offset voltage method right.

So far I've found that I can run 46x at 1.1v quite happily.

I then go back and set the vCore to Auto and enable SVID. I run AIDA and watch the vCore vary wildly. I get a max of about 1.23v and a min of 1.194. I decide to work with the min and put in an offset of -.095 to bring me down to 1.1 and it won't boot. What am I doing wrong? How do I find the load voltage? And, do I need to have SVID enabled whilst I find this?

Thanks once again

 

[EarthDog]

Leave everything on auto... do not enable SVID. Go and put a stress test load on it... look at the voltage. That should be your starting voltage. Then use the offset.

 

[granolamonster]

Ok, and should I be using the min temp as my base?

And.....once I set the offset am I leaving SVID off?

 

[trents]

Also, any advice on fans for my case. At the moment I've got one low front input fan and one high rear exhaust. The case has plenty of space for additional fans at the front, bottom and side. Is it worth me adding a third fan and if so where and in which direction?

- - - Auto-Merged Double Post - - -



Sorry, still trying to get my head around this.....so.....I find manually that 1.15v is stable at 46x. I then set the voltage to auto and run a stress test and find that with auto vCore settings (and presumably SVID/Auto enabled) the cpu wants 1.3v so I would use a -150mv offset to get back to 1.15v when it is put under load in the future?

Regardless of how you do it, with offset or LLC, manual or whatever, you need to find the lowest voltage under full load that will keep you stable. You need to observer the max voltage in HWMonitor or something while stress testing the CPU in order to see what that is.

 

[granolamonster]

Hi all,

I thought I'd drop you an update on progress. I've been and added two more fans to the case, one more at the front and one more exhaust at the rear top.

I think I've found a settings combo that works, but please feel free to critique:

MCE = Disabled
Core = Sync All
Core Ratio = Auto
SVID = Best Case
SVID = Enabled
IA/DC = 0.01 (although no idea what this does, just followed tips online and seems to work)
LLC = 6
Adaptive vCore = 1.294v
Offset = -0.065v

This lots seems to give me a machine which will quite happily hit x50 without drawing more than 1.3v which is a lot less than the system would draw when on auto settings. Plus it can spin down CPU/vCore when idle which seems like a sensible thing.

After 10 minutes (I know a longer stability test is needed) of prime95 I end up with the following stats:

Max Core: x50
Avg Core: x45/x46
Max VID: 1.299v
Avg VID: 1.27v
Min VID: 0.645v
Min Core Temp: 30ºC
Max Core Temp: 70ºC
Avg Core Temp: 60ºC

What do you think? Are there any other settings I should tweak? I think that this way I finally manage to get to this CPUs full turbo without the mad voltages I was seeing with auto settings (1.41v in some cases to achieve the same thing).

Cheers!

 

[trents]

That looks about right to me but it's very, very preliminary if you have only stress tested for 10 minutes. You need to stress test for about 2 hours with Prime95. You will find that at about 18-20 minutes into a Prime95 stress test the temps climb significantly. Instead of Prime95 you might consider using AIDA64 Extreme (test overnight) or the Realbench stress test for about four hours. Both of those will give you a good real world stress test with lower temps than Prime95.

Also, what did you do with the AVX offset in bios?

 

[granolamonster]

Thanks - I've not yet understood the differences between Prime95 and AIDA. I'll give the latter a go, only gave up using it because don't want to pay for a copy. Currently got AVX set to 1.

 

[EarthDog]

P95 is a bit more stressful than using AIDA64 at default. If you just stress test with FPU as the only thing checked, they are pretty close.

I personally use AIDA64 for long term testing, P95 for short term as it can find instability quicker or your temp limit.

 

[trents]

AIDA64 is free to use for 30 days on each new computer you install it on.

 

[granolamonster]

Cheers for the pointers. Ok, I've managed to get a bit more testing in.

So I end up with basically the same setup as before, but I've found that actually the system is stable with an adaptive voltage of 1.25v and offset of -0.06v.

AIDA64 testing for 1 hour with GPU option enabled as well:

Max Core 50x, Avg 47x
Max Core Temp 85ºC, Avg 70ºC
Max vCore 1.275v, Avg 1.201v

I've then run 2 hours of RealBench with 8gb of ram selected because it wouldn't kick off with 16gb selected:

Max Core 50x, Avg 46x
Max Core Temp 80ºC, Avg 70ºC
Max vCore 1.275v, Avg 1.174v

How does that sound?

 

[EarthDog]

Why is there an average core speed?

Is it throttling for other reasons (VRM or power limit throttling)? Or is that just power savings and an erroneous data point?

I guess what I am getting at here is an average isn't really relevant. So long as while you are stress testing that the cores stay where you say they should be speed wise, then you are ok. If it is jumping around with a static load like that, something isn't right.

 

[Johan45]

I think he's still on auto for the clocks E_D, just dialing in voltage requirements

 

[granolamonster]

Yep, I'm just using the auto function for the clocks. I can set the clock to sync all cores at 50x but get the same effect, once a load is applied across all of the cores then the cpu seems to throttle down to 46x/47x. I'm not sure why. It doesn't jump a lot, it seems to start off at x49 which figures with an AVX offset of 1, but after a few minutes drops down to x46/x47 and stay there. Is there something I can do to explore this?

 

[EarthDog]

Because you are still on auto, the clocks will fluctuate. This should be normal. You can check for throttling reasons by keeping Intel XTU up and running your test. If none of the reasons pop for throttling, then the board is acting normal (as I expect in the first place) with the settings on auto.



...and I guess now he's back to clocks... lol. So confused.

 

[granolamonster]

Alright

I haven't used the intel XTU up until now, so good shout, if I run the 5 minute stress test then it flags up power limit throttling a few times whilst running it, although it claims that it passes the test at the same time which makes less sense. Is the power limit throttling just down to the use of a reduced vCore or is this related to something else I have yet to learn (it's ok, I like learning)

 

[trents]

Power limit throttling actually can prevent you from failing the XTU test. Failing the test means that data is being lost during the calculation work that makes up the stress test. It's expecting a certain mathematical sum from the calculations and getting something different. XTU will notify you if that happens and terminate the test. If the machine is grossly unstable you will actually get a lockup or blue screen and spontaneous restart.

 

[granolamonster]

So.....if I understand this correctly the real issue is the CPU hitting the boards TDP limit. So what happens if I increase the short term limit to say 200w and the long term limit to 150w? Is that a dangerous move or is that acceptable? I've been trying to find a bit more out on the forums but advice seems pretty varied.

 

[EarthDog]

Crank them both up as high as they go.