OC: Sabertooth 990fx R2 + FX 8350 + Noctua D15. Low Vcore High Temps

[second]

Hello,

I am opening this thread to log my attempt to OC my RIG, and check for solutions.
I am running now at 4.65 GHz at 1.344V. I am checking stability and temps with OCCT: OCCT. The reason I use OCCT is because it can stop the test based on temps. For now I am using just the FSB to push. Keeping the RAM, NB and HT at or below 1866, 2400 , 2400.
The problem is that in a 4 hour check the package temp reached 61º (CPU 65ºC). Ambient temp should be between 18º and 22º C. I was expecting much more cooling from D15. RGone reached 4.6 GHZ at 1.45V with a low cooler and the package temp was 59º.
Later I will post all the settings. Questions:
1) Can the D15 be defective, and thus I am getting high tempos? I could have mounted something wrong? Pea method Noctua thermal compound.
2) 62º is the absolute max temp for 8350, or it is ok to spike at or little above? At some point I must set the rig for 24/7 usage.


In case someone stops here looking for OC information I highly recommend: http://www.overclockers.com/forums/...s-gt-gt-Using-the-AMD-FX-Bulldozer-Piledriver

 

[caddi daddi]

those runs by Rgone are by a true expert, are you a true expert?
4.65 is pretty darn good as all cpus vary one to the next.
what is your cpu/nb voltage, cpu/nb is on the cpu die and adds heat to the cpu.

 

[second]

I am not on my rig right now. At night I will post all settings.
If not mistaken CPUNB is at 1.225V.

English is not my default language, please forgive If something looked offensive, it was not my intention. I just mentioned Rgone to compare the cooling results.

 

[second]

This is the current setting for 4.65

 

[caddi daddi]

looks like you can back the cpu/nb voltage a bit.
I'm thinking vdda should be 2.50 and vddr should be 1.25, let me find my notes.
in my notes that is what I find, I also noted to myself that auto worked fine until the clocks got extreme.

 

[second]

I leave the voltages from previous setting, forgot that I have lowered the CPU NB multiplier.
So CPU NB back to "stock", actually it is at 1.156 as if I leave the offset on auto the bios set it to 1.4V.
Put VDDA on auto. I upped because I read on some threads that it can help lower Vcore.
VDDR as far a I know is the Dram singnaling, and BIOS stated 1.2 standard, so I upped it to 1.205, just in case.

1:30 hour of OCCT still no crash or error. Package hit: 56 and CPU hit: 56. But package is staying at 53, and before it was staying at 57. Thanks.
If it pass 4 hours, I will try to low the Vcore.

Is it ok to let the CPU power phase control on Extreme? When I hit 4.5 if I let my system idle, when I hit the start on OCCT it will hang or show error. But if started the test right when system starts it will pass. So messing with the settings I discover that setting it to Extreme prevent this to happen.

 

[caddi daddi]

turn the power phase control down, this is how long the pulses of power stay "on" and extreme makes a lot of heat, so start backing that down also.
this will reduce the vrm section temp, the one you read as cpu.

 

[Waza]

IF your OC is stable at those voltages and temps stay in check, I would call that a good OC.
I am curious about the LLC setting you use. Your CPU voltage likely goes up in load to around 1.4V or close to it.
I would recommend sticking to LLC high leaving the CPU/NB voltage like it is. 4.6-4.7GHz is a good target, after that you will start to need additional cooling. Cold climate, AC on constantly or watercooling.
I would also advise to use ASUS optimized for power phase control.. Unless you start to push over 5GHz you will be fine without extreme settings.

 

[second]

So I backed the Power phase control to "Optimized", for now I get no error.
I was able to high up a notch the clock it is now at 4.7 Ghz but today is colder and I only run the test for 1 hour.
Waza,
My Vcore is getting 1.37 according to OCCT, in theory it uses HwMonitor to get the readings. Next run I will open HwMonitor too.
The problem is that my OC is not close to be useful for 24/7. I am right now pushing it to the limit to see where I can get. I was hoping to reach 5Ghz and it to be useful 24/7 at 4.6. I saw people reach 4.8 and 5.0 with 1.44 1.5 V I believed that I have a best on my hands, since as you can see I get 4.6 at 1.36 1.37.
About the LLC, at 4.4 I tested the options. In high setting it Vdrop a little. And at Ultra high it didn't, it overvolt a little. And the delta between min and max is lower at Ultra high. So I choose Ultra high, especially because it makes more sense to have the voltage go up on high usage.
It is autumn here, I choose this time to OC because it has the least temperature variance, making the results of cooling more comparable. It should be between 18 and 22 C. But in summer time it will easily pass 30 C. And I have a Crossfire of r9 290. So I have to get a lot better temp for a 24/7 usage.

I have to remove the socket fan for now, and just tomorrow I will be able to put it back, here is the result:

 

[caddi daddi]

what will you be doing with the rig?
I run my 8350 at 5.2 ghz in my file storage rig but this use never loads the cpu so it's no issue but I could not render video at that.

 

[cdawall]

You know a 9590 only runs 4.5ghz under those conditions correct? You are exceeding 220w TDP on an air cooler, at a vid that's an entire tenth higher than a 9590. What you are seeing is normal temps for that vcore and clock speed nothing is defective nothing is wrong. If you really want to wrap your head on how much variance is in the silcone my 12 core opterons use small 92mm noctua coolers and run roughly those temps at 4.6ghz and much lower voltage (1.42-1.45v)

 

[second]

Mostly gaming (GTA V, Battlefield, Call of duty and some others) and Internet.
Probably the CPU will not be at 100%, but the 290 will increase greatly the case temp.

cdawall,
I believe I don't understand correctly. As far as I know 9590 vid is [email protected] and [email protected]. And by you experience I should be getting better results. As you have 12 cores [email protected] with a 92 mm cooler and I have 8 core [email protected] with a 165mm cooler?
Or it is 2x6cores? Although I have no Idea what technology empowers the opterons.

Anyway for the price and for what I could found when I bought the D15 I was expecting better cooling.

 

[cdawall]

It's 24 cores total in my opteron setup. They have 2 6 core dies on each mcm. Your results are semi normal I was merely pointing out 8350 desktop silicone is bottom of the barrel stuff. So expect more hest more voltage and worse temps.

 

[second]

Yeah, I believe you are right.
After some testing and looking at results I came to the conclusion that the low voltage is actually a bad sign.
I believe that the CPU is drawing more current and more power than the others OCs I have seen. This explains the high socket and VRM temperature. And, I believe, explain why sometimes when I go from idle to load with Cpu power phase control on another option than Extreme the pc freezes or the tests show error.
I believe that the cpu start drawing a lot of current before the VRM activates the needed phases, causing the protections to be activated or the cpu to not get what it needs. Since If I restart the test or go from some load to full load the tests can go for 6h.
Now I am trying to improve the socket and VRM cooling. Based on ASUS temperatures VRM1 looks like to be sensed on the back of the board and the VRM2 on the front of the board. Cause when I trun off the back fan the VRM1 goes a lot high and VRM2 changes little.
I am using the FAN that came with the stock cooler, anyone know how to disable the embed thermal control?
My case have little space on the back, so I cannot use a standard 120mm fan. I have been thinking about changing the case, anyone know about Corsair carbide 540 with 2 290 on air? Since it does not have a side fan I believe that the heat from the GPUs will go straight to the intake of the Noctua.
Thank a lot for the help, soon it will get colder and I will restart the OC.