Sabertooth 990FX R2.0 & FX 8320 - lowering voltage

[0v3rl0rd]

First things first:

MB: Sabertooth 990FX R2.0
CPU: FX-8320
CPU cooler: Cooler Master 212 EVO + PHANTEKS PH-F140HP fan on the cooler

Ambient temperature: 22°C

Everything is stock in BIOS. With stock setings, when booted in Windows, I get the following core voltage readings (CPUZ, HWMonitor):

• min: 0,852
  
• max: 1,272
  
• other states between MIN and MAX:

• 0,864
  
• 0,900
  
• 0,936
  
• 0,948
  
• 0,996
  
• 1,044
  
• 1,164
  
• 1,224
  
• 1,248
  
• 1,260

With those stock voltages I get IDLE temp of 33-34°C. I use the TurboV EVO utility to lower the voltage. The stock CPU voltage displayed in TurboV EVO is 1,2875, and I lower that to 1,18125 (stable with Prime95 Small FFTs & Blend, stable with Aida64 stability).
With this voltage (1,18125) I get the IDLE temp arround 31-32°C, and following voltage readings (CPUZ, HWMonitor):

• min: 0,744
  
• max: 1,164
  
• other states between MIN and MAX:

• 0,756
  
• 0,840
  
• 0,852
  
• 0,912
  
• 0,924
  
• 0,936
  
• 1,020
  
• 1,056
  
• 1,140
  
• 1,152

I like that it drops the frequency (and the voltage also) when IDLE-ing, so it stays cooler. I'm happy with those settings. The thing I'm not happy with is that I have to use TurboV EVO to achieve it.

I'd like to set this situation in BIOS but I have no idea what to change to what values. I want to keep this frequency and voltage "scaling" to keep it as cooler as possible in IDLE.

Tnx

[elesde]

Nice sample you got there if it's really stable at those voltages 😄

What you can try in Bios is setting the CPU voltage to Offset -0.1V, should produce almost the same values you set via TurboV.
Best to check every P-state separately for stability with something like AMD PSCheck + Prime/y-cruncher.

[0v3rl0rd]

elesde wrote:
Nice sample you got there if it's really stable at those voltages 😄

What you can try in Bios is setting the CPU voltage to Offset -0.1V, should produce almost the same values you set via TurboV.
Best to check every P-state separately for stability with something like AMD PSCheck + Prime/y-cruncher.

Thanks for Your suggestion. I did try as You said. The voltage readings (with minus offest of -0,1) were these:

• 1,224
  
• 1,212
  
• 1,200
  
• 1,188
  
• 1,104
  
• 1,092
  
• 1,080
  
• 1,068
  
• 1,056
  
• 1,044
  
• 1,032
  
• 1,008
  
• 0,972
  
• 0,960

What I noticed is that somehow the Windows Power Profile got changed from "balanced" to "high performance" (have no idea how, maybe through an update or something, because I didn't change that one manually), so I changed that back to "balanced".
Of course I did think of all the possible causes (in BIOS) of these "wierd" voltages that didn't make sense to me - when I tried everything, than I remembered there is one more thing to check - the simplest one -.-'

Than, I lowered a voltage even further, and with the offset of -0,14375 I got these readings:

• 1,176
  
• 1,164
  
• 1,152
  
• 1,056
  
• 0,948
  
• 0,924
  
• 0,732
  
• 0,732
  
• 0,720

Right now (15 minutes and going...) I'm running Prime95 small FFTs:
Ambient: 23°C
Socket: 52°C
Cores: 38°C
Vcore: 1,176 V

Later will do Prime95 blend, and aida stability. I hope it remains stable 🙂

[elesde]

Looks nice 😄

Just make sure it is stable in the turbo P-states as well, usually the CPU clocks down under heavy load to stay within the TDP limit when APM is activated. You can try to run Prime with just 1/2 or 4 cores loaded.

Might gonna get my a second sample of these CPUs as well, those voltages look really great... have you tried overclocking it at all?
Also wondering what the LLC (Load Line Calibration) setting for the CPU is in your bios. With LLC deactivated voltages should drop quite a bit under load.

[0v3rl0rd]

I'm pretty sure that I disabled Turbo Core in BIOS, so those voltages aren't important (yet) 😄
But, after 10 hours of small FFTs and blending, will try to enable it.

APM: I think I enabled it, 99% sure I did - will check after FFTs testing 10 hours finishes.

Will do 1/2 cores & Turbo core enabled - thanks for the suggestion. Just one thing: for this test, should I disable the cores in BIOS or should I do it by setting affinity to (some, which?) workes in Prime95?

LLC: it is on stock setting.

OC: I did some OC, not to high, just because I've never done it before. What I got was 4 GHz - Turbo disabled, and all other possible "extras" that make OC harder disabled.

What I first want to complete is "how low can it go" undervolting story, and after, "how high can it go" overclocking story 😄

When I was overclocking to 4 GHz, I got it stable at 1,2125 volts (manual in BIOS). The temps were: socket @ 57 °C and cores @ 43°C.
At that time I didn't pay attention that I should select balanced power profile in Windows, so "high performance" mode was active - therefore, this was the reason (I hope) of higher temps. As I stated, will do the overclocking after "lowering" voltages first - test take time.

As about testing, I'm used to doing these:
- Prime95 small FFTs 10 minutes
- Prime95 small FFTs 1 hour
- Prime95 small FFTs 10 hours
- Prime95 Blend 10 minutes
- Prime95 Blend 1 hour
- Prime95 Blend 10 hours
- Aida64 stability 10 minutes (everything checked, except the disks)
- Aida64 stability 2 hours

If, on any steps above, I get a error/bsod/something, I take a note.
After all of above pass, I call it stable. Is that a good practice? 😄

[elesde]

10 hours... that poor hardware 😃
But yea should be pretty stable after that.
You can also check the Windows event viewer for WHEA-Logger warnings regarding "corrected hardware errors", my FX CPU throws those a couple of MHz before Prime / y-crucher get any errors.

To test turbo stability you can ignore affinity, just make sure total CPU usage is below 25 or 50%. The Windows scheduler has the odd need to reshuffle a singlethreaded load across the cores every few milliseconds making it look like all cores are in use...

For both turbo P-states to work APM, Turbo and C6 must be active btw.

[0v3rl0rd]

Yeah, 10 hours of this and 10 of that... all in all, one whole day of testing for only one voltage setting. I hope that it is worth it 😄

About that WHEA-logger - first time I hear about that. Under what "tree" inside the Event Viewer should I find it? We're talking about Win 7 x64 SP1 here.

About that CPU usage testing for Turbo, I still don't understand how can I "make sure that the total CPU usage is below 25 or 50%"?

Btw, tnx for the info about Turbo and C6 activation 🙂

[elesde]

The WHEA-Logger warning should be in the "System" tab, guess not all CPUs throw them when becoming unstable. For me it was just funny because Prime was stable but I did get the occasional bluescreen while running long gaming sessions. Since then I also started looking for these warnings in the event log.

Turbo usually engages on light loads when the power used by the CPU is well below the TDP limit, for Prime it started only after I used 4 cores or less and even then only for very brief moments. To get the highest turbo state to activate (4GHz in case of a stock FX8320) at least 2 cores must be in C6 sleep state, so i usually let Prime/Y-C just run 2 threads. With all the stuff running in the background there are still enough idle cores around to go in C6 sleep 😄