A10-5800K undervolting

All about them.

Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee

Post Reply
johannes
Posts: 53
Joined: Tue Oct 23, 2012 8:13 am
Location: Suisse Romande

A10-5800K undervolting

Post by johannes » Sun Dec 23, 2012 2:53 am

Hello,

I recently upgraded my family's PC and it is now running the following configuration:

AMD A10-5800k
Arctic Alpine 64 Plus cooler
Asus F2A85-M LE
2x4GB 1866 RAM (Patriot viper 3 black mamba)
Samsung 250GB HDD
Cooler Master Elite 33x case
Sharkoon SHA450-8P psu

The PSU is not 80+ certified and probably quite inefficient. The old system (Athlon64 X2 4400 EE, Radeon 1950) used 100W AC when idle and about 215W when using the OCCT PSU test. (All power measurements were done using a ProfiTec KD302 and I rounded the values to the closest multiple of 5 when the values where changing a lot.)

The new configuration uses about 43W idle. Since I can only change the VCore for full load this is the same value for all configurations.

I wanted to see if by underclocking and -volting the power consumption of the A10-5800K can be easily brought down to the level of the 65W A10-5700. To simulate an A10-5700, I downclocked the processor to 3.4GHz which also disabled Turbo Core and reduced the voltage.

There were short power usage spikes when starting some tests, I've included these values in parenthesis. All power usage values are AC as shown by the power meter. Stock VCore is 1.325V.

Code: Select all

            Stock       [email protected]   [email protected]
Bios        115W        100W            90W
OCCT CPU    145W(160W)  130W            115W(120W)
OCCT 3D     130W        125W            120W
OCCT PSU    180w(197W)  170W(185W)      155W(162W)
I haven't tried to change any other voltages to reduce GPU power consumption (I can do a test if anyone is interested), but in general the results are not encouraging. 30W AC in the CPU test and 25W AC (at an efficiency of probably less than 80%, so the DC difference is even lower) difference at load are not even close to the 35W TDP difference between an A10-5800k and an A10-5700. For my planned Trinity Mini-ITX build for gaming (to be done once there's a decent FM2 mini-itx board...) I'll go with the A10-5700.

I hope these numbers help someone in a situation similar to mine.

- Johannes

Milli
Posts: 4
Joined: Sat Jan 26, 2013 2:55 am
Location: Belgium

Re: A10-5800K undervolting

Post by Milli » Sat Jan 26, 2013 3:11 am

I'm interested in some more testing because I'm planning something similar. But I'd rather do it with an A10-5800K because I want to go even lower than 65W. The A10-5700 being multiplier locked makes this harder (or is it possible to decrease the multiplier?).
Anyway, there's a slight GPU clock difference between the two (800Mhz vs 760Mhz) and I suppose also an northbridge voltage difference. Would you be kind to test and see how much can be gained from decreasing these two parameters?
I guess you're running your memory at 1866Mhz? I'm guessing that this will prevent really low voltages on the northbridge. You might want to try with the memory set to 1600Mhz.
But it's possible that your motherboard is just inefficient at lower power ranges.

johannes
Posts: 53
Joined: Tue Oct 23, 2012 8:13 am
Location: Suisse Romande

Re: A10-5800K undervolting

Post by johannes » Sat Jan 26, 2013 9:48 am

Hi Milli,

I'll gladly do some more tests, but unfortunately it won't be possible before Easter. (I live in a different country than my family.)

The power numbers porkchop posted suggest that the influence of the graphics is quite big and using a lower voltage is very much worth it. He was running it at 608 [email protected] (compared to [email protected]).

And yes, I confirm that the memory was running at 1866 MHz and with relatively tight timings. For my personal mini-itx build I'm unwilling to compromise on memory speed since the graphics are already limited by it. (High details are no problem with many games at 720p, but anti-aliasing is a problem even with older games in my limited testing.) But I remember that you could set the speed of the integrated graphics in the bios, so you should be able to run the graphics at a speed similar to the mobile Trinity processors.

I hope I could help you a bit, even if I couldn't run the numbers right now. If you're lucky the Zotac A75 board (the apparently least bad announced FM2-mini-itx board) will be offered in mid-February and I can run some numbers with an A10-5700 already earlier.

- Johannes

Milli
Posts: 4
Joined: Sat Jan 26, 2013 2:55 am
Location: Belgium

Re: A10-5800K undervolting

Post by Milli » Sun Jan 27, 2013 7:30 am

Thanks for the heads up.
It seems the GPU voltage isn't adjustable in the bios. I wonder if AMD Overdrive or some other software can do it. Maybe it's just done automatically (which would be nice) or maybe it's derived from the northbridge voltage.
What I found interesting about porkchop's post is the PicoPSU results since I'm planning to use a MS-Tech ITX case which includes a 60W AC adapter and a (unrated) DC-DC converter. I'm going to try to use a higher rated (~80W) AC adapter and see if the included DC-DC converter can take it.
It surprises me thought that he couldn't go lower than 1.150v @ 2.8Ghz (while he could do 3.6Ghz with 1.156v) which makes me wonder if he turned off (or setup the max) Turbo Core.
One feature that Intel has, is the ability to setup a TDP ceiling in the BIOS which would be a life saver in this situation. I know Opterons have this feature too but as far as I can see, FM2 doesn't.
Anyhow, an A10-4600M runs at 2.3Ghz and it's GPU at 496Mhz, all under 35W. So an A10-5800K with the exact same setting must at least be able to do 45W (if not less). The 4600M's max rated memory frequency is 1600 so probably it's northbridge frequency and voltage must be lower than the 5800K (anyone knows the 4600M's northbridge frequency?).
I'm looking forward to some more testing.

porkchop
Posts: 496
Joined: Thu May 25, 2006 1:19 am
Location: Australia

Re: A10-5800K undervolting

Post by porkchop » Sun Jan 27, 2013 3:59 pm

i didn't undervolt during the 2.8ghz tests.

turbocore was off the whole time.

the gpu voltage is adjusted automatically and as far as i can tell, is unrelated to the nb voltage

the reason the idle for the undervolted 3.6ghz results are high is because the freq was locked at 3.6ghz for the whole test.

i should have explained all of this in my post...

johannes
Posts: 53
Joined: Tue Oct 23, 2012 8:13 am
Location: Suisse Romande

Re: A10-5800K undervolting

Post by johannes » Mon Jan 28, 2013 10:25 pm

I'm afraid you might have problems starting the PC and entering the bios with a PSU rated so lowly. My Asus board would run the CPU at maximum (non-turbo) frequency in bios, but it would respect undervolting and -clocking.
That means that at least for the initial set-up (power draw > 80W) you would need a different PSU. I don't know if there's a short power spike before the frequency and voltage settings are applied, but I couldn't observe any.
Porkchop apparently had no problems starting the system with a 65W APU and a 80W PSU, but then his system would only use 47W in a CPU stress test whereas my A10 used a lot more. (NB. my PSU might be inefficient and the real power usage might well be 25% or more percent lower.)

Post Reply