Undervolting & Overclocking the Llano A6-3400M

The AMD A6-3400M comes with a significantly higher stock voltage than is needed to operate stably. Has anyone tweaked and stress tested all the P-states to lower voltages?

I've stress tested the boost state B0 (FID: 30 DID: 2 FREQ: 2300Mhz) at 1.1125V, which is significantly lower than the default, which is 1.325V, and I've also managed to decrease the default voltage on P6, as well.

So this has me wondering just how low all the P-states can go, but i just dont have time to do a 3hr stress test for every test. Which is why I'm wondering if someone else has already done something similar?

Incase anyone is wondering, the tool I'm using for undervolting / overclocking the Llano processor is K10Stat, and the computer that this is in, is a HP Pavilion G4

caesay

Posted 2012-04-20T20:10:21.927

Reputation: 336

Answers

In case anyone else stumbles upon this - after some stress-testing at each state (30 minutes, nothing thorough) I have come up with some settings for my processor that I am comfortable with and seem pretty stable.

State  FID  DID  Freq   Volt
B0     30   2    2300   1.0750
P0     24   2    2000   1.0000
P1     20   2    1800   0.9500
P2     16   2    1600   0.9000
P3     26   3    1400   0.8500
P4     20   3    1200   0.8000
P5     14   3    1000   0.7500
P6     16   4     800   0.7000

From what Ive read online 800Mhz is the lowest you can safely go, and I was fine with that number so I honestly didn't try anything lower - However, it would seem that moving upwards there is a trend upwards of 0.0250 per 100Mhz. My fourth core was always the one the failed first though. I think that if I was able to tweak the voltage per core I could've gotten lower on the Voltages but I am pretty comfortable where it is now.

Happy overclocking / undervolting / whatever!

caesay

Posted 2012-04-20T20:10:21.927

Reputation: 336

I think these are interesting as a reference value. However since each APU is different there is not guarantee that even an identical labeled APU will perform identically with your voltages. – Hennes – 2016-08-15T15:14:56.257

@Hennes: Surely anyone doing this themselves should do their own tests, but this is a great starting point. It's something that I wish I would've had when I started this. – caesay – 2016-12-27T13:24:54.047

Some stable values on my mobile Llano (E2-3000M):

2900 MHz ( FID 13 , DID 0 , 1.2375 V )
2400 MHz ( FID  8 , DID 0 , 1.1000 V )
1600 MHz ( FID  8 , DID 1 , 0.8750 V )
1200 MHz ( FID  8 , DID 2 , 0.7750 V )
 800 MHz ( FID  8 , DID 3 , 0.6875 V )
 600 MHz ( FID  8 , DID 4 , 0.6625 V )
 400 MHz ( FID  8 , DID 5 , 0.6000 V )

The 400 MHz might even be stable at 0.575 V , just haven't thoroughly stress-tested it at that voltage.

Defaults being 0.9375 V at 800 MHz , 1.1125 V at 1800 MHz , 1.4125 V at 2400 MHz.

It is actually stable at 3.2 GHz, without even having to set the max volts (1.4125 V ), but the temps can get a bit high , though still staying below 90 C at full APU (CPU+GPU) load; combinations of furmark, x264 , prime95, 7z - but the fan noise gets rather high.

Generally stays below 60 C at 2400 MHz under full load, and the fan hasn't even revved up from its minimum speed yet.

Idles about 10 C above ambient.

That said, you will have to do some stress testing to find your own stable values, unless you don't want to get them as good as possible. In that case, just start prime95 , start lowering the voltage (once a minute or so) on a specific speed until the computer freezes/crashes or prime95 throws an error. Then increase the voltage by 1 to 3 steps, until you don't get an error.

If it's then stable enough for your needs, then you're done (for that speed). If not, let the stress test run for at least 10 minutes and monitor the temps.

Some sources (using your specific CPU)

Matti Nescio

Posted 2012-04-20T20:10:21.927

Reputation: 364

+1 for the links, and reminding me that this post still existed. When I did a search originally for documents like that, I didn't find anything. – caesay – 2012-11-21T07:23:39.383