Mini-ITX A10-5700 power numbers, Turbo Core influence

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

Mini-ITX A10-5700 power numbers, Turbo Core influence

Post by johannes » Sun Apr 07, 2013 10:04 am

(SPCR has reviewed this processor already. I was surprised by some of the numbers I got, so I'll post them here.)

I recently bought a FM2 mini-itx system. The goal was 720p gaming in a very small case. I'll post pictures later, for now I'll share my observations about the influence of Turbo Core (TC).

The configuration:
AMD A10-5700 (65W Trinity, 3.4GHz, 4 GHz Turbo)
MSI FM2-A75IA-E53 (BIOS v1.2)
2x4 GB G.Skill Sniper Memory (1866 MHz 9-10-8-28, 1.5V)
Samsung SSD 830 128GB
PicoPSU 160XT with 192W external brick (89% avg. efficiency)
Noctua NH-L9a CPU cooler
M350 case (case not closed during tests, i.e. open air results)

Short remark about noise: It is quiet with the case open, but at idle there is coil whine from the PSU (disabling C6 improves this a bit). With the case closed and an added 40mm fan it is not silent, but below the noise floor when gaming.

Power numbers (230V AC, taken with a ProfiTec KD302):
BIOS: 96W, later falling to 84W (78W w/o TC)
Windows 8 Desktop Idle: 23W

With turbo core:
OCCT CPU test: 86W (CPU frequency jumping all the time between 3000-3700 MHz)
OCCT 3D test: 65W (DX11, 720p, Shader 7, ~40fps)
OCCT PSU test: 85-100W (CPU ~2x00-3700MHz, 74-81fps)

Without turbo core:
OCCT CPU test: 84W (stable 3400 MHz)
OCCT 3D test: 72W (~44fps)
OCCT PSU test: 126W (~85fps)
Witcher 2: 75-85W (Loc Muinne market, casting some Igni to create some special effects)
Test Drive Unlimited 2 (TDU2): 75-90W

I retried the PSU test w/o TC after the system was turned off and permitted to cool down and I got 120W. Since the PicoPSU is only rated for 96W 12V continuous, I interrupted the test in both cases to avoid damaging hardware.

For gaming use, disabling Turbo Core seems to be a good idea. The GPU gets more headroom which results in higher and more stable fps in the OCCT 3D and PSU tests. (Edit: Disabling TC did not solve the micro stutter in TDU2 contrary to what I claimed earlier.)

However, it appears that disabling Turbo Core also removes some limiter when using both CPU and GPU at 100%. >120W power consumption is scary with a 65W APU that has 23W idle power consumption. Gaming tests don't show this as a problem, but I'll try this with Metro 2033 in the future to make sure there are no issues.

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

Re: Mini-ITX A10-5700 power numbers, Turbo Core influence

Post by johannes » Wed Apr 10, 2013 10:06 am

Metro also has power consumption figures between 75 and 90 Watt on the wall, so gaming is probably safe. Nonetheless, I could reproduce my 126 Watt figure when starting the OCCT PSU test afterwards, so something isn't working as it should.

CPU VCore is all the time at about 1.23V, Uncore hovers around 1.13V and VIN0 (GPU?) is stable at 1.28V. While the RAM is being populated I see figures around 100W, as soon as all cores kick in for the Linpack portion I get the 126W figure. FPS are stable at 80±2 [1], which leads me to believe that disabling Turbo Core simply puts the CPU and GPU parts in their highest performance state without caring for power consumption. Maybe something will be throttled when a certain temperature is reached, but I didn't want to keep my PSU out of spec to test it.

The jumping around of the CPU frequency with turbo core enabled (going to lower 2000er values and then back up to a turbo state) is probably responsible for the clearly lower power consumption in this test. The changing fps might be a sign for changing GPU power states. The graphs for VIN0 show regular changes from 1.23V to 1.28V.

This removes the amazement I felt when seeing the power figures for the 5700 after doing my own tests with a 5800k. There's no magic way to save power involved, but throttling to keep in spec. Looking at my own data and assuming 80% efficiency for the PSU, the system with the undervolted 5800k clocked down to 5700 speed uses about 125W DC, while with the 90% PSU the 5700 system uses about 115W DC. I believe the difference in fans, hdd vs sdd, uncore and gpu speed might account for these 10 Watts. (and my calculations are of course not very exact anyway).

[1] I was running an extraction in the background which might be responsible for the slightly lower fps compared to my first test.

fastturtle
Posts: 198
Joined: Thu May 19, 2005 12:48 pm
Location: Shi-Khan: Vulcan or MosEisley Tattonnie

Re: Mini-ITX A10-5700 power numbers, Turbo Core influence

Post by fastturtle » Wed Apr 10, 2013 8:07 pm

From reading the AMD specs, the APU can really put a load on things - TDP quickly goes through the roof.

Your correct in your thinking that by turning off the TC, that they chip kicks into a high power state and does absolutely no stepping but and this is a big one, if you underclock/volt things in that state, you may be able to find a middle ground with decent performance while keeping the TDP low enough for silent cooling. Not sure where it would be but it's there.

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

Re: Mini-ITX A10-5700 power numbers, Turbo Core influence

Post by johannes » Wed Apr 10, 2013 11:49 pm

Thanks for confirming my suspicion and your suggestion!

Fortunately, just yesterday a new program to undervolt Trinity APUs was released. I hope it doesn't disable my wireless+Bluetooth like AMD Overdrive does and I can do some voltage testing.

I'm not sure if I can reduce the multiplicator of the locked APU, but I would gladly trade some hundred MHz for significantly better thermals. I'll do some tests over the next days to find the sweet point.

If I can't change the multiplicator...well, this gives me another excuse to go the 11h to visit my family and I can just exchange the 5700 and the 5800k.

QUIET!
Posts: 238
Joined: Sat May 11, 2013 8:33 am

Re: Mini-ITX A10-5700 power numbers, Turbo Core influence

Post by QUIET! » Wed May 15, 2013 8:06 pm

A picopsu might be 89% average efficiency but that's just going from 12v to 5v and 3.3v, but that ignores the efficiency of the brick and at full load you might not be getting "average" efficiency.

If your brick is 80% efficient at that load, 25+ watts is lost heat and after you subtract a few watts for the motherboard and ram, your power consumption isn't looking that bad.

Lastly, the CPU die and package are exactly the same between the 5700 and 5800. If your cooling is up to task and your die temps are OK, there is no reason why you can't run just as much power through a 5700.

I'm not sure how AMD bins their die but if their yield is high, they might not bin them at all, they could just laser trim them to produce whichever model needed to fulfill orders.

TDP numbers are mostly provided to tell OEMS what kind of cooler they need to run at the advertised frequency without throttling too much.

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

Re: Mini-ITX A10-5700 power numbers, Turbo Core influence

Post by johannes » Wed May 15, 2013 11:50 pm

My power brick is 89% avg. efficiency. :-) The PicoPSU 160XT is specified as being 92-96% efficient for 12V-5V/3.3V conversion and should have no real loss for 12V.

I think you are correct and my processor is not binned at all. I compared my undervolted/underclocked numbers (both work well using amdmsrtweaker) to my A10-5800k numbers and the results are more or less the same.

I later found out that AMD implements indeed some form of TDP control called APM which will change CPU and GPU states to keep inside some power target. Disabling turbo disabled this for me, but you can use tools such as amdmsrtweaker to disable it at run time, too.

Now for more interesting data:

The lowest idle power (Win8 desktop) I've seen was 14-15W. This was with AMD Overdrive installed which disabled my integrated Bluetooth and Wifi. These are quite nice numbers which might even be lower with a more appropriately specced power brick and with display output disabled. (My 192W brick most probably doesn't have a good efficiency at 8% load).

The effect of undervolting is huge.[1] My processor is worse than others I've seen on-line, but the effect is still very noticeable.

Code: Select all

           OCCT CPU  OCCT PSU
3.4@stock    84W       126W
[email protected]   76W       116W
[email protected]   61W       104W
[email protected]    54W        98W
[email protected]      52W        94W
Turbo goes in the 1.4V range which probably results in similarly much higher power consumption.

Unfortunately I haven't found a way to reduce GPU voltage so far (and I haven't tested NB voltage changes) which would reduce heat even further.

[1] My last processor before Trinity was a 45W Athlon X2 and I'm probably more easily amazed by the effect of VCore changes than most people here. ;-)

QUIET!
Posts: 238
Joined: Sat May 11, 2013 8:33 am

Re: Mini-ITX A10-5700 power numbers, Turbo Core influence

Post by QUIET! » Thu May 16, 2013 5:10 am

Have you tested the efficiency?

Relying on specs does not always work, I've heard of at least one brick that missed its specs badly, I doubt its uncommon.

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

Re: Mini-ITX A10-5700 power numbers, Turbo Core influence

Post by johannes » Thu May 16, 2013 8:20 am

QUIET! wrote:Have you tested the efficiency?

Relying on specs does not always work, I've heard of at least one brick that missed its specs badly, I doubt its uncommon.
No, I don't have the equipment for testing efficiency, only a cheap power meter. I can only hope that energy star only gives a level V rating to PSUs with avg. efficiency of >= 87% and that the testing result of 89.65% avg. (230V) efficiency from the spec sheet is no lie, but you could of course be right.

I just know that there are are noises on idle which make me glad that I don't use this machine for work. They are not very loud, but still distracting.

Post Reply