8350 Power Consumption

I've been tuning my rig for power efficiency so I thought I'd share some results... ;)
This chip will run daily stable ~5Ghz and bench ~5.3, but it takes a lot of v's and becomes very inefficient!
For daily use I've settled for the following clocks (IMC voltage is 1.275):

http://i30.photobucket.com/albums/c3...psba9dbe99.png

Just for reference, this is my main rig, it's water cooled and loaded for bear! :D
Here are some specs so you folks know what else is sucking power:

Five/120mm fans for the rads
Two/ 140mm fans for the case/HD cage
OCZ Mem cooler fans over the PWM heatsink
Aftermarket Zalman dual fan cooler for the 6870 vid card
MPC35X Pump and a C5F 990FX mobo.

With CnQ enabled here is my idle draw:
http://i30.photobucket.com/albums/c3...ps21aa407e.png

Here's the draw while running the WinRAR benchmark (represents max real world use, including WCG & encoding video):
http://i30.photobucket.com/albums/c3...ps831fe12d.png

And finally, the power draw while running P95 Blend as shown in the 1st SS:
http://i30.photobucket.com/albums/c3...pse702a308.png

Someone over in Xtreme News said the 8350 drew 200w by itself when OC'd...
That may be true when pushing the ABSOLUTE limits, but with a little tuning the 8350 isn't all that bad! :D

Nice work Dave, could you mess power cons. in x264fhd and r11.5? thx

Here you go Flank3r... ;)

Nothing too impressive as far as MAX clocks, but these were all done with low v clocks (as shown above).

http://i30.photobucket.com/albums/c3...ps6d291b1f.png

http://i30.photobucket.com/albums/c3...psd957b2ca.png

Keep in mind that when I ignore power requirements I've hit 9.2+ on CB11.5... :D

Here's a WinRAR bench @ 4.6 too, I know you didn't ask for it, but it's still pretty decent @ the low v clocks... :yepp:

http://i30.photobucket.com/albums/c3...ps695d913b.png

yes, this C11.5 score is one og the best what I got with Thuban at maximum :D...And at x264fhd is no change get this with "old" good Thuban

Quote:

---

Originally Posted by Daveburt714

http://i30.photobucket.com/albums/c3...ps6d291b1f.png

---

Nice ,the Fx-8350 (at 4.6 Ghz) beat Core i7-2600 ( at 4.9 Ghz) , :cool:

What he was claiming at first was that at stock it was a 200 Watt processor.

Then he moved on to claim turbo was over 200.
Then he moved on to saying overclocked.

Well my rig is sucking. About 480W from the wall (spiked at 608W, but i dont know when)

That is with 8350 @ 4,5GHz and 2x7970s

Quote:

---

Originally Posted by Evantaur

Well my rig is sucking. About 480W from the wall (spiked at 608W, but i dont know when)

That is with 8350 @ 4,5GHz and 2x7970s

---

I assume that was while running WCG? If so, I'm sure those 7970's are drawing a lot of power, especially if your using the mod to allow multiple HCC wu's.

Just for reference, here's what I get while running WCG (w/HCC on 6870 VC).

http://i30.photobucket.com/albums/c3...ps225c9865.png

http://i30.photobucket.com/albums/c3...ps28e4dfcf.png

Quote:

---

Originally Posted by Daveburt714

I assume that was while running WCG?

---

of course :D

so its safe to say vishera/piledriver cpus use much less power once OC'd then zmabezi/bulldozer?

Quote:

---

Originally Posted by tbone8ty

so its safe to say vishera/piledriver cpus use much less power once OC'd then zmabezi/bulldozer?

---

yes very safe to say

8-core FX within 45W TDP envelope...

Note for Prime95 LFFT: At stock specs the rated TDP was exceeded by 15.5% during Prime95.
The same margin is allowed for 45W TDP test too.

Frequency: 2.7GHz
Voltage: 0.925V

Prime95 LargeFFT
DCR Imax (A): 47.25A
DCR Vsen (V): 0.919V
DCR Pmax (W): 43.42W

43.42 / 1.155 ("Prime95 margin") = 37.6W
37.6W + 6.75W (BaseTDP = NB, Logics) = 44.345W

X264 R2230 64/10-bit BD-RAW 1080p -> MKV
Preset: Slow, Tune: Film, Bitrate: 9000, Output: MKV

DCR Imax (A): 41.5A
DCR Vsen (V): 0.919V
DCR Pmax (W): 38.13W

38.37W + 6.75W (BaseTDP = NB, Logics) = 44.89W

http://img830.imageshack.us/img830/1...ychosmoker.gif

With both P95 and OCCT stress testing I have seen the power consumption under full load change as much as ~90w depending on what is actually being run at the time you view the Kill-A-Watt meter. You also need to take into account the PSU efficiency and work the number viewed as a percentage of the actual PSU draw, i.e 90% for a good PSU.

^Woah!

Nice results "The Stilt" :D

he is magic man. He got glory of name in my review :)

371 Wattoosies?! In winter that's a good thing, you need the heat anyways.. ;)

Nice stilt

8 core fx laptop chip!!!

Very Impressive Stilt! :clap:

I was running mine ~4.85 @ 1.475v and power consumption was pretty bad...
That's why I decided to try and find a nice balance between OC and power usage.

At stock VID (1.3375), I could get stability ~4.3 Ghz. I never considered underclocking because I use this machine for serious work.
After some tuning and checking bench scores this seemed like the best balance for my uses. Yes, power consumtion is still a little high, but these clocks aren't bad... ;)

This seems like a decent sample (1244 PGT). Next time I get bored I'll see how low I can get it by underclocking.
If it'll get down to the numbers you got it may make a nice server/HTPC chip down the road... :yepp:

Dave, if i'm not mistaken The Stilt does measure amps with a clamp at the EPS wires (4/8 pin connector). This way he will measure CPU only current. You measure power drawn by the entire system, not just the CPU.

I'm not sure all the current drawn by the CPU is fed by just the EPS wires but form what i've read from The Stilt, that's the way i understand it. If you multiply this current times the Vcore you're using you get a power number in watt.
Divide this by the efficiency percentage of your PSU and you get number of CPU power drawn.

Maybe The Stilt can clarify and/ or acknowledge this this?

Well under 25W average now, the peak is barely over...

Note for Prime95 LFFT: At stock specs the rated TDP was exceeded by 15.5% during Prime95.
The same margin is allowed for 25W TDP test too.

I did not underclock the NCLK domain as I wanted to keep the L3 cache at stock performance level.

Frequency: 1.8GHz
Voltage: 0.7875V
NCLK Frequency: 2200MHz
Voltage: 1.075V

I cannot make separate DCR measurements for VRM Loop 2 (CNB) currently, so the new BaseTDP is an estimate.
Default BaseTDP = 6.75W (CNB 1.1625V -> 1.075V): 6.75 * ((1.075 / 1.1625)^2) = 5.77W

Prime95 LargeFFT
DCR Imax (A): 28.75A
DCR Vsen (V): 0.781V
DCR Pmax (W): 22.45W
EPS12V Clamp Imax (A): 2.95A
EPS12V Vin (V): 12.13V
EPS12V Pmax (W): 35.78W

22.45 / 1.155 ("Prime95 margin") = 19.4W
19.4W + 5.77W (BaseTDP = NB, Logics) = 25.17W

X264 R2230 64/10-bit BD-RAW 1080p -> MKV
Preset: Slow, Tune: Film, Bitrate: 9000, Output: MKV

DCR Imax (A): 27.75A
DCR Vsen (V): 0.781V
DCR Pmax (W): 21.7W
EPS12V Clamp Imax (A): 2.58A
EPS12V Vin (V): 12.13V
EPS12V Pmax (W): 31.3W

21.7W + 5.77W (BaseTDP = NB, Logics) = 27.47W

DCR = Current flow over VRM inductors
EPS12V = Current draw from EPS12V connector measured with a separate clamp meter (VRM Iin, both loops included).

Cooling: Passive Thermalright Ultra 120 Extreme (not very well suitable for the purpose...), peak temperature after Prime95 LFFT 63 celsius (TSI).

http://img843.imageshack.us/img843/4495/cb18g25w.png

very nice man! Is it possible add dynamically profiles to PS Check software? I would like in next week rebuilding my Bulldozer for Vishera and after try PS check software (I plaiyng with this only for OC records etc)

It will be more dynamic if you use Turbo Core PStates with reconfigured TDP limits.
Cannot be done with PSCheck, however Orochi Control Interface (release TBD) supports this...

Hope, we will see this utility soon

Quote:

---

Originally Posted by The Stilt

Well under 25W average now, the peak is barely over...

Note for Prime95 LFFT: At stock specs the rated TDP was exceeded by 15.5% during Prime95.
The same margin is allowed for 25W TDP test too.

I did not underclock the NCLK domain as I wanted to keep the L3 cache at stock performance level.

Frequency: 1.8GHz
Voltage: 0.7875V
NCLK Frequency: 2200MHz
Voltage: 1.075V

I cannot make separate DCR measurements for VRM Loop 2 (CNB) currently, so the new BaseTDP is an estimate.
Default BaseTDP = 6.75W (CNB 1.1625V -> 1.075V): 6.75 * ((1.075 / 1.1625)^2) = 5.77W

Prime95 LargeFFT
DCR Imax (A): 28.75A
DCR Vsen (V): 0.781V
DCR Pmax (W): 22.45W
EPS12V Clamp Imax (A): 2.95A
EPS12V Vin (V): 12.13V
EPS12V Pmax (W): 35.78W

22.45 / 1.155 ("Prime95 margin") = 19.4W
19.4W + 5.77W (BaseTDP = NB, Logics) = 25.17W

X264 R2230 64/10-bit BD-RAW 1080p -> MKV
Preset: Slow, Tune: Film, Bitrate: 9000, Output: MKV

DCR Imax (A): 27.75A
DCR Vsen (V): 0.781V
DCR Pmax (W): 21.7W
EPS12V Clamp Imax (A): 2.58A
EPS12V Vin (V): 12.13V
EPS12V Pmax (W): 31.3W

21.7W + 5.77W (BaseTDP = NB, Logics) = 27.47W

DCR = Current flow over VRM inductors
EPS12V = Current draw from EPS12V connector measured with a separate clamp meter (VRM Iin, both loops included).

Cooling: Passive Thermalright Ultra 120 Extreme (not very well suitable for the purpose...), peak temperature after Prime95 LFFT 63 celsius (TSI).

http://img843.imageshack.us/img843/4495/cb18g25w.png

---

so this is the new PS4 amd chip their using :)

where do i find the x246 benchmark?

is this it?

http://www.techarp.com/showarticle.aspx?artno=520

Quote:

---

Originally Posted by tbone8ty

so this is the new PS4 amd chip their using :)

---

PS4? Forgive my ignorance but do you mean Play Station 4 by chance?

yes I meant PlayStation 4

Got a 1249pgt fx-8350 today. Will test later tonight

Hopefully they weren't already on holiday when they spun the silicon haha lol

Quote:

---

Originally Posted by tbone8ty

where do i find the x246 benchmark?

is this it?

http://www.techarp.com/showarticle.aspx?artno=520

---

yes :)

Things seem to have improved a bit...
I see up to 15-20% reduction in power consumption on FX-83xx series parts since the launch ;)
The average seems to be around 11%.

Maybe the chip fairies have been working overtime :D

stilt, good to hear :D

i took out power test meter.

Idle is about 110 watt for the whole case.

At stock (volt 1.425) I get about 230 watt when I run CB benchmark.

Then I oc'ed to 4,6 or 4,7 (i forgot) and raised volt a little and still 250watt.

All this is CNQ and all other :banana::banana::banana::banana: enabled, a lot volt going up and down of course (lack of that load line)

Not bad for the Fx :) :up:

Quote:

---

Originally Posted by The Stilt

Things seem to have improved a bit...
I see up to 15-20% reduction in power consumption on FX-83xx series parts since the launch ;)
The average seems to be around 11%.

Maybe the chip fairies have been working overtime :D

---

GloFo's 32nm process is pretty well refined. They have 28nm running and 20nm and 14nm enroute. These will all provide additonal power savings, though 125w TDP under full load at default values isn't an issue for most folks. Lower is always nice but not a deal breaker for most people, especially considering most PCs are under full load for only a small percentage of time in typical use other than stress testing, looking form Primes, etc.

Well here are my results to add to Daveburt714 thread.

All power saving options enabled in bios.

Check sig for rig details (watercooled etc).

I havnt figured out what is going on with the voltages.

HWMonitor / CPU-Z show no change when the multiplier drops.

AMD Overdrive, shows changes, but the voltages are incorrect.

If anybody has any ideas on what I can do to get the correct voltages (excluding using a DMM let me know).

Idle 2.2Ghz

http://i.imgur.com/RM7WSQj.jpg

Idle 5Ghz

http://i.imgur.com/MZutCwH.jpg

Prime Full Load

http://i.imgur.com/YOJV3ly.jpg

Dont think I will be keeping the 5Ghz as 24/7 settings, as its not prime95 stable (IBT / OCCT are OK) as power consumption is huge

:p:

I have been living in a faith that Prime95 is the worst thing you can throw at your CPU.
I was wrong.

AMDs own test program (called "Hotdog") makes the chip draw almost 14% more power than Prime95 LargeFFT (V27.9) :D
Kabini and Kaveri optimized versions work on 15h Gen1/2 too, however they result a lot lower power draw. Maybe it is a sign ;)

Quote:

---

Originally Posted by The Stilt

I have been living in a faith that Prime95 is the worst thing you can throw at your CPU.
I was wrong.

AMDs own test program (called "Hotdog") makes the chip draw almost 14% more power than Prime95 LargeFFT (V27.9) :D
Kabini and Kaveri optimized versions work on 15h Gen1/2 too, however they result a lot lower power draw. Maybe it is a sign ;)

---

Any chance we can get access to it

:D

agree with mongoled, I want to have AMD stress program too :). AMD could try new program for biggest enthusiast in forums with some FTP link and we could testing new OC*tweaking etc softwares for users :).

"Hotdog" is part of AMD ThermNow software, which is under NDA.
Besides there is no real use for such a program, as the resulting stress does not correlate by any means with the stress levels generated by the real world applications:

DCR Peak Power by application (during 10 minutes):

100% - Hotdog 15h V0.02
87% - Linpack 11.0.1.005
87% - OCCT 4.40 (CPU OCCT - SDS)
87% - Prime95 V27.9 (LargeFFT)
83.8% - AOD 4.2.6 (Calculation Test)
83.8% - Prime95 V27.9 (SmallFFT)
72.9% - Blender 2.66a (Burning Plane scene)
70.6% - Cinebench R11.5
68.4% - X264 R2273
67.3% - Euler3D Stars (CFD Benchmark)
66.2% - HyperPI 0.99b
65% - AOD 4.2.6 (CPU Test)
62.9% - AOD 4.2.6 (FPU Test)
56.2% - wPrime 2.09

how about AMD overdrive stability test? Can you compare it with others?

Added.

83.8% - AOD 4.2.6 (Calculation Test)
65% - AOD 4.2.6 (CPU Test)
62.9% - AOD 4.2.6 (FPU Test)

Quote:

---

Originally Posted by FlanK3r

how about AMD overdrive stability test? Can you compare it with others?

---

It looks like AMD stopped updating AOD a while back, it can't even read the sensors on my sabertooth 990FX R2.0

Last version is 4.2.6
http://download.amd.com/Desktop/aod_setup_4.2.6.exe

I use some 4.2.x. I think, we will se some update with Richland or with Vishera new models at summer or so.

Quote:

---

Originally Posted by FlanK3r

Last version is 4.2.6
http://download.amd.com/Desktop/aod_setup_4.2.6.exe

I use some 4.2.x. I think, we will se some update with Richland or with Vishera new models at summer or so.

---

yes it's been in dire need of a revamp.

I'd like to see some SSE (1,2,3,3S,4.1,4.2,4.1A), AVX, AES, XOP, FMA load tests

and some kind of way of checking which is being used on the cpu.

A bit off topic, but..

http://img43.imageshack.us/img43/1407/xc6b.png

http://valid.canardpc.com/show_oc.php?id=2837872

And thats a ultra low-leak part.
The first run too ;)

Should I try 6GHz?

:D

Quote:

---

Originally Posted by The Stilt

Should I try 6GHz?

:D

---

on water? :shocked:

edit: nvm noticed temps only now

omg, you say, this is with air/watter ?! My best is 5740...

On AMD FX (Asetek 570LC) watercooler ;)
It has only one of the original high cfm fans installed too, the second one is some random 120mm fan.

Replaced the exhaust fan on AMD FX CLWC with a Scythe Ultra Kaze.
The ambient was lowered by couple of degrees once the window was opened.

http://img853.imageshack.us/img853/591/7rs8.jpg

http://valid.canardpc.com/show_oc.php?id=2838300

Now I would say that the 'big nine' is only a matter of helium.

What was the ambient ? Cold weather still in Finland, right ?

Quote:

---

Originally Posted by felix_w

What was the ambient ? Cold weather still in Finland, right ?

---

Now it is a bit colder again, this morning it was around 16 celsius outside.
The ambient was at 24 celsius during the run.

Impressive...not very low temp as i expected, yet your result is probably one of or the highest with this type of cooling...

Quote:

---

Originally Posted by felix_w

Impressive...not very low temp as i expected, yet your result is probably one of or the highest with this type of cooling...

---

Off topic, but still:

Our cold weather is during ~2/3 of the year, The second half of June, all of July and the first half of August are quite nice. Some years the temperature range is ~40 degrees Celsius over the year, extreme years even more :). The average minimum is ~-10, while the average maximum is around ~20-25 degrees in southern Finland. The record temps are something around -35 and 35.


Closer to topic:

@The Stilt: You wouldn't happpen to be an engineer, would you? Or is it pure genius? :)

Because the things you do for the community mean alot. I don't think there's anyone else contributing in the same sense, really.

Quote:

---

Originally Posted by The Stilt

Replaced the exhaust fan on AMD FX CLWC with a Scythe Ultra Kaze.
The ambient was lowered by couple of degrees once the window was opened.

http://img853.imageshack.us/img853/591/7rs8.jpg

http://valid.canardpc.com/show_oc.php?id=2838300

Now I would say that the 'big nine' is only a matter of helium.

---

omg! First air/watter 6GHz with normal ambients! Wow...It could be great for LN?

6ghz.. impresssive!!!!!

wow and a 8320 to!

Master stilt good job!


maybe i should swap my air cooler nzxt havic 140 out for one of these fancy closed loop water coolers

any news about hotdog stress test?:)