Bitcoin Forum

I didn't see a thread like this yet, so I am starting one.  While there is plenty of info on graphic card power usage, I couldn't find a good resource on total system wattage, as measured from the wall (using kill-a-watt meter or similar device).  It seems like it would be beneficial to have this as a refence thread so you can see how efficient your rigs are compared to others.

I haven't measured all my systems yet, but I do have one bit of data to start the thread with.

1.924 mhash/w   533mhash, 277W.
FX-8120, 2X 5830 overclocked and undervolted, win7 64, 12.1 catalyst, diablominer, 80 plus silver power supply.

Actually, I think a big problem is too many people posting from the wall power usage as card usage


Anyway

I get 307ish MH/s at 200 to 205 Watts

75 Watts when the system is idle, so the card uses 125 to 130 (assuming the card uses 100% of that extra 5 watts)


Oh, uhh 1.4975 MH/W

Whenever my other cards come in, efficiency will jump up... approx 465 Watts, 921 MH/s or 1.98 MH/W

2.72 MH/W 2300 MH/S @ 845W (at the wall, dedicated 240V circuit).
2.65 MH/W 2300 MH/S @ 870W (same system on 120V circuit)

2.46 2.54 MH/W -- 550 554 MH/s @ 224W 218W at the wall

I'm kinda crippled because I only have one 5970 mining.  I think I can get the total system power down if I can figure out how to undervolt the CPU.

Edit: Ok, I figured out how to underclock my CPU and PCI ports.  Curiously, my hash rate bumped up a little bit too.

550mhash is really low for a 5970. You should be getting like 700ish or higher

I can hit 840 MH/s, but this thread is about efficiency.

1350MH/s, 610W at the plug = 2.21MH/s/W

4 x 6950 @ 1.012Volts, 800MHz core, 700Mhz Mem. Shaders on one 6950 is unlocked.
1xVelociraptor HDD
PhenomII X6 2.8GHz @ 800MHz

1.67 GH/s, 920 W -> 1.82 MH/W
4x 5770
3x 5750
1x 5830

Three boxes w/ ancient PSUs and Celeron D / AMD X4 space heater CPUs. Thinking of moving over to undervolted 5970s land.. :P

All wattage values measured at the wall with a kill-a-watt.
Rig 1: 600 MH/s 390 W ->  1.53 MH/W 2x 5830's
Rig 2: 320 MH/s 200 W -> 1.6 MH/W 1x 5830

As a side note, rig 2 has an old athlon 64 4800+ underclocked to 1 Ghz @ .775 v. The difference between idle and full cpu load was 9 watts, down from its rated 65.

1685 MH/s, 660 Watts from the wall, 2.55 MH/W

What setup?

5x7970
filesys:    wd green 500g  win7  diablo
voltage:  1.056
clocks:   1049/685
killwatt:  1160
Mh:        3120

2.69 MH/W



Dat.  how did you measure this on the 240v, i have my 79xx on 220 now, but cannot measure the watts directly

TADA ....

http://www.toolorbit.com/image/Klein-Tools/Klein-Tools-CL2000-G0.jpg

Well I measured amps & volts.  Remember the two legs of 240V circuit in US wiring are out of phase so to read current you need to clamp just one conductor. 

They make adapters like this:
http://www.service.kleintools.com/Marketing/Catalog_Imagery/69400_ICON.JPG

but I have the rigs connected to a PDU using C14 to C13 cables and I doubt anyone makes an adapter so I modified one power cable which I use for testing only.

I do have that tool,  how else can I keep from burning the house down :)

but the reading is not exact..  jumps around and what you write down is..  well an educated guess.

edit:
 my new rig(or is it 2 rigs joined by a common psu)  I will be easily able to do this..

it is 10x7970  3psu's  6790 Mh.  I will calculate the watts tonight.

If you have a model w/ true RMS it shoudn't be jumping around.  Cheaper units measure mean current and then calculate "guestimate" RMS.  This is fine is power is perfectly a sine wave but some loads disrupt the sine wave.  Loads like computer power supplies.  :)

If you are sure your model is True RMS you also need to make sure that the conductor is in the center of the clamp and doesn't move.  If you move the conductor around you will change the induced current and thus the reading.   A kill-a-watt works the same way so done right readings shouldn't be any less accurate than a Kill-a-watt.

I have thought about getting one of these to provide measurements for entire circuit over time but that is just crazy right?
http://www.ccontrolsys.com/w/Advanced_Pulse_WattNode_-_Models

Hi, my name is Jim, and I have way too much money laying around :P

I didn't notice this.  IIRC nobody has gotten more than 8 GPU working due to limits in AMD drivers.  This includes both windows & linux, 32 bit and 64bit.  At one time AMD had a 4 GPU limit on windows but thankfully fixed that.

Sounds like 1 PSU per mobo and a third for some of the cards. Good idea actually.

Jesus, Jimm is the king of multiple psus. I can't even get one rig stable with three 5970's and 2 seasonic gold 750watt joined with a lian li connector and he just uses f'in paper clips to short the 24 pin to make his work.

There must be some major power fluctuations or issues at my brother-in-laws house where these rigs are  or something. Extenders maybe? Hell I don't know, but Jimm is crazy.

I'd be happy to get this rig stable:
3 5970 at 800/300
MSI 890FXA-GD70
sempron 145
2 x Seasonic 750 watt gold psus with connector
2 Gb ram.
open case on pci e x16 powered or unpowered extenders(I have both)
linuxcoin on 16GB flash drive with cgminer.

oh well, maybe someday.

Worse case scenario ... ebay the 2x 750 (or try these forums) and buy a single 1200W.  1000W is all you need but 1200W gives you the option to expand to 4x5970s.

a silverstone 1200 will NOT run 4 5970's overclocked to 820.  a seasonic 1250 will

Dat:
  in reference to the 10 gpus  it has 2 mobos and 2 operating systems..  . I read somewhere that solaris will load Xnumber of gpus?

IT IS:



The rig(s) are still up and stable, albiet running a little warm for my tastes (high 70's)

Care to share with us that magic solaris or whatever it is called server that does 10 GPUs in one mobo ? Pics would be nice  ;D

Thanks !

There are 2 mobos, not 1....

jjiimm can you post core voltages at which you got these frequencies of your cards?
And what do you mean by high 70's? 78? :)

The volts are listed with the clocks..  1.143 and 1.125  Please do not take my volts as the 'optimum'  I do not follow a very scientific approach to arrive at these numbers..

Ah lol somehow i didnt see the voltages :)
Well if its stable then its good enough will try em out, what brand 7970 u got?

edit: Do you use pci-e extenders? With or without molexes?
Im asking cuz im waiting for extenders to setup 4x 7970 rigs, but i bought ones without molexes so i wonder if im gonna have to put them there myself..

I have never used powered ext. i do not think we need them anymore...  maybe with the older mobos?,  I always use extenders for cooling.

Ok then its all fine if you can run 5x7970 on extenders without molexes then i can do 4 with np :P I have asus sabertooth 990fx mobos with four pcie 16 slots, i guess it will work, if not i report here something got fried :D

I tried your volt/freq but for me it crashes, what manufacturer is your 7970, Sapphire (i got sapphires..) or something else ?

xfx will do pics
we should move this to the 7970 thread, this is for Mh/J
http://imageshack.us/g/832/img0536sx.jpg/




PSU?

5x 5850

2,06 MH/s / Watt -> 371MH/s @ 180Watt

I will be using enermax platimax 1500W, i think 1 could take 5 cards, but for now it will be 4 per rig  because i didnt order pcie 1x to 16x extenders..

Did you take power factor into account when you measured volts and amps separately?  I'm running at PF about .9 so maybe that caused the difference if you measured different ways for 120 vs 240.

830 MH/s / 520 W = 1.6 MH/J

3x5830, 80+ PSU, 120V

measured with kill-a-watt

720 MH/s @ 335W ~ 2,15 MH/J

measured with Voltcraft Energy Check 3000

Setup:

MSI GD70
AMD X2 250
2 GB Corsair XMS 1333 MHz
250 GB  Western Digital 2,5" HDD
HD 5850 rev 2 @ 850/150/1.088V
HD 5850 rev 1 @ 850/150/1.088V
PSU: Cougar S700  (230V EU/Germany)

Cooling/Case:
modified Cooler Master CM690 rev 1
1x Scythe Big Shuriken (passive)
2x Arctic Accelero S1 Plus
2x Noctua NF-S12B FLX
1x Noctua NF-P14 FLX
1x Noctua NF-P12-1300
2x Cooler Master 120 mm

temperature GPU: 55 / 42 C°
temperature System/CPU: ~ 32 C°

CPU undervolting/clocking: no
disabled onboard devices: Audio / Serial Port
GPU overclocking: 765->850 / 725-> 850
GPU over/undervolting: no
CPU Load: ~ 0-9%
GPU Load: 99%
Miner Software: cgminer 2.2.1
Driver / APP SDK: 11.12 / 2.5
OS: Windows 7 x86_64 Ultimate

Power Draw start up : ~ 380W
Power Draw boot : ~ 180W
Power Draw idle: ~ 130W
Power Draw mining: 335W


 ;D

I just spent some time optimizing one of my rigs -

890FXA-GD70
Sempron @ 1600mhz
Cooler Master 1000w 80+Gold
All mobo options disabled, USB drive/linuxcoin

3x 5970 + 1x 5870, undervolting & underclocking all the cards
2317Mh/s @ 880w - 2.63Mh/J (measured with kill-a-watt)

:)

Could optimize it a bit but for now it is...

"Kong"
Running BAMT

2x750 Seasonic Gold w/ Add2PSU
Asrock 970 Extreme4
Sempron 145
1 x 2GB DIMM
4GB USB Key
2 x 5970 (800/300)
2 x 5870 (970/300)
1 x 5770 (945/300)
2 x 120x38mm Delta Screaming Fans

2,580 MH/s - 1060W = 2.43Mh/W

If I remove the fans I get almost to my target of 2.5Mh/W.  I am sure I could look at under volting and cranking back to further improve efficiency Mh/J as well.  A project for the weekend.

PC Power Cooler 950w 87% efficiency
Asus P7P55 WS
i5 670 @ 1.6ghz 0.9v
2gb @ 1.5v
WD Blue HDD
Win7 x64 CGMiner/Phatk
11.12 divers w/2.5sdk swap
3x5970's @ 815/300 1.05v
1x5850 @ 725/300 0.95v

http://i42.tinypic.com/24l2yqq.jpg

2515mh/s 970w
2.59mh/w

Still tweaking just got on board with CGminer, next Linux USB drive, more cpu/gpu tweaking, and 240v.

Anyone know where to buy 250 VOLT NEMA 6-20R POWER STRIP?

http://www.internationalconfig.com/icc6.asp?item=61820
That doesn't cost $300? Looking at you DAT.

What I did was just wire a regular 20amp plug with 2 hots and a ground.

Thanks jimm was thinking the same but what about surge protection?

Am I missing something here?  950W PSU pulling 970W?  I understand the output (which is how PSU's are measured) isn't 970 (probably 850) but man that is pushing it, no?

The word you are looking for is Power Distribution Unit (PDU).  Searching for "PDU" will help.  Also most units are "officially" 208V (single phase, don't buy 3 phase ) because that is what data centers are wired for.    They work fine at 230V/240V/250V used in residences.  Sometimes they are advertised 208V/240V sometimes just 208V.

Shortcut?  

APC model AP9571.  There are a lot of them on ebay for ~$50.  Very solidly built.  It is 30A so it uses NEMA L6-30P plug.  You likely want 30A unit anyways because PDUs are derated 20% to comply with electrical code. 30A unit has 2x12A breakers = 24A usable.  20A unit has 2x8A breakers = 16A usable.   Seems there are more used 30A units then 20A units likely because that is what most datacenters use.

Best part.  The cable is very thick and 12 feet long.  No worries about overheating, or damaging the insulation.

specs & manual:
http://www.apcmedia.com/salestools/ASTE-6Z6JWV_R0_EN.pdf

http://www.apcmedia.com/resource/images/500/Front_Left/1DAC4746-5056-9170-D315B2D6EDB25582_pr.jpg
http://www.apcmedia.com/resource/images/500/Back/1DB1E75F-5056-9170-D36BBE5ADBD8924C_pr.jpg

To add to my "you might be a serious miner" series.
If your first thought when seeing the pics above is "I wish I had that" you might be a serious miner. :)

Real world efficiency is likely closer to 85% on that unit so that is more like 820W or 85% of rated DC power.  It is running a little heavy but modern (as in built in last 3-4 years after huge wattage GPU and 80Plus specs) have pretty flat curve.  They tend to run just as well at 75%-80% load as 50% load.  Higher efficiency, flatter curves, beefier heat sinks, direct DC switching, and giant 140mm fans means the "age old standard of keep it at 50% of peak" is kinda outdated.

Now older units (say 5-6 years ago) had very spikey curve so even if peak efficiency was 80% at 80% load it is more like <60% which means a lot of heat.  Combine that with tiny heat sinks, inferior switching designs, and wimpy 80mm fans pushing a heavy load was not a good idea.  

Very good units like Seasonic 80Plus-Gold tend to be >88% efficient at 100% load which is better than peak efficiency (at half load) over the best units not so long ago.  I wouldn't try pushing so no-name junk that hard though.

direct wire seasonic into 240 :)

(what's a surge protector?) ;)

D&T - Thanks.  It seems we always revert to the 80/20 rule.  Don't put more than 80% load on the circuit (especially since it is constant) and then not more than 80% load on the PSU.  The breakthrough for my simple non-electrician mind was realizing a PSU wattage is output and not draw (i.e., your Kill-a-Watt can run closer to 90% of rated watts when using a good PSU).

I have 4 Seasonic x-1250's en-route  ;D

Hoping that will solve my dual PSU woes with 2x750 Seasonic gold (https://bitcointalk.org/index.php?topic=63147.msg738546#msg738546 ).

The seasonics are rated at 100% constant.....  also they will peak at much higher then advertised constant.

most psu's advertise they're peak amount

They are awesome.  I love all the attention to detail.  It is IMHO the best PSU I have ever used. 

I have some CoolerMaster 80Plus-Gold units I got cheap and my workstation has an Enermax 1350W unit.  Both are solid workers but the Seasonic is just so much nicer to work with.

1) Oversized power switch
2) 99% of output available as 12V (107 Amps :) )
3) fully modular (even 24pin connector).  makes swapping and moving rigs so much easier.
4) Very high quality fan.  Whisper quiet & cool even at 80% load.
5) High quality, thick power connectors.

Thank you DAT you settled my search/indecision.

As for my extreme load on my psu that's me just wanting to see how OCZ holds up to PCPC's 7 year warranty.

Sadly the only PSU to die on me in the last 10 years was my PC P&C 610 Silencer...

Good I might finally get/submit my first warranty. Got 2 crippled 5850's with a 1 of 3 vrm's blown that force me to be energy efficient. Ones on the rig I posted.

I would rather see more complete stats, some have done more than the OP requirements, which is appreciated.  :)

MB model: GA 890FXA-UD7
Psu 1:OCZ ZT-750w (not here yet)
Psu 2:

Card 1: Sapphire 5850 Extreme 1GB
BIOS Flashed/Stock:
Slot x1/x4/x8/x16: x16
Extender/Powered: none
Clock: 725MHz
Memory: Stock
Voltage: Stock
Temp: 0
Watercooled: No
MHash/s: 0 MH/s


Total Hash Rate: 0
Watts @ Wall: 0
MHash/W: 0
Ambient Temp: 21.6C
OS: Linux
Miner:
Miner Options:
SDK:
Cat:

;D

I have picked up one of these for $7.99, will it suffice?

http://www.harborfreight.com/media/catalog/product/cache/1/image/370x/9df78eab33525d08d6e5fb8d27136e95/i/m/image_13483.jpg

$7.99?  Likely not.  Even $79.99 is unlikely.  Of course if it normally is $200 and you found a 90% off sale then maybe. :)
I couldn't see the model otherwise I could tell you.

Generally clamp meters fall into one of four categories:
AC only averaging - $ (nearly worthless for mining work).
AC/DC averaging only - $$  (DC is will be accurate but AC averaging meters are nearly worthless for measuring PSU loads).
AC/DC True RMS - $$$ (Will give accurate amps reading on AC side also but doesn't account for power factor).
Power Meter - $$$$ (Measures current & voltage simultaneously for accurate wattage including power factor).

Are you (or anyone else here) familiar with multimeters? If so could you recommend any? Best if it could be used for AC and DC but currently the most important for me is to measure my 12V lanes' voltages. I found one at home that can be purchased for as little as $1-3 so I bet the actual voltage may vary by 5-10% (or more).

Depends on your budget.  Voltage is much easier to measure accurately than current and DC is always easier than AC.

I have always had good luck with KleinTools:
http://www.amazon.com/Klein-MM100-Manual-Ranging-Multimeter/dp/B003JJCGF6/ref=sr_1_1?ie=UTF8&qid=1329152604&sr=8-1

Klein says measurements are +/- 0.5% and I don't have any reason to doubt them.

Buy the $3 one for general purpose around the house jobs or auto. Buy the $100-200 one for serious tolerant requirement testing.
When someone asks to borrow your DMM, instead of just saying no, you lend them the $3 one. Your accurate expensive tool stays safe.