Power efficiency: post up your most efficient rigs!

[Biker]

Hi all

In my crazed drive for the most power efficient machines vs max ppd I have been a tweakin' and have a couple of rigs up for ppd/w inspection:

Rig 1:

AMD X6 1055T @ 3.8ghz 1.375v

Gigabyte GA890GPA-UD3H

2GB G.skill Trident @ 1.5v (1 stick)

on board gfx (low clocked)

Enermax Modu87+ 600W Gold rated PSU

60GB 5400rpm HDD

Total average WCG PPD (over last 10 days): 21771

Average power consumption (over last 10 days) 161w

PPD per watt score (21771 / 161) = 135 PPD/w



Rig 2:

Intel i7 920 @ 3.8ghz 1.18v LLC on

Gigabyte EX58-UD3

4GB G.skill ECO 7-8-7-24 @ 1.4v (2 sticks)

Nvidia GT220 (stock clocks)

Seasonic X-650 Gold rated PSU

Intel X25-V 40GB SSD

Total average WCG PPD (over last 10 days): 34221

Average power consumption (over last 10 days) 185w

PPD per watt score (34221 / 185) = 185 PPD/w



I also ran rig 2 with a cheap ass PSU for 10 days and it averaged 245w for the same PPD which dropped the PPD/w figure to 140 PPD/w - just goes to show how important a high efficiency PSU is in saving some $



For UK users you can then work out how much you are going to get bent over a till by your local friendly power provider here


Trust me... the results will make you



Also there is a great site here which is stuffed full of all kinds of energy usage reducion tips and covers everything not just our humble folding farms. In addition and almost all of the information applys globally



Just thought I would post up my findings in case the results piqued your interest

[fallwind]

IIRC my Gulftown crunchers at 4.2GHz were drawing 240w at the wall and make 56,000 ppd, so thats 233 PPD/w. These are barebones systems though, just a board, 3x1GB RAM underclocked and undervolted, no video, 1 HD, 1 fan and a 400w FSP PSU, not sure of the efficiency on those. I'm waiting for a good sale on a 80+ Gold PSU to see if it will make a reasonable difference.

[jcool]

My most efficient rig's gotta be the Dual X5650. 72k PPD at 300W from the wall equals 240 PPW. That's with a slightly oversized 650W 80+ Bronze TT Toughpower unit
Tyan S7002, 6x2GB DDR3, a couple of hard drives and loads of fans too. Should be possible to get this thing below the 300W mark...

[Duh]

@jcool: then your Dual X5650 has the same PPW as the 980x ( underclocked and undervolted), right?

[jcool]

Actually it should be a little better than two 980X's since you essentially run one less motherboard. Even though two Gulftowns will produce a lot more PPD when overclocked

[WhiteFireDragon]

builds centered around sr-2 or 980X will win the efficiency game, assuming they're only moderately clocked. as for budget conscious folks like myself, here are my two most efficient rigs:

- phenomII x6 1055T @ 3.7ghz 1.37v
- biostar biostar TA790GXA3+
- onboard video
- 1 HD, 3 fans
- OCZ z-series 850w gold
- 181w
- 22.5k/day, 124PPD/w


- i7-920 @ 3.8ghz 1.17v
- evga X58 LE
- 7200gs
- 2 fans, 40gb SSD
- seasonic x650 gold
- 185w
- PPD?? let my friend use this computer, but he can't get an internet connection on it

Biker, your i7-920 @ 3.8ghz gets 35k/day?? these numbers seem a little high don't they? our two rigs have very similar hardware and clocks

[Biker]

Biker, your i7-920 @ 3.8ghz gets 35k/day?? these numbers seem a little high don't they? our two rigs have very similar hardware and clocks

Indeed they are very similar

The power usage is identical!

Maybe it is due to what I am kerunching and the os?

For the last 10 days I have been concentrating on CMD and FAAH (using XP x64 and the 64 bit client)...

Edit: the 35k PPD was a near estimate

The exact figure for rig 2's last 10 day average PPD is 34221 which gives 185 PPD/w

Post #1 edited... calculating exact figure for rig 1 now...

Edit #2:

The exact figure for rig 1's last 10 day average PPD is 21771 which gives 135 PPD/w...

Post #1 updated.

[jcool]

builds centered around sr-2 ... will win the efficiency game

Well.. not really, no. The SR-2 is a power hog due to its high-power design and useless Dual NF200 chips
To do 100k/d on a SR-2 you see an estimated power draw of around 550W, giving the SR-2 the same PPW ratio as a single 920, way behind a single Gulftown or stock Dual Westmere on a Tyan or Supermicro board.
Of course, that didn't stop me from getting one

Edit: What are you guys crunching on those PII X6 chips?
I was going to get me one of those, but seeing your numbers doesn't do anything to further that choice.. I would have thought they'd at least be on par with a 920?!

[Biker]

Edit: What are you guys crunching on those PII X6 chips?

Exactly the same as my 920's :/

Since I actually paid less for a 920 setup I agree that in terms of PPD / £ it is not great :

If you look at just the board and CPU (and GFX card on the 920):

X58 board + 920 + 8400GS = £241

890G board + 1055T = £259

X58 = 34221 / £241 = 142 PPD / £

X6 = 21771 / £259 = 84 PPD /£

The X6 is not a great return by this measure

[Rob_B]

Intel Core I3 540 @ 3780MHz 1.15 v.
MSI H55M-ED55
GF6200
Kingston HyperX LoVo 4 GB, PC3-12800, 1600 MHz 1.3v
Seasonic S12II-380 380 Watt
And an old 7200rpm sata drive.

It uses 95 watt running WCG.
Doing 15800 PPD = 166 PPD/w

Maybe it can do a little better with the onchip GFX and a better HD.
But I want to watercool it too, so pump will be added to this.

[BlindFreddie]

What mixes of WCG projects are you guys crunching? I think this makes a difference, and relative results depend on the CPU type, eg Yorkfields do well on Autodock (FAAH, DDDT-1, HFCC), on which they underclaim.
They don't score so well on HCMD2.

The large cache (3MB/core) on the Yorkies is the probable reason: On my A-64 X2 (512kb/core), FAAH has the lowest cache success rate (~87%), while HCMD2 gets almost 100%, and scores higher PPD (and terrible PPD/W whatever you crunch).

[Biker]

What mixes of WCG projects are you guys crunching? I think this makes a difference, and relative results depend on the CPU type, eg Yorkfields do well on Autodock (FAAH, DDDT-1, HFCC), on which they underclaim.
They don't score so well on HCMD2.

The large cache (3MB/core) on the Yorkies is the probable reason: On my A-64 X2 (512kb/core), FAAH has the lowest cache success rate (~87%), while HCMD2 gets almost 100%, and scores higher PPD (and terrible PPD/w whatever you crunch).

Oh yes definitely...

At the moment I am strictly FAAH which only seems to have a negligible power usage difference to HCMD2 (on my X6 and 920 machines). Having said that I have not turned my attention to my 3 775 quads yet so I have no power usage data for them at the moment....

[Jaco]

I think HCC consumes a little more , can't prove it though.

[slaveondope]

Grrr AMD fail
Athlon 620 @ 3.25ghz 13,500ppd 140w=96p/w
955c3 @ 3.9ghz 16,000ppd 170w=94p/w
1090t @ 4ghz 25,000ppd 320w=78p/w (280w without 2x5850's)=89p/w

EDIT:
I think my Killawatt meter is off

[Biker]

I think my Killawatt meter is off

I dont think so..... those X6's pull some serious power at 3.8 ghz+

Quote:

We had a starting clock of 3.2 GHz, then, for a vcore of 1.25v due to Vdrop, with power consumption measured at the ATX12V as 116W (we went up to 126W limiting Vdrop to get 1.296V). Firstly we tried to stabilise this clock at a lower voltage and managed to do so at 1.18V, taking us down to a power consumption of just 95W! A very impressive result for the Phenom II X6! We went up to 1.22V to stabilise it at 3.3 and 3.4 GHz (106 and 108W), 1.28V at 3.5 and 3.6 GHZ (126 and 127W), 1.33V at 3.7 GHz (145W), 1.38V at 3.8 GHz (166W), 1.44V at 3.9 GHz (188W) and 1.49V at 4 GHz (224W).

Source:

http://www.behardware.com/articles/7...90t-1055t.html

If you are pushing over 1.5v then 320w is about right, depending upon other hardware in the system...

[slaveondope]

Im running all the hardware in sig, which is alot and several fans on the 1090t so I guess its right.
I know it was 10 more watts for 200mhz more on the Athlon and 10 more for 300mhz more on the 955.
I just cant bare the idea of not getting every ounce out of each
The machince's haunt me in my sleep if each one isnt as high as it can go

[hedge]

I just cant bare the idea of not getting every ounce out of each
The machine's haunt me in my sleep if each one isnt as high as it can go

That is the problem I had until I started looking at my power bill. Now they haunt me if they aren't running the most efficient that they can.

[Biker]

That is the problem I had until I started looking at my power bill. Now they haunt me if they aren't running the most efficient that they can.

Presactly

[Rob_B]

It uses 95 watt running WCG.
Doing 15800 PPD = 166 PPD/w

Maybe it can do a little better with the onchip GFX

I have removed the GF6200 and use the onchip GFX now.
Power consumption dropped to 86 watt
I will check if it keep doing the same PPD. If it does that means it is doing 183 PPD/w.

[Rob_B]

I think HCC consumes a little more , can't prove it though.

I'm running HCC on my cruncher only at the moment. One thing I noticed was that on Linux it used about 4 watt more than on Win7. (both 64 bit) I think that is more a driver thing than some thing else.

[slaveondope]

That is the problem I had until I started looking at my power bill. Now they haunt me if they aren't running the most efficient that they can.

But $5 off the monthly bill for 10,000ppd less
Im sure after a couple more months I may rethink things just going thru the stages of addiction

[hedge]

But $5 off the monthly bill for 10,000ppd less
Im sure after a couple more months I may rethink things just going thru the stages of addiction

On top of that the room doesn't feel like an oven with the computers pumping out maximum heat. (less AC, less complaining from room mates, and less of me having to get out of my seat to go get a drink)

[Biker]

Yayy!

Just junked 4 inefficient PSU's and 2 GPU's and my total crunching power usage is down 163w with NO loss of PPD

In money terms that is a saving of £11.50 a month...

At this rate I might just get it to under £150 a month



ps.

Seasonic X-650 FTW