jonnyGURU Forums
Home Site Search Reviews Articles Contest Links PSU FAQs  


Go Back   jonnyGURU Forums > Computer Hardware > PC Power Supply Discussion

PC Power Supply Discussion Troubleshooting and discussion of computer power supplies

Reply
 
Thread Tools Display Modes
  #1  
Old 10-16-2007
jonnyGURU's Avatar
jonnyGURU jonnyGURU is offline
Site Founder
 
Join Date: Oct 2006
Location: jonnyGURU forums, of course!
Posts: 14,891
Thanks: 475
Thanked 1,870 Times in 1,154 Posts
Default SLI Power Consumption Results

Ran a few tests. Thought I'd share. Not going into much detail because I don't want to get into an Anandtechesque pissing contest.

The following numbers are all DC, not AC, so there is no need to factor in PSU efficiency.

The numbers may be high because these are accumulative results. Current was measured at multiple points in the system and were then added together to provide the numbers below.

On the other hand, the response time of the test equipment is > 40mS, so transient loads that may hit the PSU faster than 40mS would not be picked up by the equipment.

Tests were actually performed with an 800W (unless stated otherwise) with a single +12V rail, but I've broken the load down into two rails to help illustrate the power distribution. The "CPU rail" is labeled +12V2 in my tests.

Build one:

680i board w/ QX6800, 2x1GB PC2 8500, two Seagate SATA drives in RAID0, one optical, one floppy, three fans run at full RPM, no PCI-e cards, on board sound, NIC, etc. USB KBD and mouse. Room temperature is just around 23.5C.

Build two:

Same parts with dual FX72's on a 680a based LIN64-SLI board. RAM upgraded to four 1GB PC2 8500 due to the nature of the platform.

Let the fun begin....

6800 GT 256MB x 2 SLI bridge on build 1
Total: 410W DC.
Total +12V2: 6.77A
Total +12V1: 21.7A

Comments:

Clearly this doesn't need much power, and will presumably work fine even if dual +12V rails (at no less than 18A each) are implemented. Since a 400W to 450W PSU does not typically have 18A and higher +12V rails, it is suggested a single +12V rail unit is used (i.e. Corsair VX450W.)

6800 GT 256MB x 2 SLI bridge on build 2
Total: 556W DC.
Total +12V2: 13A
Total +12V1: 25.7A

Comments:

Certainly we have a GPU bottle neck here. The +12V2 consumption goes up because of two CPU's, but also the +12V1 and total power consumption goes up because of the dual Northbridge design (a 680a is essentially two 590 chips.) I would say you can certainly "get by" with a 550W or 600W with dual +12V rails (20A each, I would hope) but would rather recommend a 600W with either a single +12V rail or four +12V rails or greater (like a SilverStone Strider ST60F or Corsair HX620W.)

7950 GT 256MB x 2 SLI bridge on build 1
Total: 444W DC.
Total +12V2: 8.2A
Total +12V1: 21.6A

Comments:

Power consumption here is very similar to the 6800 cards. Apparently some CPU bottlenecking is freed up, but power consumption isn't great enough for me to recommend anything different.

7950 GT 256MB x 2 SLI bridge on build 2
Total: 602W DC.
Total +12V2: 13A
Total +12V1: 29.4A

Comments:

Power consumption here is very similar to that of the 6800 cards. Again, it seems as if some CPU bottlenecking is freed up. I'm definitely suggesting that single +12V rail 600W and up here. If a single +12V rail solution isn't availble, feel free to use a 700W and up with four +12V rails (like the Etasis 750W/SilverStone Zeus ST75ZF.)

8800 GTS 640MB x 2 SLI bridge on build 1
Total: 484W DC.
Total +12V2: 8A
Total +12V1: 24.6A

Comments:

Not much different going on here still. The current going through the 8-pin (which may be supplying current to the CPU, but also to other parts of the board) went down a little, but the rest of the +12V consumption went up a little. A single +12V rail 500W and up (like the Corsair HX520W or Ultra X-Pro/X3 600W) or 600W and up with four +12V rails or more (like the SilverStone Strider ST60F) would be a good candidate here.

8800 GTS 640MB x 2 SLI bridge on build 2
Total: 656W DC.
Total +12V2: 13.7A
Total +12V1: 33.4A

Comments:

Power consumption here has gone up across the board. I'm going to recommend a minimum 700W here (like the Seasonic M12 700W, PC Power & Cooling 750W or the Etasis 750W/SilverStone Zeus ST75ZF.)

8800 GTX 768MB x 2 SLI bridge on build 1
Total: 527W DC.
Total +12V2: 7.2A
Total +12V1: 30.2A

Comments:

It's interesting to see the power consumption via the 8-pin EPS12V connector gradually go down with each set of cards. Going from the 6800 to 7950, we apparently unleashed a GPU bottleneck. But now it seems as if we're seeing a CPU bottleneck. Once again, a decent 600W, as long as it doesn't have two +12V rails, is going to be adequate.

8800 GTX 768MB x 2 SLI bridge on build 2
Total: 715W DC.
Total +12V2: 14.3A
Total +12V1: 41.1A

Comments:

Although I'm using an 800W PSU right now, I will say that I've tried to power this exact config with less than a 700W and it failed to even boot into the OS (using Corsair HX620W, Ultra X-Pro 600W and Thermaltake Toughpower 600W.) A minimum 700W is suggested, and a single +12V rail at the (Seasonic M12 700W or PC Power & Cooling Silencer 750W.) If only PSU's with split +12V rails are available, I suggest an 800W or up (like the Etasis 850W/SilverStone Zeus ST85ZF.)

8800 Ultra 768MB x 2 SLI bridge on build 1
Total: 636W DC.
Total +12V2: 10.3A
Total +12V1: 33A

Comments: We finally out grow the 600W suggestion and now move into the realm of 700W and up. 750W and up for PSU's with split +12V rails. The +12V2 consumption jumps way up. Obviously more juice is being drawn through this connector than the CPU can actually use, so I imagine that the resistance from the +12V lead on the main 24-pin connector is causing the board to pull power from wherever possible.

8800 Ultra 768MB x 2 SLI bridge on build 2
Total: 859W DC.
Total +12V2: 17.8A
Total +12V1: 45.6A

Comments: Remember when I said that I was using an 800W unless stated otherwise? Well.. here is my otherwise. During testing, I'd see the AC meter hit 1000W AC and.... THUD! PC would shut off or lock up. Even when the 800W was swapped out with a 1000W with four +12V rails I would shut down when power consumption hit 800W DC. Got to make sure that if you're using a pair of Ultra's on a quad FX rig that your PSU is at least 1000W with a single +12V rail (Ultra X3 1000W, PC Power & Cooling Turbo Cool 1000W) or has five or six +12V rails if they do split them up (yes, this exists. Topower built 1000W and up units often have five or six +12V rails.) If you're going with a PSU with four +12V rails, do check the power distribution. Make sure the OCP is at least 20A per rail, NOT 18A, and that there are no more than two PCI-e connectors on each of their +12V rails. No sharing with the 24-pin or any peripheral connectors.
__________________
Rest in peace Mike Clements, aka "Yellowbeard"

Rest in peace Joerg Theissen, aka "GI Joe"

Last edited by jonnyGURU; 10-16-2007 at 05:03 PM.
Reply With Quote
  #2  
Old 10-16-2007
Spawne32's Avatar
Spawne32 Spawne32 is offline
1kW User
 
Join Date: Feb 2007
Location: New Joisey
Posts: 309
Thanks: 0
Thanked 0 Times in 0 Posts
Default

interesting results
Reply With Quote
  #3  
Old 10-16-2007
jonnyGURU's Avatar
jonnyGURU jonnyGURU is offline
Site Founder
 
Join Date: Oct 2006
Location: jonnyGURU forums, of course!
Posts: 14,891
Thanks: 475
Thanked 1,870 Times in 1,154 Posts
Default

MYTHBUSTERS!

Why did I measure DC instead of measuring the AC and calaculating for efficiency? Because your Kill-A-Watt, Power Angel, or whatever tells you you're only pulling 150W from the wall may actually be WAY OFF. I would like to direct your attention to these videos:

http://www.jonnyguru.net/forums/show...45&postcount=8

and

http://www.jonnyguru.net/forums/show...9&postcount=15

I believe the problem is with the APFC circuit. Somehow the harmonics correction throws whatever these devices use to measure current. Can't say for sure. But whenever someone says "I only measured 200W AC power consumption from the wall so I only need a 300W for a pair of 8800GTS cards," I'm going to say, "Find a way to measure the DC output instead and get back with me because I'm not buying your numbers." Sorry. FYI: I used multiple data-logging clamping ammeters on every output of the PSU to get my numbers. Then added up the results of all of the meters.
__________________
Rest in peace Mike Clements, aka "Yellowbeard"

Rest in peace Joerg Theissen, aka "GI Joe"

Last edited by jonnyGURU; 10-16-2007 at 07:01 PM.
Reply With Quote
  #4  
Old 10-16-2007
danbfree danbfree is offline
500W User
 
Join Date: May 2007
Posts: 84
Thanks: 0
Thanked 0 Times in 0 Posts
Default

Wow... GREAT stuff! Perfect reference for those considering SLI-rigs and what to power them with. What about single card setups? Would it be safe to say it would be roughly 60% of the power (amps/watts) of the 12v+1 rail SLI results?
Reply With Quote
  #5  
Old 10-16-2007
SKYMTL's Avatar
SKYMTL SKYMTL is offline
Crazy Canuck
 
Join Date: Oct 2006
Location: Montreal, Canada
Posts: 985
Thanks: 0
Thanked 0 Times in 0 Posts
Default

Very interesting.

I have one question: is there ANY AC input gauge that is right?
Reply With Quote
  #6  
Old 10-16-2007
jonnyGURU's Avatar
jonnyGURU jonnyGURU is offline
Site Founder
 
Join Date: Oct 2006
Location: jonnyGURU forums, of course!
Posts: 14,891
Thanks: 475
Thanked 1,870 Times in 1,154 Posts
Default

Judging by my results? I'd say no, at least in some rare circumstances.

The good news is when they're off, they're way off. The way you accumulate data for multiple PSU's running the same rig across the board, you'd catch it because you would suddenly see a PSU that was reporting a power consumption much lower than the typically "efficient" units (i.e. Seasonic, Epsilon, etc.)
__________________
Rest in peace Mike Clements, aka "Yellowbeard"

Rest in peace Joerg Theissen, aka "GI Joe"
Reply With Quote
  #7  
Old 10-16-2007
SKYMTL's Avatar
SKYMTL SKYMTL is offline
Crazy Canuck
 
Join Date: Oct 2006
Location: Montreal, Canada
Posts: 985
Thanks: 0
Thanked 0 Times in 0 Posts
Default

Yeah...that is true. I will have to report it if I see it. I haven't run into any such circumstances other than the Apevia that went apeshit on me...and that reading was off because the PSU sucked so much....
Reply With Quote
  #8  
Old 10-16-2007
jonnyGURU's Avatar
jonnyGURU jonnyGURU is offline
Site Founder
 
Join Date: Oct 2006
Location: jonnyGURU forums, of course!
Posts: 14,891
Thanks: 475
Thanked 1,870 Times in 1,154 Posts
Default

Quote:
Originally Posted by danbfree View Post
Wow... GREAT stuff! Perfect reference for those considering SLI-rigs and what to power them with. What about single card setups? Would it be safe to say it would be roughly 60% of the power (amps/watts) of the 12v+1 rail SLI results?
Yeah... I didn't bother running one card because I thought that I could measure at the PCI-e and figure out what each card was using. Then I slapped myself in the head realizing that I wasn't taking into consideration how much each card was pulling from the slot. My +12V reading for the board was done on the +12V leads on the 24-pin. Thinking I could somehow math it out, I then looked at my PCI-e numbers and found that each card in SLI was using a different amount of juice than it's partner. Presumably, the card with the DVI connector on it consumed more power. I'll get deeper into the results as I have a chance to sort through them.
__________________
Rest in peace Mike Clements, aka "Yellowbeard"

Rest in peace Joerg Theissen, aka "GI Joe"
Reply With Quote
  #9  
Old 10-16-2007
SKYMTL's Avatar
SKYMTL SKYMTL is offline
Crazy Canuck
 
Join Date: Oct 2006
Location: Montreal, Canada
Posts: 985
Thanks: 0
Thanked 0 Times in 0 Posts
Default

Well, if you knew what the efficiency of the PSU is at each output level, you can narrow it down even more.
Reply With Quote
  #10  
Old 10-16-2007
jonnyGURU's Avatar
jonnyGURU jonnyGURU is offline
Site Founder
 
Join Date: Oct 2006
Location: jonnyGURU forums, of course!
Posts: 14,891
Thanks: 475
Thanked 1,870 Times in 1,154 Posts
Default

Right... IF you knew that. You would have to know that ahead of time. For that SevenTeam in the video, I had to ask SevenTeam for a print out from their own test equipment because I could not get a realistic number out of anything I had in the lab. Somehow, they got accurate AC input numbers (and efficiency translated to about 78 to 81%) so obviously there is some way to accurately measure AC input, but not in my posession.
__________________
Rest in peace Mike Clements, aka "Yellowbeard"

Rest in peace Joerg Theissen, aka "GI Joe"
Reply With Quote
Reply

Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump

Similar Threads
Thread Thread Starter Forum Replies Last Post
VGA Card Connectors & Power Consumption Eduardo GPU Discussion 3 06-23-2008 01:06 AM
power consumption of my x3360 (kill-a-watt) graysky General PC Hardware 5 05-19-2008 05:47 PM
Power consumption of a Raptor virtualrain PC Power Supply Discussion 1 01-10-2008 09:42 PM
Tom's: The Truth About Power Consumption SKYMTL PC Power Supply Discussion 10 10-21-2007 01:06 PM
Memory and power consumption Seatrout General PC Hardware 8 05-30-2007 07:19 AM


All times are GMT -4. The time now is 04:00 PM.


Powered by vBulletin® Version 3.8.7
Copyright ©2000 - 2018, vBulletin Solutions, Inc.
Copyright © JonnyGURU.com 2000 -