Debunking Power Supply Myths

[burebista]

@Anandtech (http://www.anandtech.com/casecoolingpsus/showdoc.aspx?i=3413&p=1).
Our three sample systems show that few if any people actually need anything more than 850W power supply, especially if you can get an appropriate number of connectors. About the only reason you should consider anything more is if you want to keep your power supply at a lower load, since optimal efficiency and lower noise levels usually come at around half the rated output. Overclockers might also want a bit of extra headroom. Otherwise, there's really no need for a 1000W power supplies, and certainly not those with even higher ratings.
Looks fair to me.

[HOOfan_1]

@Anandtech (http://www.anandtech.com/casecoolingpsus/showdoc.aspx?i=3413&p=1).

Looks fair to me.

Of course it goes against his colleagues' (Anand and Derek Wilson) preposterous claims that a 1200W PSU is needed to run 4850CF and GTX 280 SLi

[Smirnoff]

Nice to see somebody else took the time to measure the individual components. They seem to get some higher figures than Xbitlabs, so I guess Furmark loads the GPU more than the 3DMark loop that Xbit usually run.

[Travis]

HD2900XT only a bit above 100W? Lower than a 8800 Ultra? WTF?

Anyway though some numbers are rather questionable I'll say the discussion is worth reading.

[Spectre]

I wonder how the power consumption was measured for each component.

[Travis]

ammeter, I guess.

[Smirnoff]

Yeah, the 2900XT had me puzzled too. If that were true, it could have done with one 6-pin PCIE connector. So I guess they're not spot on either. XBitlabs measured it at 160W or some I think. To be taken with a grain of salt I guess.

[SKYMTL]

ammeter, I guess.

You can't attach one to the PCI-E slot and the writer cannot assume that higher-end cards will draw the max possible power from the PCI-E slot before drawing from the PEG connectors. GPUs do not draw power from the PCI-E slot in a linear fashion so monitoring this way is inacurate at best.

Some companies measure power consumption using a special PCB outfitted with a single PCI-E slot which is then hooked up to a power reader. The CPU and ram are seperated from this setup. Any additional PEG connectors are monitored with a ammeter. IMO, this is the only way to get somewhat accurate power consumption numbers for a GPU.

[Travis]

You can't attach one to the PCI-E slot and the writer cannot assume that higher-end cards will draw the max possible power from the PCI-E slot before drawing from the PEG connectors. GPUs do not draw power from the PCI-E slot in a linear fashion so monitoring this way is inacurate at best.

Some companies measure power consumption using a special PCB outfitted with a single PCI-E slot which is then hooked up to a power reader. The CPU and ram are seperated from this setup. Any additional PEG connectors are monitored with a ammeter. IMO, this is the only way to get somewhat accurate power consumption numbers for a GPU.

I like that PCB :D

Back to the topic, there are still a few ways to get the amperage numbers.
For example, you can measure the 12V current on the 24pin ATX connector. It mainly goes to the GFX so accuracy is quite good.
Another trick is to cover the 12V pins on the PEG connector with insulating tape. I'm not sure if it's reliable, though.

[Travis]

Yeah, the 2900XT had me puzzled too. If that were true, it could have done with one 6-pin PCIE connector. So I guess they're not spot on either. XBitlabs measured it at 160W or some I think. To be taken with a grain of salt I guess.

look at the 160W 6000+ and 86W QX9770 and 19W idle Q9550...
things got really weird here.

[D-Juice]

HD2900XT only a bit above 100W? Lower than a 8800 Ultra? WTF?

Anyway though some numbers are rather questionable I'll say the discussion is worth reading.

:confused:

Smells like the bottleneck was somewhere else......

Most sites pay little attention to the need to remove bottlenecks in order to unlock power consumption.

[Spectre]

I like that PCB :D

Back to the topic, there are still a few ways to get the amperage numbers.
For example, you can measure the 12V current on the 24pin ATX connector. It mainly goes to the GFX so accuracy is quite good.

Or not.

[Travis]

Or not.

Does chipset use 12V or 5V voltage?

[theELVISCERATOR]

Until they run orthos at 100 percent and Furmark they arent loading the system fully.

[D-Juice]

Until they run orthos at 100 percent and Furmark they arent loading the system fully.

I may be wrong but IIRC i read somewhere that if you load your CPU 100% with orthos, you wont have the necessary CPU cycles available to fully load the video card(s).

And thats why this method resulted in a lower total system load than even crysis during some power consumption tests around here.

Can anyone clarify this ?