Furmark 480/5870 power figures may be invalid

[Whoisthisreally]

Many of the reviews including Kyle's use Furmark to compare the load power consumption between these cards.

This comparison may not be valid because AMD uses Catalyst and hardware detection in order to throttle against Furmark.

http://www.geeks3d.com/20090925/ati...on-against-power-virus-like-furmark-and-occt/

 

[SicKlown42012]

It only throttles if you get into the danger zone. Mainly, this is for people who overclock and voervolt. At stock clocks, it doesn't kick in. If it did, you would see the power consumption numbers drop off drastically. Since they stayed consistent, nothing's going on.

I'm also betting that Nvidia has something similar in order to protect their cards. As hot and power hungry as they are in normal use, they'd be idiots not to.

 

[YeuEmMaiMai]

welcome to last year man.......

 

[mafa]

there are also many were not using furmark and still got similar results. nv 480 uses more than 100+ watts than 5870.

 

[ZerazaX]

TomsHardware, bit-tech, Anandtech, and other websites off the top of my head all used more than just Furmark - and in all instances, the 5870 drew 100W less than the 480

 

[Whoisthisreally]

It only throttles if you get into the danger zone. Mainly, this is for people who overclock and voervolt. At stock clocks, it doesn't kick in.

Can you confirm this? based on the articles, it throttles automatically when detecting the executable name.

there are also many were not using furmark and still got similar results. nv 480 uses more than 100+ watts than 5870.

I know this, and I wouldn't expect the figures to change drastically were Furmark a valid tool, but the question still remains: Are the reported figures valid?

 

[SicKlown42012]

Can you confirm this? based on the articles, it throttles automatically when detecting the executable name.

If the chip detects a dangerous situation (overload), the chip will immediately throttle back the card by one PowerPlay level.

Seems pretty self-explanatory.

 

[ZerazaX]

Can you confirm this? based on the articles, it throttles automatically when detecting the executable name.

I believe that was intended for the 48xx series. The 58xx's use hardware to detect overloaded PWMs

I know this, and I wouldn't expect the figures to change drastically were Furmark a valid tool, but the question still remains: Are the reported figures valid?

If you look at the 5870 CF numbers and subtract it from the 5870 single GPU on Furmark, you will see that Furmark does draw more than the rated TDP for the 5870, so they go over the limit.

In fact, in non Furmark situations, the 5970 drew 40W less than the GTX 480. In Furmark, the 5970 drew single digits less W than the 480, meaning it actually shot up due to furmark.

 

[kllrnohj]

Can you confirm this? based on the articles, it throttles automatically when detecting the executable name.

That was for the 48xx series. The 5xxx series has hardware protection against overvoltage that is not based on executable name. As per the article you linked, anyway. In a normal, stock clocked situation, an overvolt isn't likely to occur anyway.

I know this, and I wouldn't expect the figures to change drastically were Furmark a valid tool, but the question still remains: Are the reported figures valid?

Yes. IF this was kicking in then you would see the clock speeds of the cards bounce up and down. The card drops a powerplay level when it detects an overvolt, which backs down on clock speeds. Clock speeds are easily measured and recorded over time, so if this was happening reviewers would very easily notice it.

 

[Whoisthisreally]

Okay, so if executable detection is restricted only to 4xxx cards, and throttling is regulated purely by hardware monitoring, there probably is no problem.