Power Consumption of Contemporary Graphics Accelerators (xbitlabs)

[Schmide]

Power Consumption of Contemporary Graphics Accelerators

Most of us knew the ball park figures, now they're a bit more exact.

Very neat to see the 5850 bleed at 1ghz.

Kudos

 

[CurseTheSky]

At stock clocks, the HD 5850 and HD 5770 in particular impress me. So much performance for so little power.

 

[cbn]

Great article. Thanks

 

[frostedflakes]

Xbit Labs is fantastic for video card power measurements, been reading their articles mainly for this reason for years. Keep in mind that they measure actual consumption of the graphics card (they have a really slick setup that allows them to measure power through the PCI-e slot, as well as the external power connectors), not system power like most sites.

Anyway, good article from them, nice to have all the measurements in one place. Definitely worth bookmarking.

 

[Borealis7]

will definitely use this as a reference in the future. but isnt this article a week too early?

lets hope they do a separate article for Fermi!

but knowing XBitLabs, it'll only be next year...

 

[MarkLuvsCS]

The only question I have about the article: Why does the ATI cards have volts ranging from 1.088 to 1.35 and the nvidia ranging 1.17 to 1.18?? I understand its because ATI allowed software to adjust voltages and the nvidia cards only allowed through hardware modding.

Just seems silly showing the extreme end of ATIs power consumption to the minimal consumption on nvidias end. I doubt many would want to crank near that much juice through their video card regularly. Although if the sole purpose was to see how much voltage they could push through the cards while still running stable I guess at least we know the new 5xxx series should handle the volts just fine. I don't see the reason they are trying to draw any conclusions from their test other than how much power is being consumed by each card given it's voltage and frequency.

"... The power consumption difference between the minimum and maximum core frequencies was only 32.4 W, which is nothing compared with 148.8 W we have just seen by ATI Radeon HD 5850...."

The power consumption isn't greatly different because they didn't touch how much voltage was going through nvidia's card?

 

[Ben90]

The only question I have about the article: Why does the ATI cards have volts ranging from 1.088 to 1.35 and the nvidia ranging 1.17 to 1.18?? I understand its because ATI allowed software to adjust voltages and the nvidia cards only allowed through hardware modding.

Just seems silly showing the extreme end of ATIs power consumption to the minimal consumption on nvidias end. I doubt many would want to crank near that much juice through their video card regularly. Although if the sole purpose was to see how much voltage they could push through the cards while still running stable I guess at least we know the new 5xxx series should handle the volts just fine. I don't see the reason they are trying to draw any conclusions from their test other than how much power is being consumed by each card given it's voltage and frequency.

"... The power consumption difference between the minimum and maximum core frequencies was only 32.4 W, which is nothing compared with 148.8 W we have just seen by ATI Radeon HD 5850...."

The power consumption isn't greatly different because they didn't touch how much voltage was going through nvidia's card?

You answered your own question with: "I understand its because ATI allowed software to adjust voltages and the nvidia cards only allowed through hardware modding."

Very few people volt mod cards. However a good number of people don't mind cranking it up via software.

Your question presents itself that you are interested in how power scales with voltage between Red/Green, but that is not what the article is about. Its about real world power consumption between current GPUs.

 

[nyker96]

the HD 5850 is impressive, it's only about 40W more during crisis than my hd 4850 yet is so much faster.

 

[crazylegs]

Very nice link

The results for 925MHz @ 1.14V are pretty nice, impressive OC from stock for little added power consumption

Seeing as it has 1.35V pumped through it in order to reach an OC of 1000MHz, goes to show the diminishing returns that are experienced when gunning for the absolute maximum possible OC.

Cut back ~5-10% and reap the rewards of (free) added performance coupled with minimal extra power consumption (Y)

 

[MrK6]

Very nice link

The results for 925MHz @ 1.14V are pretty nice, impressive OC from stock for little added power consumption

Seeing as it has 1.35V pumped through it in order to reach an OC of 1000MHz, goes to show the diminishing returns that are experienced when gunning for the absolute maximum possible OC.

Cut back ~5-10% and reap the rewards of (free) added performance coupled with minimal extra power consumption (Y)

I was just going to say the same thing. I think I arrived at the same conclusion through my own testing (my 5850's are currently at 900/1250 with 1.125V), but there's definitely a "sweet spot" where you get great performance increases with a minimal addition of power.

Great article, thanks for posting!

 

[dguy6789]

GTX 295 uses more power than a 5970 at full load. That's surprising.

 

[Termie]

I was just going to say the same thing. I think I arrived at the same conclusion through my own testing (my 5850's are currently at 900/1250 with 1.125V), but there's definitely a "sweet spot" where you get great performance increases with a minimal addition of power.

Great article, thanks for posting!

Indeed...the 5850 allows for some significant gains without major power use. I run mine at 850/1200 on stock volts, and get about 15% more FPS than a 5850@stock for only about 25-30 more watts at load according to my Kill-a-watt, which corresponds with what Xbitlabs showed (although they didn't run that exact combination, they showed that a ram speed increase to 1200 cost 10 watts and that the GPU speed increase to 850 was worth 20 watts). I'll "pay" 30 watts for a 15% increase in speed. Still so much lower than a 275/285 or 5870, and it's within 5% of 5870 performance.

Just one amazing fact to consider: the 5850@stock uses 125 watts less than a 285 for typically better performance. I call that a major design win! And it idles at 20 watts less than my old 8800gt.

 

[yh125d]

Wow, astronomical numbers on the 5850


And I am disappoint at my 4890. 190w in games at stock speed

though I didn't expect it to be that much lower than GTX275/85 considering their relative performance

 

[MrK6]

Wow, astronomical numbers on the 5850

And I am disappoint at my 4890. 190w in games at stock speed

though I didn't expect it to be that much lower than GTX275/85 considering their relative performance

FWIW, the 4890 idle power consumption is very easy to fix using RBE.

 

[PingSpike]

I have always really liked these power use reviews by x-bit labs. I've used the older ones to find low power use cards for servers, or secondary uses.

 

[yh125d]

Meh, I don't care too much about idle, but ~200W during gaming (I'm OC'd) is alot. Maybe it'll just motivate me to buy a 5870

 

[blanketyblank]

Dual GPU numbers are kind of strange to me. For instance why is power consumption on the 5970 nearly 3 times the and not 2 times the 5870 in desktop? Load makes sense though since it's about twice as much as a 5850.
Kind of makes me wonder if you couldn't actually SAVE power by running a SLI/ crossfire setup. If the power is just doubled then 2 x 5850 or 5870 would idle lower than 5970. Even at load power looks about the same or a little less for comparable performance.
Or 2 x 5770 = 5870 performance, but would use about 155 vs 170 at load in crysis. Idle would be higher here though 28 vs 16.

 

[Headfoot]

Why would they EVER do a review like this a week before a brand new architecture is supposed to come out from one of the companies. That coupled with their excessive praising of the ATI cards leads me to believe this article was purposefully timed right before Fermi so they could get their fanboy jollies out of it. Truth be told though, Fermi will undoubtedly use a ton of power so I guess its already somewhat known.

 

[SHAQ]

GTX 295 uses more power than a 5970 at full load. That's surprising.

How can that be? It isn't just a little more either...45-70 watts. I don't buy it.

 

[darkswordsman17]

Did they only test peak?

 

[NoQuarter]

Why would they EVER do a review like this a week before a brand new architecture is supposed to come out from one of the companies. That coupled with their excessive praising of the ATI cards leads me to believe this article was purposefully timed right before Fermi so they could get their fanboy jollies out of it. Truth be told though, Fermi will undoubtedly use a ton of power so I guess its already somewhat known.

Uhm, you must not read the site, they way overrated the GT240. Has Nvidia released anything praise worthy since 5870 came out? Not really.. that's why it feels like ATI has gotten so much attention for the last 6 months.

 

[adairusmc]

Did they only test peak?

You could always read the article and find out.