PSU requirements for some of current GPUs

[Neeyik]

With nothing better to do for 30 mins or so, I decided to go through some of X-bit Labs' graphics card reviews and compile the power consumption data. The reason why I picked XBL's numbers is that they are the only site (that I know of, at least) that record the actual power consumption of the graphics card during operation and not the whole system. Basically, they've wired shunts across the voltage supply pins in the PCI Express slot and the external connectors, thus allowing them to record the voltage drops across the shunts.

Manufacturers are rather unhelpful by just stating what power rating of the PSU is required but as we all know, one can have a PSU with a large W number attached to it but have pretty poor specifications on the 12V lines, which is where the bulk of the power is taken. Anyway, without further ado:

===
To determine the peak current draw from the 12 volt supply, use the following equation:

12 V current draw = peak power usage x 0.08

The exact amount will vary from system to system but it's a good enough formula for nearly everyone.













New for 2010
X-bit Labs now use a different testing method for measuring the power consumption of graphics cards - they still measure the current draw on the 3.3V and 12V lines (both in the PCI Express slot and the additional power cables, if required) but they now use Crysis Warhead at 1600x1200, FSAA 4x, DirectX 10/Enthusiast settings on the Frost map instead of 3DMark06.

===
Previous results using 3DMark06

GPU		Total W		3.3V Amps	12V Amps
5970		191		1.3		 15.5
5870		107		0.9		 8.7
5770		61		(1.8)		 (4.6)
4890		120		0.9		 9.8
4870 X2		264		1.3		21.6
4870		130		0.9		10.6
4850		110		0.8		 8.9
4830		 85		(0.9)		 (6.7)
4770		 50		0.5		 3.9
4670		 47		(0.5)		 (3.7)
4650		 30		0.7		 2.4
3870 X2		171				(13.9)
3870		 81		0.5		 6.6
3850		 63		0.5		 5.1
3650		 42		0.3		 3.4
2900 XT		161		1.6		13.0
2900 Pro	133		1.6		10.7
2600 XT		 49				(3.9)
2400 XT		 19				(1.5)
X1950 XTX	125		2.0		 9.9
X1950 Pro	 66				(5.2)
X1900 GT	 75				(5.9)
X1800 XT	103				(8.1)
X1800 XL	 57				(4.5)
X1800 GTO	 48				(3.8)

GPU		Total W		3.3V Amps	12V Amps
GTX 295		214		1.0		17.6
GTX 285		150		0.4		12.7
GTX 280		178		0.6		14.7
GTX 260-216	105		0.4		 8.6
GTX 260-192	136		0.6		11.2
GTS 250		 81		0.9		 6.5
GT 240		 81		0.9		 6.5
9800 GX2	182		0.3		15.1
9800 GTX	108		0.4		 8.9
9600 GT		 61		0.2		 5.0
8800 GTX	132		1.2		10.6
8800 GTS	 99				(7.8)
8800 GT		 78				(6.2)
8600 GTS	 47				(3.7)
8500 GT		 27				(2.1)
7950 GX2	110				(8.7)
7900 GTX	 84				(6.7)
7900 GT		 48				(3.8)
6800 GT		 58				(4.6)
6800		 39				(3.0)
6600 GT		 48				(3.8)

The figures in brackets are my own estimations, going from what patterns are displayed in the XBL readings. Roughly, it would seem that around 95% of the power is taken via the 12V pins and wires. The sources for all this lot are:

http://www.xbitlabs.com/articles/vid...0_5.html#sect0
http://www.xbitlabs.com/articles/vid...2_6.html#sect0
http://www.xbitlabs.com/articles/vid...0_4.html#sect0
http://www.xbitlabs.com/articles/vid..._14.html#sect0
http://www.xbitlabs.com/articles/vid...o_4.html#sect0
http://www.xbitlabs.com/articles/vid...4_5.html#sect0
http://www.xbitlabs.com/articles/vid...x_7.html#sect0
http://www.xbitlabs.com/articles/vid...t_5.html#sect0
http://www.xbitlabs.com/articles/vid..._12.html#sect0
http://www.xbitlabs.com/articles/vid...x_7.html#sect0
http://www.xbitlabs.com/articles/vid...i_4.html#sect0
http://www.xbitlabs.com/articles/vid...p_6.html#sect0
http://xbitlabs.com/articles/video/d...0_6.html#sect0
http://www.xbitlabs.com/articles/vid...m_5.html#sect0
http://www.xbitlabs.com/articles/vid...0_4.html#sect0
http://www.xbitlabs.com/articles/vid...s_6.html#sect0
http://www.xbitlabs.com/articles/vid...0_5.html#sect0
http://www.xbitlabs.com/articles/vid...0_7.html#sect0
http://www.www.xbitlabs.com/articles...0_5.html#sect0

X-bit Labs' Testing Method

Following our standard procedure, the 3D load was created by the first SM3.0/HDR test from 3DMark06 running in a loop at 1600x1200 with forced 4x FSAA and 16x AF.

*** It's worth noting that whilst this method is good enough to represent a typical load that game will play on a graphics card, it doesn't demonstate an absolute peak power consumption scenario. For that, one needs to use something like FurMark but this program is designed to do this, whereas games aren't. Running with FurMark can add anything up to 50% more peak power consumption, depending on the GPU architecture.

Feel free to add your own numbers to these thread, if you can find them in X-bit Labs, but do note that total system figures are useless here.

[Thresher]

Sticky! [thumbup]

[Ninja7]

Whats the reason for the HD3870 consuming slightly less (or the same) than the HD3850 ? I would have thought that the lower power consumption of the GDDR4 would have not been enough to offset the extra power needed by the higher voltage and higher clocked core.

[LuckyNV]

It might be a typo, should be 63W

[Ninja7]

It might be a typo, should be 63W

Yeah it is, i should have looked at the 12V and 3.3V consumption first.

[Neeyik]

Thanks for spotting the typo chaps. For those people wondering how much difference RAM makes, 512MiB of GDDR3 uses around 3~5W so you can easily shift the numbers around accordingly.

[cyberodin]

Sticky! [thumbup]

another vote

[Noizic]

Sorry if I'm blind but I didn't noticed my 8800GT or 9800GT?

[Neeyik]

8800 GT now added.

[guezz]

Sticky! [thumbup]

I disagree, seing how the same cards are listed (those marked by: ²) in a somewhat similar thread linked to in the Power section of "The Buyer's Guide".

Neeyik, that is a nice and clean list.