# How Much Power Does Your Graphics Card Need?

## Power Supply Efficiency

Now that we’re clear on the power cost of a graphics card, we can use this as the basis for a somewhat simplified theory for the selection of a power supply class. You need to differentiate between two values: the power measurement at the outlet—which is crucial for the overall electricity cost calculation—and what the actual power supply load is that is produced by the components.

At this point we must generalize somewhat, since we can’t guarantee the product quality of each power supply manufacturer based on specifications. The following statements are relevant to brand name power supplies, but fluctuations in quality are always possible. As always, you’re better off avoiding cheap power supplies with a poor track record for performance.

As an example, we take a 500 watt power supply, and plug into it components drawing 500 watts of power; the power supply is therefore at 100 percent load, theoretically. But the efficiency factor now determines how much the power supply absorbs from the electrical outlet and dissipates into heat. If the power supply in this example has an efficiency of 85%, this means 15% of the power taken from the outlet is wasted. So while we use 500 watts, the power meter shows 588 watts (since 588 * 85% = 500). These additional watts of power must not be included in the power supply choice; they are only relevant to the electricity cost.

A few notes:

1. You should easily be able to use the wattage specification of the manufacturer: a 500-watt power supply can be loaded with 500 watts.
2. The measured value at the outlet is higher and not the actual load of the components.
3. Test values are almost always measurements at the outlet.
4. Current brand name power supplies run at more than 80% efficiency with higher loads; lower efficiencies exist mainly with loads below 100 watts.
5. The higher the efficiency, the less energy is converted into heat, so costs remain low.
6. The lower the efficiency, the more energy is converted into heat, and the more cost increases, since more energy is required at the outlet.

In this table we see once again the differences between power supply (load by the components) and measured outlet (electricity cost). On the left is the power supply watt information; on the right is the theoretical number of watts, which the meter would display under full load. 85 percent efficiency is assumed.

Power Supply Performance Class in Watts
Measurement Socket (Full Load) in Watts
300
353
350
412
400
471
450
529
500
588
550
647
600
706
650
765
700
824
750
882
850
1000
Summary
• nukemaster
This article was due. No more you need an 800 watt psu for the 4870 , core2 quad ad 1 hard drive anymore

Guest11since Core i7 920 has TDP = 130W, how can it consume 85W only?

The TDP is more of a design thing. Almost all of Intels initial Core2 line had a TDP of 65 watts yet many took much less power. Intel gives a worst case of that type number and does not measure every cpu.

AMD does the same thing. They listed almost all the initial Athlon 64's at 89 watts yet many did not take that or give off that amount.

zxv9511.21 Jigawatts !!!

You act like you would need a small fusion reactor or maybe a bolt of lightning to get that?
• Inktfish
• sepuko
Took the words right out of my mouth.
• Anonymous
since Core i7 920 has TDP = 130W, how can it consume 85W only?
• Anonymous
1.21 Jigawatts !!!
• nukemaster
This article was due. No more you need an 800 watt psu for the 4870 , core2 quad ad 1 hard drive anymore

Guest11since Core i7 920 has TDP = 130W, how can it consume 85W only?

The TDP is more of a design thing. Almost all of Intels initial Core2 line had a TDP of 65 watts yet many took much less power. Intel gives a worst case of that type number and does not measure every cpu.

AMD does the same thing. They listed almost all the initial Athlon 64's at 89 watts yet many did not take that or give off that amount.

zxv9511.21 Jigawatts !!!

You act like you would need a small fusion reactor or maybe a bolt of lightning to get that?
• neiroatopelcc
So my system actually has a too big power supply to be effective?
I'm running a 3,4ghz c2d with 5x500gb sata drives, a dvdrw and a 4870 on a p35 board.
According to the article that's not going to draw the ~400W needed to get within effective range of my corsair 620 ....
• cynewulf
There's a mistake in the power under load for the 3870X2. It shows the same as the idle consumption. If only that were true!
• Inneandar
The TDP (thermal design power) is meant to be a guideline for the cooling solution, not the power consumption. To qualify for a cpu with a TDP of 120W, a cooler must be able to dissipate 120W. Practically, of course, this means it is an upper bound to (sensible) power consumption.

also small note: Is it just me or is it strange to see the 260 SLI consume more than the 280 SLI. maybe in need of a beefier test scene...
• zodiacfml
nice collection of data. i hope many learn from this and avoid recommending too powerful supplies.
• roofus
better off with too much power supply than not enough. at least if you over-spec the power supply you leave some breathing room for any additional components.
• kschoche
neiroatopelccSo my system actually has a too big power supply to be effective? I'm running a 3,4ghz c2d with 5x500gb sata drives, a dvdrw and a 4870 on a p35 board. According to the article that's not going to draw the ~400W needed to get within effective range of my corsair 620 ....

Absolutely Correct!
If you add all of those components together, and get 400W, and your PSU is only 75% efficient at that level, you're actually consuming ~500W and your PSU is eating that extra 100W. Though if you have a modern PSU, its usually not that bad, but goes to show that idiots who buy 1200W PSU's because it has a big number really are just that, idiots.
• zodiacfml
yes if you're planning to add a video card but adding a couple of hard drives, dvd drive, or upgrading the processor won't require more than 100W allowance.

roofusbetter off with too much power supply than not enough. at least if you over-spec the power supply you leave some breathing room for any additional components.
• one-shot
kschocheAbsolutely Correct!If you add all of those components together, and get 400W, and your PSU is only 75% efficient at that level, you're actually consuming ~500W and your PSU is eating that extra 100W. Though if you have a modern PSU, its usually not that bad, but goes to show that idiots who buy 1200W PSU's because it has a big number really are just that, idiots.

Let's take a step back. If his PSU is 400W and it is 75% efficient, then it draws 533.3W at maximum power draw. You are somewhat close, but you generalized. Different PSU's are more efficient at different load percentages, but 75% sounds alright for an older PSU. His Computer draws 400W and his PSU draws 533W from the receptacle on the wall.
• billiardicus
Toms,

Great article. This is exactly why I visit your page everyday. How about adding the GTX 295 and 285 in single and SLI configurations to the list? Hey, somebody has to ask right?
• hyteck9
Where is the Nvidia 295?
I think the 295 with an i7 OC'd to 4Ghz is going to be a pretty standard choice in the coming months... add a hard drive and a DVD-burner and your looking at 600WATTs at full load... so 800WATT PSU would be the right choice, yes?
• Niva
Well I learned something, I always thought that PSU ratings are based on what they can suck out, not what they put out to the computer components which they power... now that I'm looking at this article I feel pretty stupid for thinking this all these years.
• Pei-chen
One of the best articles ever. I hate proving myself when I tell people that using an inefficient video card will increase their electric bill by \$50 vs. an efficient card. An idiot actually told me that the difference between a GTX 260 and 4870 running 24/7 at idle for a year is less than \$10.

My system:
ASUS P5B Deluxe WiFi-AP
E6400 @ 2.56Ghz @ 1.135v with Speedstop enabled
8GB OCZ PC2-6400 RAM @ 1.8v
Arctic Cooling Freezer 7 Pro with PWM enabled
Antec NeoPower 550 PSU
2x Seagate 7200.11 1.5TB drive
1x Toshiba SATA DVD+/-RW drive
GIGABYTE GV-R485OC-1GH Radeon 4850 @350 core/500 ram

The idle power consumption at plug is 1.06A. Gaming load is about 1.71A with the Radeon overclocked to 730 core and 1130 ram. The 4850 consumption should be lower than typical 4850s because it uses GIGABYTE's custom PCB. Clearly my PSU is overkill as I only load it between 20~40% but it is not that easy to find good quality small PSU two years ago. Good thing it’s efficient.
• Anonymous
This is an exellent article!

There has been way too much bull about needing a 500 watter or more for a regular board, cpu and single graphics card. Its also great to have figures to compute total cost of ownership per GPU. The PC i game on is more on then off, so this info is significant. Thank you.
• Anonymous
Interesting article.. it should be noted that those of us with Geforce 2** cards and two monitors active will always be drawing the full 3D load of powe.

There is a bug in the latest two WHQL drivers which causes throttling to not occur even if there are no 3D applications active.
• hyteck9
Dont forget to spend \$500 on a UPS that can handle your 800-1000 watt power supply!