Is your graphics card inefficient or faulty? GPU idle power draw comparison

[Ilstar57]

Hey there! So, recently I had purchased an Asus Prime 5070 Ti, which had abnormally high idle power. I could not fix that regardless of what I did. I was just about to just accept it and live with it, until the GPU decided to try and burn my house down. Though I've since gotten a refund on that, and still waiting for a PSU warranty replacement, I thought I'd share a thread, so that we can compare the idle power draws and note any cards which behave abnormally.

To take part, all you have to do is post how much power your GPU is drawing while not doing any serious task, like gaming. Bonus points if you also include your monitor(s) resolution and refresh rate. This information can be obtained through the GPU software (nVidia app/AMD Adrenalin, through GPU-Z sensors "Board power draw", or through an app like HWInfo). Make sure to include the GPU brand and model.

Please keep in mind that having a higher idle is only of concern if it is consistently significantly higher than other examples of the same model. Some brands cards just take more than others and, while a little inefficient, this is completely normal and not a cause for any worries.

For reference, the faulty Asus Prime 5070 Ti was drawing between 35 and 40W at idle on a single 1440p 180hz monitor. Even dropping down to 1080p 60hz only shaved ~3W off that. In comparison, several friends that I know that had that exact same GPU, but which had no issues with it, reported idle power draws between 13 and 20W on that card with a single 1440p monitor, and about 20-24W with two 1440p monitors.

I've also asked around a bit and found several pieces of information:
- Gigabyte and Palit cards tend to have a higher idle this generation. Usually around 20-25 for a 5070 and 30 for a 5070 Ti.
- MSI cards either have great or very poor power draw. Some models are easily under 20 (5070/70 Ti/80 tier), while others burst above 40 at times.
- PNY cards generally have pretty good idle power at a 5070 (>20) and 70Ti (>25) level, but at a 5080 level they will go to about 35-40W - considerably higher than average.
- 5070s are usually quite efficient at idle
- 5080s (>18W) can even have a lower idle than 5070Tis (17-28W) for some reason
- 5090s are power hogs, and usually draw 30-50W at idle
- The AMD Radeon 9000 series cards are usually efficient, with >15W idles. Radeon 7000 are less efficient, and suffer from an issue where having multiple monitors with mismatches resolutions/refresh rates can cause terrible idle power (over 50, over 80, sometimes even reaching 100W).

A few GPU examples (each GPU is one example):
PNY 5070 - 7W
PNY 5070 - 12W
Asus Prime 5070 Ti - 20W single monitor, 23W multi-monitor (all 1440p)

Some information from various forums, discords etc:

• 5090
  • PNY - 50W (multi-monitor; Linux)
• 5080
  • PNY - 39W (multi-monitor)
  • PNY - 35W (multi-monitor)
  • MSI Ventus 3X OC Plus - 30W
  • MSI Shadow 3X - 31W (multi-monitor; first monitor 4k 120hz, second monitor 1440p 165hz)
• 5070Ti
  • MSI Shadow 3X - 15W (single-monitor); 35W (multi-monitor; 3x 4k monitors)
  • MSI Gaming Trio - 45W
  • MSI Ventus 5070Ti - 18/19W

I'd be happy to hear some more examples

 

[Ilstar57]

4090 FE
1440p 144hz
9w idle at desktop

That is a great idle for that tier of card! Congratulations!

Have had HWmonitor open for a bit as I was monitoring CPU temps.... 7900XT, 1440p 165Hz and 4k 60Hz. Minimum 29W, hovering between 30-36 at the moment.

Seems normal to me for that tier of card on double monitors with mismatched resolutions and refresh rates (a known Radeon 7000 weak point).

To the OP, I suspect we would get more useful data if you also asked for both GPU and memory clocks.
Both "idle" and video playback power usage has a lot to do with clocks. And various GPU's BIOSs and the respective drivers will have some kind of lookup table. Power and clocks for - as an example - 1080P @ 120Hz can be quite different than 1080P @ 144Hz if for the latter the driver and/GPU BIOS decide VRAM clocks have to go from 300MHz to 1000MHz. Sometimes VRAM clocks can be overridden but that can lead to glitches.

Yeah, more data would always be good, though I'd also guess that asking for too many clarifications could discourage people from taking part.

19 to 23 Watts, RX 7800 XT.

This is actual idle, having a browser open or even some apps is not idle as those use GPU acceleration, in which case you will see high 'idle' power, for me about 40 watts and the memory runs at between 1300 and 1500 Mhz, not the 45 Mhz you're seeing here.

I think you can change that by turning off GPU acceleration in those apps.

Yes, I agree. Browsers do use hardware acceleration. With that said, many GPUs will handle that type of load by only adding a few watts on top, while as others will jump up quite significantly. Others still may spike for a few seconds when a new page is loaded before returning to idle. You definitely can turn off GPU acceleration for most apps that use it. Your reading appears generally normal to my eyes.

 

[Ilstar57]

Sapphire 9070XT Pulse
5120x1440p @ 240hz
7w idle desktop.

I'll test my 4090fe system later.

We've got to give a shoutout to AMD. They went from having very hot, power-hungry cards to this amazing efficiency at low loads. 7w on that tier of card, with such an ultrawide and high refresh rate monitor is great!

 

[Ilstar57]

But they still run hot when gaming like the older cards, to the point everyone is undervolting them in the owners thread though.

True. The 9070 and 9070XT do use more power than the 5070 and 5070 Ti under load, respectively. With that said, I undervolt all the GPUs and CPUs I get, regardless of brand.

 

[Ilstar57]

Put it this way, when I've had the PC next to my leg or on the desk, it was the hotter build versus the Nvidia 4XXX I've owned, and the area of the room I was using the PC was a lot hotter
So real world is what matters and unfortunately, the AMD was hotter in real world use, also undervolting it never seemed to make any difference, it was very stable, but would just boost higher, so I achieved nothing in terms of quietening the fans, temps or power saving, which is great for some people, but not what I was trying to achieve.
Everyone in this thread seems to be undervolting them too to bring down the temps/power draw. So I'm not alone here, as I wasn't with the RX 6800 XT, 7900XT/XTX series.
I can put up with it, but there's no denying, especially in summer, I'd rather a cooler room/PC, and it to run quieter and not have to have headphones on, especially on a 3 fan card

As I say though, I'll continue to own both brands, it'll just depend what one is better for me at the time of release, I'm not loyal to either, well to be fair, I was always more of an AMD guy, but I do like the modern featureset of the last few generations of Nvidia, and the speed at which those features come out and continue to be updated, along with the wide range of games supported, it is what it is.

Personally though, I'd love to see some crazy 9950XTX or something come out, and grab that next

Well, to lower the heat what you need to do is lower the power consumption. On AMD cards this is done through power limiting. By doing this, you can lower the power target and hence reduce power consumption. This generates less heat, lower temps, quieter fans and less heating of your room, in exchange for lower clocks and hence - performance loss. Undervolting is often used in conjunction with this, usually to offset the performance loss caused by the power limit decrease. If left without a power limit decrease, the undervolt will just result in a higher clock, which will increase performance beyond stock levels, but won't decrease heat output as the power consumption hasn't changed - the card has simply clocked higher with the new headroom that the undervolt provided. The two combined will allow you to keep your stock levels of performance and use less power, causing less heat. If you'd like to take it even further, you can power limit even more and accept some performance loss in exchange for a much cooler card. Most AMD cards will allow you to slice off about a third of their overall power consumption. A 300W card can become a 200W card this way, saving you the heat output of about 10 modern new LED light bulbs, or one old-fashioned incandescent bulb.

 

[Ilstar57]

The RX9070 doesn't use more power than an RTX5070:

MSI GeForce RTX 5060 Gaming OC Review

Compared to the base RTX 5060 that we initially tested, the MSI GeForce RTX 5060 Gaming OC offers a very quiet cooling system, excellent temperatures, a metal backplate, RGB lighting, and idle fan-stop, adding a touch of premium polish. But is that enough to stand out in this crowded market segment?

www.techpowerup.com

Under normal gaming load there is a 4W difference,but under RT,Furmark,etc the RTX5070 uses more power. TPU actually measures card power draw.



My RX9070 Pulse runs fine in a 12L NCase M2. I would be more worried about how hot an RTX5070 runs because it is drawing as much power or even more than an RX9070 over a much smaller die size.

The RX9070 is one of the most efficient gaming cards of this generation.

Hmm, this is interesting. Don't know if I watched a review with different numbers or simply misremembered. That 9070 is really sipping - both at idle and under load.

 

[Ilstar57]

TPU measure actually card power draw. If you look at most reviews which measure card power draw,either the RX9070 on average is matching or exceeding the RTX5070 in overall effiency.

TPU actually go above and beyond in measuring power draw. They not only measure using hardware (which is more accurate), but also take measurements at idle, in video playback, framerate limited gaming, normal gaming, raytracing and stress tests. These are among the most exhaustive tests I've seen out there, with only a few German, Korean and Russian websites doing something similar.

 

[Ilstar57]

Unfortunately I got my hands on a Palit 5070 Ti GamingPro. Low price, but idles at 28W and has very noisy fans. The fan problem was mostly fixed by installing a PNY bios.

I've tried several other bioses in hope to get the power consumption down, but so far it did not work. One bios increased the power consumption to 32W though.

If anyone else here has a more efficient 5070 Ti, I'm curios to hear what your "GPU chip power draw" reading in GPU-Z is.
Mine is 11-12W, and I suspect the root issue is related to this value being higher than it should.
I've seen at least one screenshot from a 5070 TI where the chip was idling at only 5-6W.

https://hardwareand.co/images/thumb/gpu-z-rtx-5070-ti-frequences_fhd.webp

I don't get it... how my chip power consumption can be abnormally high when my idle voltage (800mV) and frequency (180MHz) both seem normal.

Well, if it's any help, my faulty 5070 Ti (Asus Prime) had a chip power draw of ~13.3W and a board power draw of ~35W. Voltage around 816mV, clocks at 508 MHz (though this spiked a couple of times while taking the reading, so in reality it's a bit lower. The power figures are correct, though).

 

[Ilstar57]

Thanks, it seems the board power draw is roughly 2.5 times the gpu chip power draw in all cases.

I guess there are shunt resistors involved in measuring the "gpu chip power draw".
Thus, I wonder if these are placed before or after the VRMs.
If they are placed before, the explanation may be that some cards are equipped with cheap VRMs that incur power losses.

Also, I checked my GPU with nvidia-smi and it is running at power state P8, which is fairly low, but there are lower states.
The lowest one is P15 but that one is "sleep mode" I think, not intended for normal use.

Good question!

AFAIK, P8 is the lowest state for an nVidia graphics card that is connected, powered on and displaying a video output. P8 is as idle as we can reasonably expect it to get. What I've noticed is that some brands tend to have far higher or lower idle than others. Cheap VRMs perhaps?