DisplayPort wasting 32 W when display powered off

[snoukkis]

Hi,

I was doing some experimenting to determine how low I can get my computer power consumption without putting it to sleep. I prefer to leave it on so it can keep working on light background tasks (none running during testing of course).

With all the display cables disconnected, I can get down to about 75 W. I'm measuring the AC power between the computer PSU and mains. I'm not measuring the power from the displays themselves.

I noticed that connecting my primary display with the DisplayPort cable adds about 32 W to total power (+43%!). My secondary display over VGA seems to have no significant effect on power consumption.

So I configured Windows to turn off my displays after 1 minute, expecting to save 32 W when it goes. I was surprised to find out that power consumption stayed the same. D: I also tried pulling the power cord from the display, but still no change. Finally I pulled the DisplayPort cable and the power dropped 32 W.

I don't know how DisplayPort is designed to work, but it seems kind of stupid to me that I need to physically pull the display cord to save power.

Not sure who/what to blame. Microsoft? AMD? VESA? ASUS? myself? (surely not myself!) Anyone else noticed this?? I would test with integrated Intel GPU, but I only have a mini-DisplayPort cable. I'll try to get my hands on a regular DisplayPort cable.

My setup:
ASUS Z97-Pro
Intel 4790K
8 GB RAM
AMD Radeon HD 7950
PSU... can't remember... something way too big to be efficient at low power...
primary display: ASUS VG248 1920x1080@144 over DisplayPort
secondary display: HP w2408h 1200x1920@60 over VGA (HDMI broken)
Windows 7 64 bit, balanced power profile, similar results with power save profile

 

[zebrax2]

Are you sure its not due to running dual display? GPUs need to clock a little higher when idle if your running more than 1 display IIRC

 

[3DVagabond]

I'm not up om the technical aspect, but when you attach a monitor to a card Windows loads up ~256MB of VRAM for some function. Again, sorry, but I don't know the technical reason. I would assume this uses power for the card though.

 

[tweakboy]

Pull the Plug.

 

[ShintaiDK]

Sounds like dual display and the broken clock setting for AMD cards with higher idle clocks.

Try run GPU-Z and see what the clocks are with 1 and 2 screens attached.

 

[KingFatty]

Yes, OP, can you please repeat the experiment, except use *ONLY 1* display at a time.

First, plug in the VGA display with no other displays. Then disconnect it and connect only the DP display with no other displays. Then connect none at all. Then connect both. See how the power usage is for all 4 of those conditions.

 

[snoukkis]

Are you sure its not due to running dual display? GPUs need to clock a little higher when idle if your running more than 1 display IIRC

Sounds like dual display and the broken clock setting for AMD cards with higher idle clocks.

Try run GPU-Z and see what the clocks are with 1 and 2 screens attached.

Yes, OP, can you please repeat the experiment, except use *ONLY 1* display at a time.

First, plug in the VGA display with no other displays. Then disconnect it and connect only the DP display with no other displays. Then connect none at all. Then connect both. See how the power usage is for all 4 of those conditions.

You guys were absolutely right. Good idea using GPU-Z.

My GPU seems to have two idle modes. Let's call them "mode 1" and "mode 2" for convenience:

1. GPU clock 500 Mhz (50% max), memory clock 1250 Mhz (100% max), ~107 W total system power.
2. GPU clock 300 Mhz (30% max), memory clock 150 Mhz (12% max), ~75 W total system power.

Indeed having two displays connected resulted in the first mode with high power readings.

Having only the second 60 Hz display connected resulted in the second mode and low power readings. Similarly having no displays connected logged the same low reading.

There was a twist though. Initially it seemed that having only the primary display connected with DP resulted in mode 1. Then I tried switching it from 144 Hz to 60 Hz and voila, clocks dropped to mode 2. Then tried 120 Hz, and still mode 2. Nice.

So in conclusion it seems that indeed having two displays connected or one display in 144 Hz mode prevents my GPU from idling properly. Connecting only one display in DP with max 120 Hz refresh rate solves the problem.

Thanks guys for help! :thumbsup:

Any information/speculation why dual displays or 144 Hz causes problems?

 

[thesmokingman]

Nice, you figured out the cut off point. Btw you don't need to run 144hz on the desktop. You could even run 60hz(or some variable) on the desktop and 144hz in games.

 

[samboy]

I don't have any links; but I recall that both NVidia and AMD had power optimization issues when dual displays were run a different resolutions. At least all the power saving optimizations are optimized for a configuration where the resolutions are the same.

Easy to test by setting both resolutions to be the same (even if it looks bad/scaled on one display) and see if the problem goes away........

 

[ShintaiDK]

Any information/speculation why dual displays or 144 Hz causes problems?

It just seems broken on those AMD cards. Its been an issue with AMD for ages. I had it on a HD5xxx card too. Its something to do with flickering when the GPUs in question change clock.

 

[PPB]

AMD has trouble with their drivers detecting correct EDID values from monitors and deciding which pstate the GPU should run at. I think this comes from the times when probably GPUs werent powerful enough to drive 2 displays with idle clock pstates.


This also happens on AMD's end when you modify EDID values via CRU. Changing any value in Front Porch/Sync Width/Back Porch will result in the GPU never going into idle powerstates.

 

[snoukkis]

Easy to test by setting both resolutions to be the same (even if it looks bad/scaled on one display) and see if the problem goes away........

Tried with setting both displays to 1024x768 @ 60 Hz. Unfortunately they remained in the high-power idle "mode 1".

AMD has trouble with their drivers detecting correct EDID values from monitors and deciding which pstate the GPU should run at. I think this comes from the times when probably GPUs werent powerful enough to drive 2 displays with idle clock pstates.


This also happens on AMD's end when you modify EDID values via CRU. Changing any value in Front Porch/Sync Width/Back Porch will result in the GPU never going into idle powerstates.

There's more details here than I understand about display timings, but...
actually I get my VGA connected display running in 1920x1200 only by disabling EDID in AMD catalyst and setting manually the maximum resolution to 1920x1200. Without that I'm limited to 1600x1200. I didn't need that hack before when the HDMI port was still working on that display.

For the purpose of the above test I re-enabled EDID, thus the 1024x768 resolution which was common for both displays.