GPU usage in SLI

[Kaapstad]

I did these runs the other day because I was looking for a bottleneck but got something else instead.

Both runs were done @4.0ghz on CPU and GPUs running @stock

The first pic is using the settings used in the Heaven benchmark thread. At these settings my setup gets thrashed by a 3 way HD 7970 setup. The interesting bit in the pic below is the GPU loads their all over the place and not very good.

http://imageshack.us/photo/my-images/341/heaven1920x1080.jpg/

In the next pic I max out everything
Resolution 2560 x 1600
Tessellation = Extreme
Anisotropy = 16
AA = 8

http://imageshack.us/photo/my-images/16/heaven2560x1600extreme.jpg/

If you now look at the GPU loads their much better @98% most of the time. At these settings I can get the better of a 3 way HD 7970 setup the green team strikes back.

The strange thing is I was expecting a bad bottleneck using the lower settings but its not there.

 

[Gregster]

Very strange how more demanding the bench, the better the 690's work. Could it be down to the 690's thinking they do not need to work so hard, so instead throttle back, Only to find out they should be working and fly back up?

I had no drops in GPU usage on one card but when in SLI, the drops are often. I will have to look and see if it is the same on the higher specced test.

Edit:

Just 2 spikes in GPU usage, at the start and in the corridor. Far better performance than the Heaven settings we use here. Very strange though.

 

[Kaapstad]

One thing I think is going on is Task Manager is not a good tool for spotting bottlenecks.

Task Manager I think measures how hard a CPU is working. The problem is for a high end CPU to work with a GPU there is very little effort involved per frame and so shows up as a low load in Task Manager. If the CPU is not running fast enough to keep up with a GPU it will not show up in Task Manager because the CPU has not done anything to be measured.

With the Heaven 3.0 benchmark strange things happen with nvidia cards. As I said earlier the higher the settings the better mine work. At the settings used on the heaven benchmark I can not get anywhere near the HD 7970s even when overclocking. At extreme settings its a different story I can get the better of them. One of the reasons for this is if you look at any run for nvidia cards they seem to hit a max frames brickwall @370 -380 fps regardless of settings. When the settings are maxed this problem is removed and it pulls down the HD 7970s max frames advantage.

There seems to be something in GTX 680/90 cards that nvidia have put there to limit max frames.

 

[Kaapstad]

Does anyone know whats going on here

I did two runs in Heaven 3.0 with the GPUs running @stock speeds and the CPU @4.0

The first run used
Resolution 640 x 480
No Tessellation
No AA
No Anisotropy
See Pic below

The second run was done using the settings from the Heaven 3.0 benchmark thread
Resolution 1920 x 1080
Normal Tessellation
4x AA
4x Anisotropy
See Pic below

If you look at the scores and FPS they have actually increased at the higher resolution and settings. I wish this worked for games.

Could anyone explain why ?

 

[chipachap]

Does anyone know whats going on here

I did two runs in Heaven 3.0 with the GPUs running @stock speeds and the CPU @4.0

The first run used
Resolution 640 x 480
No Tessellation
No AA
No Anisotropy
See Pic below

The second run was done using the settings from the Heaven 3.0 benchmark thread
Resolution 1920 x 1080
Normal Tessellation
4x AA
4x Anisotropy
See Pic below

If you look at the scores and FPS they have actually increased at the higher resolution and settings. I wish this worked for games.

Could anyone explain why ?

You would probably need 2 cpu's to achieve the high frames that would be rendered at such a low res.

 

[whyscotty]

Uploaded with ImageShack.us


Don't know if this helps or not, ran the bench using your settings

My gpu loads also were all over the place, the cards down clocking completely on numerous occasions.

As you can see i didn't hit the fps limit you mentioned.
It may be a 690 thing

I did however come across this on the EVGA forum,
don't know if its of any use to you

http://www.evga.com/forums/tm.aspx?m=1748031

 

[Gregster]

I have decided that Heaven is shocking for Multi cards if they are green. It should be a constant bench instead of spikes regardless. Maybe it is Nvidia drivers but games don't spike like this.

 

[Kaapstad]

Ignore

 

[Kaapstad]

I have decided that Heaven is shocking for Multi cards if they are green. It should be a constant bench instead of spikes regardless. Maybe it is Nvidia drivers but games don't spike like this.

+1

The reason I have been posting in this thread is to show there is some very strange things going on with multi gpu sli and the Heaven benchmark. Its very strange to get higher fps by raising the settings. Having said that whyscotty's setup works much better than mine @640x480. Its a good job I am never going to play games at that resolution.

 

[Kaapstad]

Uploaded with ImageShack.us


Don't know if this helps or not, ran the bench using your settings

My gpu loads also were all over the place, the cards down clocking completely on numerous occasions.

As you can see i didn't hit the fps limit you mentioned.
It may be a 690 thing

I did however come across this on the EVGA forum,
don't know if its of any use to you

http://www.evga.com/forums/tm.aspx?m=1748031

I think as you say its a 690 thing. I have noticed something simular when running performance test 7 (the 2D tests and also the low resolution 3D tests) the cards just throttle down. Its not something Im worried about in fact I found it quite funny and also interesting.

The thing is for normal use (games) I am never going to play them with low settings.

I suspect its something nvidia have put in to work in the same way as when Furmark is run.

On another note have you ever thought about seeing how yours cards compare to a HD 7970 trifire setup using extreme settings. I think it will be a lot closer than in the Heaven 3.0 benchmark thread.

Using the following setings
Tessellation = Extreme
AA x 8
Anisotropy x 16
Resolution 2560 x 1600
My setup manages 117 fps and a HD 7970 trifire setup about 112 fps.

Where you can oc higher than me I think it would be a good contest 3 x GTX 680s vs 3 x HD 7970s

 

[andybird123]

I'm not at all surprised by this behaviour - the whole premise of the boost functionality is that it boost up *when it's needed*... running low enough settings will mean it doesn't bother boosting

on a single 670 @ 1080p in COD Blackops I've seen it stick at 900 odd and still getting constant high FPS and even dropping to 400-600mhz... because it's simply not needed