Keep the results comming in please. We need more data from more cards and more users. So far we have 5 different rigs using 7 different cards in 5 different GPU families (GTX280, GTX470, 2x GTX580, GTX670, 2x GTX680)

These are the results plotted using the "updated" settings.

 

 

To me it's still looking like it's the shaders that make the impact. We will always get a few "abnormal" results so more data would be good.

 

So far the GTX670 and GTX680s have produced all the highest FPS data points so they are so far clearly better than GTX580 when it comes to SGSS AA in clouds.

 

I have noticed that they all are 4GB cards. What kind of GPU memory usage are you guys experiencing during the test? Is is over or under 2GB?

Saab, can you guys update the OP with the "updated" settings?

Holy cow!! :shok: What an improvement! This is solid proof that FSX requires a powerful GPU if you want to do any serious AA.

 

Keep in mind that these are not realistic IQ settings. 16xS + 4xSGSS + 16x AF is pretty insane, LOL

 

Keep the results comming in please. We need more data from more cards and more users. So far we have 5 different rigs using 7 different cards in 5 different GPU families (GTX280, GTX470, 2x GTX580, GTX670, 2x GTX680)

These are the results plotted using the "updated" settings.

 

 

To me it's still looking like it's the shaders that make the impact. We will always get a few "abnormal" results so more data would be good.

 

So far the GTX670 and GTX680s have produced all the highest FPS data points so they are so far clearly better than GTX580 when it comes to SGSS AA in clouds.

 

I have noticed that they all are 4GB cards. What kind of GPU memory usage are you guys experiencing during the test? Is is over or under 2GB?

 

Could be a shader bound or TMU bound type of workload looking at those graphs don't you think?

 

Saab, can you guys update the OP with the "updated" settings?

 

Only an admin can do that I'm afraid

Keep in mind that these are not realistic IQ settings. 16xS + 4xSGSS + 16x AF is pretty insane, LOL

Well, with an overclocked GTX680 it seems to be realistic settings and still be able to maintain 30FPS. :P

 

Could be a shader bound or TMU bound type of workload looking at those graphs don't you think?

Yeah, we can exclude the ROPs at this screen resolution.

 

Looking at it so far I'd say that it doesn't seem to be the TMUs. Sure enough, we get scaling within each architecture but if you look at our 3 Fermi results (19-24FPS) and compare how little the 2-3 times higher Texel fillrate of the Keppler results (the ones over 24FPS) yield in FPS. Never mind the comparison to our only Tesla result that has almost the same Texel fillrate. But more data would be welcome.

Hey guys, somehow I forgot to add some more cloud layers to the test flight. I should have noticed this was the reason why it wasn't so GPU limited.

can someone please run this one?

 

it's 8xS + 4xSGSS + 16xAF this time

You guys want me to run this test with my ATI 5850? Obviously some things would be different but it would be interesting.

Did a run as your settings

in cfg:WW=False, cloud 12 , max texture 4096.

run in windowed monitor usage and gpu

With Rex

http://dl.dropbox.co...gpu3 screen.png

 

Frames, Time (ms), Min, Max, Avg 9530, 300000, 29, 38, 31.767

 

92% GPU usage - So much for the "FSX doesn't use the GPU" theory!! :lol:

You guys want me to run this test with my ATI 5850? Obviously some things would be different but it would be interesting.

 

It would definitely be interesting to see how ATI cards do in the test at different AA settings, even if it won't be easy to compare to nVidia's AA modes.

 

92% GPU usage - So much for the "FSX doesn't use the GPU" theory!! :lol:

 

You know we are forcing it with tons of antialiasing right Tom?

Extricly speaking "FSX doesn't use the GPU" is still true for the most part here, since those AA modes are nothing FSX is aware of anyway

I took a look at the specs on my 5850 and it has a lot of shaders for an older card. The card is an ASUS EAU 5850 TOP DirectCU. It has 1440 shader units. Here is a link to it.

 

http://www.techpowerup.com/mobile/reviews/ASUS/EAH_5850_TOP_DirectCu/1.html

 

I will run the benchmark today and post the results and settings.

Okay not sure I did this right, but I loaded up Fraps, and loaded this flight with heavy Orbx clouds. Basically I was flying through clouds the whole time, and didn't see anything but clouds. Not sure how you guys get the stats, but my onscreen counter never went below 15 or above 17. Mostly it fluctuated between 16 and 17.

 

This was with 1920x1200 resolution using REX clouds. I have Orbx PNW installed.

You know we are forcing it with tons of antialiasing right Tom?

Extricly speaking "FSX doesn't use the GPU" is still true for the most part here, since those AA modes are nothing FSX is aware of anyway

 

He-He! It was a joke, hence the :lol: That said I went and turned off AA and AF, and I still got a high of 40% usage! I'm sure that has to do with the Shaders though!! (I use *******' Shader 3.0 mod, which I think does use more of the GPU then the default Shader 2.0.)

Okay this is interesting! I started tweaking some of the settings, and didn't really hit gold with the cfg adjustments. I played around with BufferPools. Then I started tweaking Catalyst Control Panel. It seems that I had a Tessellation setting of 64. I backed it down to 32 and my frames went way up without a significant loss in quality. Then I dropped the MipMap setting down a notch.

 

When I load the test flight, I see jaggies in the cockpit. But when I active TrackIR they disappear completely. And now I am running the test and getting 27 to 30 fps solid.

 

I tested it in the F1 Mustang with heavy clouds, and pretty much locked at 30fps with zero jaggies.

 

Still testing but promising so far!

Just got done doing a bit more testing in the F1 Mustang using Orbx. All in all this is great performance with very little trade off in quality. In the Mustang using Orbx PNW, Ultimate Traffic 2 (30%), REX and Active Sky 2012 I am getting 21-26 fps and the sim is smooth and crisp.

 

Not bad for an ATI card 2-3 years old. I might be giving those GTX580's a run for their money! lol

 

Next up, Mustang at KLAX with higher traffic settings.

These results are with REX 4096 clouds

 

Frames Time (ms) Min Max Avg 2548 300000 7 10 8.493

 

I guess my GTX 460 SC is showing its age.

About to run this test with my 680 FTW+ 4GB but which test do you use ? I think there are 3 variations on it here .