New D3D FSAAViewer Version 4!

Colourless

Monochrome wench
Veteran
I uploaded this up a few days ago, but just didn't get round to annoucing it. I've released a new version of my D3D FSAA Viewer. Download it from the 'normal' place: http://www.users.on.net/triforce/d3d_fsaaviewer/d3d_fsaaviewer-4.zip

Changes for this version include, a new Direct3D 9 version as well as the Direct3D 8 version. You can use the 2 versions to check for D3D8 and D3D9 driver differences. Of course you probably won't find much difference.

Another change is ability to change the Multisample Mask so you can isolate the various subsamples. Used Ctrl+Num for samples 1-10 and Ctrl+Shift+Num keys for samples 11-16.

A third change I made was to the texture aa test gradients (top and right). Now they use Alpha Testing (when before they didn't) so driver force anisotropic filtering shouldn't cause incorrect results.

Lastly, and possibly the most useful feature is I included the ability to 'remove' OGSS from the results. This is most useful for when there is a 1.5x factor involved which totally messes things up. An example would be Nvidia's 6xS mode. It uses 1.5x1 OGSS. This causes problems for the Texture AA test. If the 1.5x1 OGSS was removed from the results what you would be left with is just the 2x2 OGMS. See the included readme file for details on what to do.

Here's an example of 1.5x1 OGSS causing a problem:
badfsaa.png


And here is what things look like with 1.5x1 OGSS removed:
goodfsaa.png


It of course needs to be noted that taking screenshots of this mode is not useful. You will not be seeing the true sampling positions. However, it may be useful in determining the true sample positions yourself. Also this has not been tested on any card that actually does 1.5x supersampling.

A second note, Yes, those shots were taken on a Radeon 9700 Pro, and No, you can not do supersampling on a Radeon 9700 Pro. Those 2 shots simulated Supersampling by using Render To Texture.

Last thing, the FPS values you get aren't really useful in the slightest to determine anything. The program yields execution so it doesn't run as fast as it could. You should get around 100FPS max. If you are running below that, you are probably transformation or trisetup limited. D3D FSAA Viewer transforms A LOT of vertices. Using the Full Screen Geometry mode, there is approx 1 quad (2 triangles) per pixel. Since a triangle list is used there are 6 verts per quad. That ends up at about 6 verts per pixel. At 1024x768 there are going to be some 5 million verts transformed per frame. I personally get around 20 FPS in that mode, which is 100MVerts/s being transformed. It is interesting to know, but not too useful.

Thats it for now. Unlike the previous versions, if people want it, i might consider 'opening' my sources to it. Can't say there is any real reason not to.
 
Colourless said:
Thats it for now. Unlike the previous versions, if people want it, i might consider 'opening' my sources to it. Can't say there is any real reason not to.

yes, please
 
What are you saying? The reason Ati don't implement SSAA is because it will not work on their hardware?
 
K.I.L.E.R said:
What are you saying? The reason Ati don't implement SSAA is because it will not work on their hardware?
I think he's simply saying that SSAA is not implemented in ATI's 9700 drivers and cannot be forced in any way. It can simply be simulated by rendering the scene at a high resolution to an offscreen buffer and then rendering the resulting texture to the screen at the desired resolution.
 
Colourless said:
They are for MSVC.Net, but should probably compile for MSVC 6 as well (not that i've tried).

they compile like a breeze under vc6. btw, the app shows my gf2 mx at work to have 2x and 4x supersampling (i never cared to check this little sucker up till now).
 
Back
Top