ATI Radeon X700 vs The Midrange

PCI-Express:
Intel Pentium 4 "Prescott" 3.2E LGA775
DFI LanParty 925X-T2 motherboard
2 x 512MB Corsair XMS2 PC2-4300 DDR2 RAM @ 533MHz, 4.0-4-4-12
Samsung Spinpoint 80GB SATA Hard drive
Microsoft Windows XP Service Pack 2
DirectX 9.0c


AGP:
Intel Pentium 4 "Prescott" 3.2E Socket 478
Asus P4P800-E Deluxe motherboard
2 x 512MB OCZ PC3500 Enhanced Bandwidth Platinum @ 400MHz, 2.5-2-2-6
Samsung Spinpoint 80GB SATA Hard drive
Microsoft Windows XP Service Pack 2
DirectX 9.0c

Cards used:<ul><li>ATI X700 XT 128MB, PCI-Express x16</li><li>NVIDIA GeForce 6600GT 128MB, PCI-Express x16</li><li>Connect3D 9800 XT 256MB, AGP 8x</li><li>NVIDIA GeForceFX 5950Ultra 256MB, AGP 8x</li><li>Connect3D 9800 PRO 128MB, AGP 8x</li><li>Gigabyte N59X128D; NVIDIA GeForceFX 5900XT 128MB, AGP 8x</li></ul>

Drivers:<ul><li>Catalyst AI with Catalyst Control Center - build 8.06</li><li>NVIDIA Forceware 65.76</li></ul>


The Test Suite:
Full-screen Anti-Aliasing and Anisotropic Filtering were handled by the application where possible, in order to allow the game to calculate AA & AF the way it was designed to, rather than being forced via the driver control panel. If there was no application control for AA & AF, they were forced accordingly via the driver control panel.

<ul><li>3DMark03:- Default test; 1024x768, 0xAA 0xAF</li><li>Doom 3:- Alpha Labs I play through; Game detail - High; 1280x1024, 2xAA 8xAF</li><li>Far Cry v1.1:- Regulator demo; Game detail - Very High; 1024x768, 2xAA 8xAF</li><li>Unreal Tournament 2004:- Ons-Dria custom time demo; Game detail - Maximum 1280x1024, 2xAA 8xAF</li><li>Splinter Cell: Pandora Tomorrow:- A play through of the first half of the first level; Game detail - High 1024x768, 0xAA 8xAF</li></ul>

Unfortunately, due to several driver bugs, the full test suite could not be used - the bugs that we have reported have been submitted to ATI and are detailed later on in this review.

Doom 3:-
We made use of a five-minute section of the "Alpha Labs I" level in order to give realistic game play results. "God Mode" was enabled and we chose to use the Veteran skill level in order to put maximum stress on the system while enabling us to keep the recording period as repeatable as possible.

Splinter Cell: Pandora Tomorrow:-
For our measurement period in this game, we've used the start of the first level, right up to where Fisher is told about the stream that runs parallel to the Embassy wall. My end of my measuring point is where Fisher climbs down from the building overlooking the Embassy and he is sitting on the edge of the stream.

Many of you may, or may not know that Splinter Cell: Pandora Tomorrow doesn't support Anti-Aliasing of any form. Basically the issue comes down to the fact that the game uses post-filtering effects such as night vision and thermal vision. The original scene is rendered to a texture (i.e. the scene, without night vision applied) and then a few passes of post processing is done on it; eventually the texture is copied into the main buffer. Due to the game being a DirectX 8 title, it is impossible to create multisample texture render targets. Thus, Application Preference Anti-Aliasing has been selected for this game on all graphics cards. With the mid to low end graphics cards that we're reviewing here, I chose to use 8x Anisotropic Filtering.


Real world testing:
While we see a number of time demos being introduced into my graphics card test suite, I still believe that real world testing should be a part of the review too, as the overall game experience of a graphics card is of utmost importance. Anyone can run a time demo and record the numbers, but I find that running time demos in conjunction with actually understanding the overall game experience provided by each of the cards gives us with a much better insight into how well an architecture performs. I've used time demos for repeatability, while at the same time I've played through parts of all of the titles used to give an insight into frame rate lag.