By Josh Walrath

Visual Quality

            There is much more to visual quality than static pictures.  For example, if you look at a very sharp and clear picture in one frame, it looks great.  However, once action starts to roll a user may start to notice a lot of pixel popping and other artifacts.  All graphics chips manufacturers deal with this the best they can, but there are always tradeoffs between quality and performance.

The back of the card and the heatsinks on the hot DDR-2 memory.

            Anti-aliasing quality is one area where there is a very distinct difference between ATI and NVIDIA hardware.  The current NVIDIA AA has been around essentially since the introduction of the GeForce 3 (though it did receive some improvements with the introduction of the GeForce 4 Ti series).  2X AA uses a 2 sample rotated grid, which gives very similar results to the competing ATI AA mode.  The ATI 2X mode does look slightly better due to the gamma corrected AA algorithm, which NVIDIA lacks.  Quincunx is a 2X AA with a “blur filter” applied.  This performs nearly the same as 2X AA, but the textures are blurred a bit as compared to other modes.  The overall effect though is not displeasing to the eye, and can be significantly better in appearance than 2X, depending on the situation.  4X AA then takes a step backwards and utilizes a 4 sample ordered grid AA method that is not all that much more appealing than 2X or Quincunx.  This sample pattern does take performance down a significant notch as compared to the other two, plus the output is nowhere near as pleasant as ATI’s 4X AA.  The higher AA modes are typically mixed multi-sampling and super-sampling, and as such the performance of these modes is very expensive.  These higher modes though do look quite a bit better, as there is also some texture filtering going on due to the super-sampling.  Still, these do not compare well to ATI’s 6X AA, especially in terms of overall performance vs. visual quality.

            Anisotropic filtering is another key feature that really can improve the overall quality of a 3D scene.  Both ATI and NVIDIA provide several levels of AF, with ATI going all the way up to 16X and NVIDIA going to 8X.  Both companies do not use a true anisotropic filtering method, but rather an angle dependent method that does differing amounts of filtering depending on the angle of the surface/texture.  ATI’s version is a bit more aggressive in its optimizations, while NVIDIA looks to do a slightly better job at covering the angles.  In most games I did prefer the AF output of the NVIDIA part, but there are a few exceptions where it appears that no AF is being done (but this appears to be more of a driver bug than anything else).  Another area where NVIDIA gets less than spectacular marks is in true trilinear filtering.  Due to driver settings that are unavailable to the user, NVIDIA does a bi-linear/tri-linear filtering mix, which some have termed as “bri-linear”.  This method mixes the filtering modes so that a little of each is done on a texture.  During gameplay there were no real quality issues with such an implementation, but be warned that it is truly not trilinear that we are seeing.  This does give a small speed advantage in filtering with a very small penalty in terms of image quality.

Pretty beefy power for a midrange card, but something tells me that we will be seeing much more of this in the future...

            Overall the FX 5700 Ultra had very good visual quality, and in some applications that were specifically coded for NVIDIA hardware, the output was fantastic.  After playing with the Radeon 9600 XT and the FX 5700 Ultra side by side, I found very few glaring differences between the two cards.  One of the biggest differences was that of Halo.  Some areas did not have the correct lighting, and the ATI part did look slightly sharper overall.  Another problem is with Need For Speed: Underground.  This title has some significant lighting problems with the GeForce FX series of cards, but these are supposed to be addressed in either a patch or a driver update.

Gameplay

            In most of the games I played I was able to keep the resolution up at 1024x768x32 with 2X AA and 8X AF without a problem.  This gave very good framerates in many games, and the overall quality of the scene was excellent.  4X AA at that resolution could give problems, depending on the application.  DX7, DX8, and OpenGL games all performed very well on this card at a variety of resolutions.  Titles such as UT2003, Morrowind, and Quake III (as well as its derivatives), all ran very well.  Once DX9 based titles were played, things were not quite so speedy.  Halo played OK at 1024x768, but in some of the firefights things got far too choppy.  Playing at 800x600 provided good gameplay.  The Radeon 9800 XT also showed this problem in firefights, but overall it was slightly smoother at 1024x768.

            Games such as Aquanox 2 provided good gameplay at 1024x768, but again certain scenes became choppy and the resolution needed to be turned down.  Any game based on the Quake III engine performed very well on this card.  The MOH series were able to run easily at resolutions higher than 1024x768 with AA enabled, while Call of Duty played very well at the medium resolutions with AA.

The FX 5700 Ultra is significantly longer than the competing Radeon 9600 XT.  Note that the FX 5700 Ultra still uses the universal AGP slot, while the 9600 XT will only work in AGP 4x/8x slots.

            The Far Cry demo had to be turned down to get adequate performance, and I believe that it was actually being rendered with PS 1.1 instead of PS 2.0.  Still, the game looked very sharp and played well.  The 9600 XT on the other hand did render the game in PS 2.0, but many of the settings also had to be retarded to get smooth gameplay.

            Overall the FX 5700 Ultra performed very well in a variety of games, and it should not give the average user any problems in running anything (other than NFS: Underground).  It had enough power to provide AA and AF at a variety of resolutions.

Next: Test Setup