From the benchmarks above, the overall performance of the Wildcat Realizm compared to the NVIDIA Quadro FX is somewhat strange. It is clear that the Wildcat trumps in the SPECviewperf benchmarks, but showed relatively no performance benefits in the real world applications.
Full Screen Antialiasing (FSAA) Quality
We compared the antialiased (AA) result between the Wildcat's SuperScene technology and the Quadro Ordered Grid FSAA technology by taking a screenshot of the transpar.max file in 3ds max 7. Both cards used their respective vendor-supplied 3ds max driver.
We'll start with some viewport images in 3ds max showing aliased and antialiased images. We're only interested in AA quality so AF was set to auto/application-controlled. Using the supplied 3ds max driver, the Wildcat Realizm didn't actually antialias until anisotropic filtering (AF) was set to at least 2x. Further improvement in FSAA quality was seen when AF was set at 4x. This seems quite strange since antialiasing quality is independent of anisotropic filtering.
|Wildcat Realizm 200 |
(click for full PNG-8 image)
|"8x AA" (rather no AA), |
|8x AA, 2x AF ||8x AA, 8x AF |
From these image samples the following are noted:
- The Wildcat Realizm 8x AA appear to do a better job of smoothing the chessboard grid than the Quadro at 16x AA.
- The Wildcat Realizm doesn't appear to do any antialiasing of the chess pieces themselves or the chess timer.
- The Quadro FX 3000 renders the pawn chess pieces a lighter shade of green which would make the piece itself appear less "jagged" due to the reduced contrast against the background. The Quadro also appeared to exhibit rendering errors at the highest level of AA.
|Quadro FX 3000 |
(click for full PNG-8 image)
| No AA, auto AF ||8xS AA, auto AF ||16x AA, auto AF |
The result is that the Wildcat's implementation of antialiasing obviously works well, but why is it that the board grid is antialiased but other objects such as the chess pieces and the chess timer are not? This selective method of antialiasing is puzzling.
To gauge performance with FSAA turned on, we used CineBench 2003 with the maximum degree of AA and AF applied for both the Wildcat Realizm and the Quadro FX boards.
Best quality is defined to be 8x AA, 8x AF for the Wildcat Realizm 100 and 200 and 16x AA, 8x AF for the Quadro FX 1100 and 3000. The Wildcat Realizm's AA performance was significantly less than the competing Quadros in CineBench 2003.
The Wildcat Realizm performed well in SPECviewperf, but lost its performance-leading results in 3ds max and CineBench. 3Dlabs said that we should have seen better performance in 3ds max's "Bmark", but we're sticking to our results because after three driver revisions and numerous benchmark testing we still got consistent results from day one.
Maxon Cinebench showed that the Wildcat Realizms kept up with the Quadro FX 1100 and 3000 in the OpenGL benchmarks. Antialiasing performance was disappointing compared to the Quadro FX 1100 and 3000.
The Wildcat Realizm 100 and 200 are quiet cards that operated with great stability, but the initial drivers worried us with rendering errors in 3ds max. 3Dlabs was responsive and the newer driver resolved some issues experienced, but left one antialiasing issue unresolved as this article was being finalized. The selective nature of the enabling antialiasing is puzzling, and last I heard anisotropic filtering did not have to be enabled for antialiasing to work.
The Wildcat Realizm's strength: its large amount of onboard memory compared to the Quadros would have pushed large memory-intensive applications in the Wildcat's favor but none of our test software appear to take advantage of video cards with that kind of memory capability on offer.
Price-wise the Wildcat Realizm commands a premium over the Quadros, so if you're in need of large amounts of onboard RAM (up to 512MB) the Wildcat Realizm would be worth looking at. For everything else, the Wildcat Realizm faces very stiff competition.
Standard Performance Evaluation Corporation