Geforce GTX 260 and 9800 GTX+: Extreme FSAA Benchmarks and OC

The FSAAA modes and tests weren't changed: Starting with 16xQ, we increased the load with 16xS, 32xS and 16x OGSSAA. 16xS is a hybrid mode, which consist of 4x Multisampling (FSAA) and 4x Supersampling (SSAA). 32xS also has 4x SSAA but combines it with 8xQ MSAA. Among gamers paying special attention to quality, this mode isn't just seen as beautiful but also as extremely hardware hungry. That isn't really surprising. Because of 4x SSAA the whole scenery is calculated with four times the size of the actual resolution. The combination with the 8x Multisampling results in a setting that breaks the neck of graphics cards with only 512 MiByte video memory. On top of that we also activate Transparency Anti Aliasing based on Supersampling.

16x respectively 4x4 Supersampling forces the graphics card to calculate every axis with the quadrupled resolution. This results in an excellent texture and shader smoothing. Combined with 16:1 AF via the driver the anisotropic filtering is de facto done at a rate of 64:1. You might guess that even the GTX 280 gets into trouble there.

All the Supersampling modes can be activated via the tool Nhancer only.

Candidates and clock speeds
In addition to the cards of the last test, now there are a default Geforce 9800 GTX+ and a Geforce GTX 260, which we simulated with a Geforce GTX 280 via BIOS flash. The former is based on the G92b, which is a 55 nanometer structure. The "old” G92 uses 65 nanometer technology and the G80 90 nanometer. The one new thing about the 9800 GTX+, which cracks Nvidia's nomenclature, is the clock speeds: 738/1,836/1,102 MHz are about 9 percent higher than the clock speeds of the old 9800 GTX.

All the candidates with their clock speeds:

- Geforce GTX 280 (1,024 MiByte) @ 750/1,512/1,296 MHz
- Geforce GTX 280 (1,024 MiByte) @ 602/1,296/1,107 MHz (default)
- Geforce GTX 260 (896 MiByte) @ 702/1,512/1,296 MHz
- Geforce GTX 260 (896 MiByte) @ 576/1,242/999 MHz (default)
- Geforce 9800 GTX+ (512 MiByte) @ 864/2,214/1,296 MHz
- Geforce 9800 GTX+ (512 MiByte) @ 738/1,836/1,102 MHz (default)
- Geforce 8800 Ultra (768 MiByte) @ 675/1,674/1,080 MHz
- Geforce 8800 Ultra (768 MiByte) @ 612/1,512/1,080 MHz (default)

Results
The Geforce GTX 260 affirms what we predicted within our comparison between the GTX 280 and the 8800 Ultra: The higher the workload gets, the better the new card gets, compared to the G80 flagship. Especially with 32xS FSAA the GTX 260 dominates the 8800 Ultra - this could be because of the ROPs (Raster Operators). The GT200 got 32, the G80 24 and the G92 only 16 of those calculation units, which are especially important for 8x MSAA.

The Geforce 9800 GTX+ is doing well at 16xS and 16x Supersampling, it often can beat the 8800 Ultra. That's the effect of the superior filling rate. If bandwidth and VRAM are the deal - especially with 32xS - it is slower, even when overclocked. In the worst case the 512 MiByte video memory don't suffice - this is as a general more often the case with 32xS.

Conclusion: Extreme FSAA and OC
32xS and 16x OGSSAA are GTX 280 territory. A heavily overclocked GTX 260 can reach this level most of the time. Graphics cards like the Geforce 8800 GTX/Ultra or 9800 GTX(+) can deal with 16xS well enough.