Conclusion:
The GeForce 7 architecture is very sensitive to bandwidth. Without the presence of additional bandwidth, there's actually no point of adding more quad(s). It would be more appropriate to simply clock the core high enough to keep parity with memory bandwidth and vice versa. However, even on texture dependent games like Quake 4, we saw that it won't help AA penalty at all (although you'll likely have higher nominal frame rates). In our case, that's at least around 20 fps or 20 percent (most likely more since we're system limited).An upside to this is that you can compensate for the lack of quads with high enough memory bandwidth, or more precisely, memory clock. Of course, finding a GeForce 7800GT or 7800GS with memory that can scale that high is difficult, but not impossible. So good news for overclockers on a budget or even gamers who don't want to pay and arm and a leg for a GeForce 7800GTX or the elusive GeForce 7800GTX 512.
On shader heavy games, getting the core and memory clock combination right is tricky. It can be done, but it will be a process of trial and error. Of course, if you only care about the nominal frame rate, you can just clock the core and memory as high as you can.
Some Forward Looking Statements
Ah, speculation - the food of tech afficionados and hardware review forums posters everywhere. Well, we chose not to speculate here but rather point out the areas that needs to be addressed by the next generation of graphics cards. First and foremost will be bandwidth. To get an even higher frame rates with texture dependent games, we are in desperate need of significantly more bandwidth. Now, there are at least three alternatives here:- Higher clocks
- Wider bus
- New, faster memory type
Depending on how much bandwidth will be available, more quads (in terms of ROPs) can be used. But we think in this case, NVIDIA should address AA performance instead of using more G70-like quad. Even at it's best, the GeForce 7 series quads still lost so much frame rate when AA is enabled (either pure or with AF). So, that implies a change in architecture. Just how much, we can't tell. With more efficient AA (or even a better implementation), both texture dependent and shader heavy games should be faster with AA and AF.
Of course, we could also, relatively speaking, 'give up' on texture dependent games all together and choose to add more pixel shader units. After all, they are far less sensitive to bandwidth and won't require much of it. At it's worst, it should be at least as fast as the current generation of graphics cards. The higher clock and larger buffers should allow for higher quality filtering and / or antialiasing to be used with less penalty. That's pretty much the road ATI took with the X1900 series.
[Previous Page]
[Go to top]
[Home]