Tech-Hounds.com

Remember that F.E.A.R is a shader heavy game, so the factor here may not be ROP units inside each quad, but pixel shaders units instead. All this seem to indicate that we're core limited (or at least pixel shader limited) and not bandwidth limited in F.E.A.R. It's very likely that increasing clock on both core and memory clock should provide more tangible gains than with Quake 4. That's the only thing we could do, since we can't add more pixel shader units. Let's take a look. Below you'll see the graphs with core and memory clock changes, in both default (no AA, trilinear) and with 4x AA and 16x AF. All the numbers below are in frame rates.


If you look at the results with no AA and AF, its clear we are core limited. Having more bandwidth won't give you higher frame rates, so the only thing we can do in this situation is use higher core clocks. This is true regardless of the number of quads. In this situation, having more quads will only give you less than 10 percent improvement (around 5 - 10 fps).

With AA and AF, its a slightly different situation. Only with 6 quads did we see any significant improvement with higher memory clocks. Why is that? We know that AA and AF penalty are pretty much the same with 4, 5 and 6 quads. Well, the answer is actually quite simple - fillrate - or more precisely pixel shader fill rate. Remember, 6 quads means a total of 48 shader units on the GeForce 7, while 4 and 5 quads mean 16 and 32 shader units, respectively. Those extra shader units seem to appreciate the extra bandwidth, if present. And AA does need lots of fillrate. In texture dependent games, ROP units will be more influential here, but F.E.A.R is a shader heavy game. That's why it really benefits from the presence of extra pixel shader units.

[Previous Page]
[Go to top]
[Next Page]