.
More shader use means the graphics card must be fast enough at processing shaders. There are two approach this problem - run the card (core) faster or process more shaders per cycle. NVIDIA chose to use a higher core clock for the GeForce 7900, keeping pretty much the same shader pipeline found in the GeForce 7800 series. Benchmarking shaders (in isolation, without the influence of bandwidth or even depth buffer) is very hard to do - you have to rely on a purely synthetic test since shaders used in current games are still very short and not very complex. So, while we will briefly touch this issue, we won't focus on future shaders performance on the GeForce 7900.
Higher resolution is easy enough - most reviews provide benchmarks from 1024 x 768 and up to 1600 x 1200 (or higher). When using a higher resolution frame buffer, the card must draw more pixels and process depth information more efficiently (in the depth buffer). Higher resolution textures will not only take more space (even when compressed) but also take more bandwidth (or latency due to decompression before processing). This is one area is definitely quite easy to test. The card must have a high fillrate, more efficient depth buffer management and memory bandwidth. There is also antialiasing and anisotropic filtering to consider. Antialiasing is usually related to fillrate, while anisotropic filtering is pretty much related to memory bandwidth.
While effects are related to shaders, we will focus more on their bandwidth related aspects. Normal / parallax mapping puts more strain on bandwidth because these effects make use of normal maps (and an additional height map for parallax mapping). Dynamic lighting can be done in a variety of ways, but usually involve rendering the scene from the light's perspective and translating the information to determine which part is lit or inside the shadows. Naturally, this falls under fillrate, just like HDR.
Now, if we were using synthetic benchmarks, all we have to is run one with normal / parallax mapping, dynamic lighting done through shaders at various resolutions ranging from 1024 x 768 to 1600 x 1200. Then we examine the fillrate and bandwidth test to see whether the graphics card have enough of both. But of course, we won't do that. This time, we will examine efficiency, not just pure scores. By looking at efficiency, we can see just how the GeForce 7900GTX fare under certain conditions like high fillrate and bandwidth demand. Why? Simple really, the GeForce 7900 (in fact, the entire GeForce 6 and 7 series) seems to be very sensitive to memory bandwidth.
[Previous Page]
[Go to top]
[Next Page]
The Problem
Let's make one thing absolutely clear - there's no way we can be 100% sure how the GeForce 7900GTX will behave in future games. What we can do here is make a calculated guess. Now let's look at the facts:- Both developers and hardware vendors seem to agree that the next generation of games and graphics engines will likely be more shader than texture dependent.
- Higher resolution will also be likely be the trend, both in gaming resolution and art (textures).
- We will also probably see more games making use effects such normal / parallax mapping as well as dynamic lighting and HDR rendering.
More shader use means the graphics card must be fast enough at processing shaders. There are two approach this problem - run the card (core) faster or process more shaders per cycle. NVIDIA chose to use a higher core clock for the GeForce 7900, keeping pretty much the same shader pipeline found in the GeForce 7800 series. Benchmarking shaders (in isolation, without the influence of bandwidth or even depth buffer) is very hard to do - you have to rely on a purely synthetic test since shaders used in current games are still very short and not very complex. So, while we will briefly touch this issue, we won't focus on future shaders performance on the GeForce 7900.
Higher resolution is easy enough - most reviews provide benchmarks from 1024 x 768 and up to 1600 x 1200 (or higher). When using a higher resolution frame buffer, the card must draw more pixels and process depth information more efficiently (in the depth buffer). Higher resolution textures will not only take more space (even when compressed) but also take more bandwidth (or latency due to decompression before processing). This is one area is definitely quite easy to test. The card must have a high fillrate, more efficient depth buffer management and memory bandwidth. There is also antialiasing and anisotropic filtering to consider. Antialiasing is usually related to fillrate, while anisotropic filtering is pretty much related to memory bandwidth.
While effects are related to shaders, we will focus more on their bandwidth related aspects. Normal / parallax mapping puts more strain on bandwidth because these effects make use of normal maps (and an additional height map for parallax mapping). Dynamic lighting can be done in a variety of ways, but usually involve rendering the scene from the light's perspective and translating the information to determine which part is lit or inside the shadows. Naturally, this falls under fillrate, just like HDR.
Now, if we were using synthetic benchmarks, all we have to is run one with normal / parallax mapping, dynamic lighting done through shaders at various resolutions ranging from 1024 x 768 to 1600 x 1200. Then we examine the fillrate and bandwidth test to see whether the graphics card have enough of both. But of course, we won't do that. This time, we will examine efficiency, not just pure scores. By looking at efficiency, we can see just how the GeForce 7900GTX fare under certain conditions like high fillrate and bandwidth demand. Why? Simple really, the GeForce 7900 (in fact, the entire GeForce 6 and 7 series) seems to be very sensitive to memory bandwidth.
[Previous Page]
[Go to top]
[Next Page]