NVIDIA GeForce4 - NV17 and NV25 Come to Lifeby Anand Lal Shimpi on February 6, 2002 8:51 AM EST
- Posted in
Unreal Performance Test 2002 - Build 856
We've of course saved the best for last. About a week ago we introduced a brand new way of measuring video card performance through the use of the latest build of the current Unreal Engine that will be used in upcoming games such as Unreal Tournament II and Unreal 2. We began using this Unreal Engine benchmark as a method of not only characterizing the performance of today's graphics cards in tomorrow's games, but also as a way of helping developers such as Epic work more closely with the hardware vendors.
After publishing our initial article we asked Epic Games' founder and lead programmer, Tim Sweeney, about his perception of the immediate results of using the Unreal Performance Test 2002:
"Behind the scenes, the benchmark has really helped focus NVIDIA and ATI on next-generation game performance -- and move away from using vintage 1999games like Unreal Tournament and Quake 3 to judge new 3D cards. The improved interaction between our team and the hardware makers is leading to a lot of performance improvements that definitely wouldn't have happened without the benchmark."
Today's GeForce4 review provides an excellent environment to continue the use of the Unreal Performance Test 2002. Daniel Vogel, another talented programmer at Epic, spent a great deal of time putting the finishing touches on a new build of the benchmark that includes some indoor as well as outdoor arenas. We regret that we cannot publish screenshots of the benchmark at Epic's request however there are a number of screenshots of what the engine is capable of at Epic's website already if you're interested in seeing exactly what this engine is capable of.
In the original article on the new benchmark we outlined the type of stresses that this test will provide that previous benchmarks couldn't, for more information please feel free to read the first article.
First of all it's very important to note that the flickering fog issues we complained about on the Radeon 8500 in our original article have been fixed with the latest beta drivers (v7.66) from ATI. At the same time, the performance did decrease with this latest driver revision giving the GeForce3 Ti 500 the lead whereas the Radeon 8500 used to be on top. Whether this is related to the flickering fog fix or not, we are still not sure.
We also through in another statistical point to look at, and that is the GeForce4 clocked at 240/250 (DDR) which is exact operating frequency as the GeForce3 Ti 500. This helps illustrate the performance improvement that is a result of the GeForce4's architectural enhancements and not just the clock speed boosts. Here we can see that a 15% boost in performance can be attributed to the architectural improvements of the GeForce4 alone.
Combining that 15% improvement with the increase in GPU and memory clock speeds results in a 31% performance improvement over the Ti 500 for the GeForce4 Ti 4600. In a next-generation first person shooter, the GeForce4 is able to already run at almost 90 fps without even being optimized for the game.
A significant disappointment is the GeForce4 MX which fails to even outperform the GeForce2 Ti 200. This is exactly why we recommend either going for the GeForce3 Ti 200, the Radeon 8500LE or waiting for the GeForce4 Ti 4200; and this is only at 1024 x 768.
The performance standings don't change at 1280 x 1024 but now it takes no less than a GeForce4 Ti 4400 in order to break the 60 fps barrier. Granted that we're looking at quite possibly the worst case scenario when it comes to this engine but that should make you appreciate the power of the GeForce4 even more. The 4600 is still able to perform very well while the GeForce4 MX cards make it clear that the only thing they share in common is their name.
When we first looked at performance under the UPT2002 we said that 1600 x 1200 was pretty much wishful thinking for the most part, however with the GeForce4 that all changes. Granted that with the 4600 we're not at 60 fps just yet, but we're very close and that is without any significant driver or code optimizations yet as the engine is still not in a completed state.