3DMark Vantage vs. Games
Individual GamesThe first graph was all about averages - average framerates from a diverse set of games compared to 3DMark Vantage, and the results lined up pretty well. Yet if you compare 3DMark Vantage to specific games, they don't always agree - in extreme cases you can have situations where a GeForce 9800 GX2 is faster than a GeForce GTX 280 in a specific game, even if 3DMark Vantage (and the average results from a wide set of games) tell you that GeForce GTX 280 will outperform the 9800 GX2 by a considerable margin. Reasons for this are numerous, but the main point is that with individual games you have a large set of variables that can affect the score - the type of game engine used, the individual bottlenecks in the game code that hold back the frame rate, any game-specific driver issues and numerous other factors. In some cases the CPU is still a deciding factor, and there is hardly any difference between wildly varying video hardware when benchmarked using the same CPU. When you toss in SLI and Crossfire to the mix, it gets even more complicated, as they sometimes don't scale predictably even in games that are completely limited by the GPU.
(for a more complete data set, see the Radeon HD4870 review by Tom's Hardware)
In this example, a GeForce GTX 280 dominates the competition in UT3 and Race Driver: GRID, yet in Flight Simulator X, a GeForce 9800 GX2 outruns every other card. In Mass Effect, the choice of card doesn't matter - the game is frame rate locked to 60fps. Flight Simulator X is a nice example of a game that is not limited by the overall performance of the GPU (World of Warcraft and Everquest 2 are other common examples), and the pecking order gets mixed up by details completely unrelated to the raw pixel-pushing power of each card. Instead, factors that usually don't really matter, like CPU usage of the driver, specific GPU bottlenecks or the streaming performance of the card (shifting data from hard disk to RAM and from RAM to card "on the fly") come into play. You can also see that while older multi-chip cards (GeForce 9800 GX2 and Radeon HD3870 X2) can give the latest offerings a fair fight in some games, at times they have major issues. In Race Driver: GRID both multi-GPU cards lose half of the theoretical performance due to problems getting the game engine, drivers and the complex multi-GPU hardware to cooperate. Sometimes you run into subtle differences between the individual multi-chip implementation from ATI and NVIDIA - Unreal Tournament 3 seems to be able to exploit multi-chip hardware from ATI very well as HD3870 X2 can beat the (on average) faster Radeon HD4870 in this individual case. At the same time, on the NVIDIA side the situation is reversed as GeForce 9800 GX2 can't really compete with GeForce GTX 280 in UT3. 3DMark Vantage is just one forward-looking engine, being used in two different gaming scenarios. When trying to figure out the performance in specific individual cases there is great value in benchmarking those specific cases directly. On the other hand, when all you need is a simple answer to a question "is card X better than card Y?", you can see that 3DMark is capable of quickly producing an answer that is very close to the average result you would get from an exhaustive set of game benchmarks. For the current answer to "Which card is fastest?", you can check the ORB Hardware List - it's based on all the 3DMark Vantage benchmark scores submitted, worldwide.
If you have any comments on this article, click right here (registration not required).
Tags |
|||||||||||||||||||||||||||||
 |
