Slashdot Mirror
Does 3DMark Predict Game Performance?
aendeuryu writes "Sidhian has up an interesting op-ed and summary piece on whether or not graphics benchmarks like 3dMark are helpful in determining the suitability of a video card for gameplay. It goes on to show how certain tests correlate to specific cards and game engines. The article is from last month, but Sidhian's pretty good for refreshingly frank hardware articles, and this might be good background information for evaluating ATI's upcoming Crossfire line of dual-izable video cards."
Obviously benchmarks try to measure as much as possible, but there are limits. The most significant limit is that the benchmarks simply can't emulate every aspect of every game.
We've got NVidia releasing their new card, but ATI is planning their next one soon too.
However, more horsepower is always good and this year's chips are faster than last year's, so unless you're trying to differentiate between a few microseconds difference in processing time between equivalent competing cards, you're better off just looking at the grand scheme of things and just realize that things are getting better and faster (and more expensive).
the author rightly points out that the benchmark may not be accurately testing the right things. he lists different criteria for different types of games, using FPSs and racing games as the extreme examples. However, he doesn't really explain why the hair's breadth of a difference observed between nVidia and ATI in the benchmark translates to the very dramatic differences in real world performance and more importantly, he never really fully addresses the issue of different games (as the only directX game examined was HL2). If you're going to mention racing games and GTA, why not test GTA? A much more convincing article would have looked at several games spanning the genres (and graphical demands).
I prefer the use of actual games when benchmarking a card. The synthetic benchmarks from 3dMark do not tell me anything other than how the card will perform in 3Dmark. With various game engines using a wide range of features such as OpenGL, DirectX, SM3.0, HDR, etc. you need to benchmark by using a wide range of "actual" games.
p age=1&pp=20
Hardocp dropped 3Dmark benchmarks along time ago for this reason (among others)
http://www.hardforum.com/showthread.php?t=816850&
http://www.kubuntu.org/
That probably has more to do with THG than benchmarks. But I digress..