I could write a book on this subject, there are hundreds if not thousands of reasons to benchmark, particular performance in every given application for example
Broader reasons would be-
Countering PR- If we eliminated benchmarks from reviews what would we have? Just eyeballing the specs and games reviewers could well state that the V5 5500 for instance is about equal to a GF DDR and not comparable to the Radeon or GF2. Even with running benches this still seems to be the overall performance consensus, but the problem is that there are not
enough benches run, perhaps some flight sims should be thrown in(Falcon4, IIRC, has a benchmarking tool).
New Features- This one is several issues, how well does something work within a given implementation and if the feature works at all when under extrodinary circumstances(for instance none of the current boards can handle 4x FSAA@1600x1200). How well a new feature works is extremely important, will this board allow this feature to be practical when games/applications start using it(though that is more of a future proof issue)? What kind of performance hit will I be taking if I enable feature X(FSAA, EMBM, Dot3 etc)?
Performance- Performance matters, a great deal to the majority of people, be it for whatever reason it may. Games simply look better at 60FPS then they do at 10FPS, don't think to many people will argue with that, and what could be argued to be a "trivial" boost of 160FPS over 80FPS now, could be the gap between 25FPS and 50FPS in future games. This also plays into bang for the buck. If you could chose either a GF2U or a G400 both for $150, which would you go for? Why? People do care about performance, without benchmarks we have to trust someone we don't know has excellent and very sensitive vision to tell us which card performs better.
Tweaking- I like to get my money's worth out of my hardware, tweaking things to achieve the best possible performance out of
all of my hardware is something I do to ensure that I'm getting everything I paid for. Can I honestly tell the difference between 3%-4% here and there? No, but add them all up and it results in a much smoother system. By utilizing benchmarks I can tweak each setting one at a time until I find the optimal ones. This includes OCing, drivers, Window's settings etc., if they are all set to optimal, you get a much smoother and more stable(knowing when to back that CPU/vid card down) machine.
Future Proof?- I don't care what anyone says, would you buy a board that you knew would not play
any game that was upcoming? No, you would be a fool to, how futureproof a video card is is important to a great deal of people. Take hardware T&L, we know how it works, how it performs in games, and it is becoming increasingly important. Without benchmarks, we wouldn't know that this would have worked at all, but it does quite well. How
useful it is for the average user is another matter entirely, the matter of how well the hardware works is what is the most important thing. EMBM, Dot3 are other examples, how well does hardware handle these features which are becoming increasingly popular(judging by previews anyway)? FSAA is one but the other way around, how much longer will the current boards be able to push new games with this feature on? Perhaps another six months, maybe less(with the exception of the V5 6K and the GF2U). That is based on benchmark performance in current games and average fillrate requirement increases.
The list goes on and on, not benchmarking is the foolish way to go IMHO. You want to trust every single thing hardware companies, or reviewers we don't know, tell us? That would lead to some truly nasty behind the scene going ons
Facebook
Twitter