Now this is interesting

[Mr. Smith]

Benchmarking the benchmarks

Well? What do you think?

 

[Rastalovich]

It is an interesting pov, and certainly to some extent I agree with many of the points.

Mirrored in some of the sentiments is something I been banging on about for quite a while now. A lot of the benchers we use at the moment are pushing 2 years out of date. A great amount of the so-called gpu testers, u can artificially boost ur graphic rating with a beefier cpu - which imo is false for the real world of gaming. In my experience, the gpu does the majority of the work and I see 20-25% of my cpy being used. These testers include running the cpu @ 100% on all cores - it`s unnecessary.

3dMark06 with the highly based cpu tests canceled would be interesting comparisons.

But where I do semi-disagree, is that u can totally dismiss the fixed timedemo type benchmark based around games. I`ve historically liked these, way back to the first opengl unreal timedemo experience I had, and switching hardware etc. And to the same extent, u have to have a simple thing to run that is easy to use in all hardware situations.

I do push for not basing too much credibility behind fixed benchers, but at this moment in time I put it down to them being too old for the tech. I`m sure this will change this year.

 

[Mr. Smith]

That is why I like kemps reviews, a 3min in game run which gives a realistic impression of how the game behaves IRL.

Timedemos need to lose some credibility, as do the old 'benchmarks'.

The only time my cpu seems to be taxed is in RTS such as Sup Com... 90% of games hardly use dual core let alone quad, but people buy quads as they bench well.

I bought my quad as a review I read at the time (Dual vs Quad) showed eff all gains from a quad except in, Lost World iirc that had a 'core patch' which allowed the game to use all the cores. When this was applied the FPS shot up and the dual core was pwnd.

The article indicted this was the future (reference made to the 'quad core optimised code of Crysis' - yeah, right) so I bought.

CPU tests can be in graphics tests but the weighting needs to be less. A lot less.

Anyway, Kemp - keep up the good reviews mate

 

[Rastalovich]

I bought my quad as a review I read at the time (Dual vs Quad) showed eff all gains from a quad except in, Lost World iirc that had a 'core patch' which allowed the game to use all the cores. When this was applied the FPS shot up and the dual core was pwnd.

The article indicted this was the future (reference made to the 'quad core optimised code of Crysis' - yeah, right) so I bought.

I personally feel that the Dual core processors, same scale as the Quad family, with same caches and some single core superpi performance - will cope equally with the same game.

Would also hazard a guess that if a single core processor were available of the same flavor it would do pretty much the same, OS variables aside. It`d have to superpi (for or as an example) the same.

Similarly if u look at the cpu farms Intel and AMD, the AMD variants, even tho not benching as well, can wipe the floor with games just as well. Maybe they`d change my 20-25% usage to 30-40 or something, but they`re certainly not going to be out of their depth or shouldn`t be considered a non gaming choice.

EDIT: This is all `at this moment in time` of course, incase any1 wants to poke me in the eye in 5 years time. (games might come around by then lol)

 

[JN]

Yep [H] have a valid point, and that's why all of our reviews whether they be for memory, graphics cards or motherboards all feature either exclusively use FRAPS (a la Kempez) or a combination of FRAPS, Self-recorded timedemo & game stock timedemo (a la Jimbo).

 

[FarFarAway]

Werd, that is indeed why I use games as the benchmarks, as you can tell exactly how a game will perform in RL