Lies, Damn Lies and Benchmarks

This article is more interesting than it first looks. They actually played the games, rather than run time demos as all reviewers seem to do these days.


I then set about trying to compare what other reviews say you should get and found this article.


Both using 7800GTX and A64 3500+ (was suprisingly hard to find someone that had used roughly same config).

Basically the actual game logic will knock over 10 FPS off the skyhigh numbers of the timedemo. Something I have always suspected, but never seen proven. More disturbing is the enormous boost this tends to give GFX manufacturers in marginalizing the CPU. After all, playing back a simple demo is a long way off having to work everything else out.

Consistency is much underrated by the current system. I remember John Carmack saying that Quake was all about finding an algorythm that didn't explode when lots happened. No point in rendering at 100fps, when it drops to 3fps when someone walks on the screen.

Cheap machines are usually rubbish because of this. Click on a folder, it takes an age to open, because it has too little memory. Run a benchmark, and its only 10% slower than the big expensive one you could have bought. Subjectively this is a total lie.

It is vital to have balance. These days memory has become important again - 512M is getting quite borderline for most things. Spend the extra 30 quid on a bit more.

Moral of the story: a fast CPU is better investment that it has been reported. And don't believe benchmarks. They tend to be wrong in most of the ways that count.

Comments

Popular Posts