Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
Hello drunkenmaster
It's a good question . . . Hopefully someone has got the answer!
[Off Topic]
11 games tested, average 9FPS advantage to the HD5850 1024MB
The simple answer to that is, the Anandtech benchmark is ruddy awful, nothing more or less.
From quite ridiculous choice of presentation, single game results not grouped together, to mins/averages mixed in with only some games getting minimums, with no clear distinction in the horrible coloured writing.
Basically, why do they choose to do graphs in every review they've done, with a clear format, this is a mess.
Take L4D for instance, its cpu limited, why why you've chosen 4-5 games max do you choose a HEAVILY cpu limited game, at those settings, choose a 5970, or 480gtx sli, or a 480gtx single, these cards are MILES ahead of a 460gtx, but in that game aren't, the 460gtx gets around 110fps, the other cards in various setups for various reasons with differing cpu overhead in sli/xfire get between 120-130fps. Are they 10-15% faster than the 460gtx, no, does Anandtech's benchmarks suggest this, yes.
If you're going to do a quick GPU bench round up, it makes sense to use the toughest games, Crysis, Metro 2033, etc, etc, not cpu limited ones.
Take 4 results, 2 are gpu limited and a 5870 is 50% ahead, take 2 more in cpu limited and they are only 5% ahead, rounded up the 5870 now looks only 27.5% ahead.
Anandtech's GPU bench alone, is completely and utterly NON indicative of final performance.
As for what I said, they do suck, they are 30-40% slower at stock than overclocked. For zero effort, cost, I'll take the performance, if overclocked theres suddenly a 35% performance difference, essentially for free, then the 1GB becomes much the better card.
I'd also say that, honestly, some of the results from the Anandtech GPU bench are a bit dodgey to say the least.
I'd say both the 460gtx and the 5850 are heavily underclocked cards at stock to fit into a price point, and that works for the end user as both can be essentially reclocked to "normal" clocks, and then overclocked even further.
Take any specific 5 games and you can make 480gtx and a 460gtx perform the same, take another 5 and the 480gtx could be 400 quicker(if you use memory limited situations and multiscreen, whatever).
In general I blanket ignore results that are CPU limited, and round ups in any review should be ignored as two sites that have taken to doing roundups, anandtech and techpowerup give VERY different stories if you go from looking at their roundup/benchmark tools results, to actually reading individual results from a review and excluding cpu limited benchmarks.
If you further read several reviews and can pick out anomolous results, like Tom's got xfire on Crysis results basically non working, while almost every other site had it working fine on any driver. Reviews sites can and DO make mistakes, look out for them.
