There is hope for Anandtech when comparing theoretical numbers with real world games

Agent_N

Gawd
Joined
Aug 20, 2004
Messages
862
Page 1, third paragraph down.

"And now we have the other thing we've been working on since we finished GT200: RV770 in all it's glory. This includes the 4850 whose performance we have already seen and the Radeon HD 4870: the teraflop card that falls further short of hitting its theoretical performance than NVIDIA did with GT200. But theoretical performance isn't reality, and nothing can be done if every instruction is a multiply-add or combination of a multiply-add and a multiply, so while marketing loves to trot out big numbers we quite prefer real-world testing with games people will actually play on this hardware."

http://www.anandtech.com/video/showdoc.aspx?i=3341
 
I may be more inclined to believe that a bit more if they did not follow it up with pages and pages of single number result benchmark graphs, along with commentary that simply talks about those numbers.
 
I guess you could say I'm calling them on what I perceive to be a wee bit of hypocrisy :)

"This shows us that NVIDIA's architecutre requires more than 2x the die area of AMD's in order to achieve the same level of peak theoretical performance. Of course theoretical performance doesn't mean everything, especially in light of our previous discussion on extracting parallelism" (page 7).
 
Well nvidia has a different architecture, and they have more TMUs and ROPs than the ATi equivalence, which don't add to the theoretical flop number.
 
I guess you could say I'm calling them on what I perceive to be a wee bit of hypocrisy :)

"This shows us that NVIDIA's architecutre requires more than 2x the die area of AMD's in order to achieve the same level of peak theoretical performance. Of course theoretical performance doesn't mean everything, especially in light of our previous discussion on extracting parallelism" (page 7).

"We know that all these benchmarks we just showed you do not translate to real world gaming performance....but we are not going to take time to show you that, so instead, here are 23 pages of coverage, enjoy the ads X 23."

We do our best to keep you guys below 10 pages....
 
Well the thing I like about Anandtech is that they go into more detail about the architecture itself outside of just performance numbers so that you can really understand how the new architecture works. They also test several different games which I would prefer to see more of instead of just thorough testing in only 3 games since a GPU architecture can perform wildly different from game to game. I also look for a review that covers the games I'm personally interested in even playing such as Oblivion. Everyone is going to have their favorite games of course so you can't cater to everyone but in limiting the number of games you test you also limit the number of people that truly gain anything from your benchmark numbers. If the card for example is faster in Crysis that means nothing to me if it's slower in Oblivion.
 
Well the thing I like about Anandtech is that they go into more detail about the architecture itself outside of just performance numbers so that you can really understand how the new architecture works. They also test several different games which I would prefer to see more of instead of just thorough testing in only 3 games since a GPU architecture can perform wildly different from game to game. I also look for a review that covers the games I'm personally interested in even playing such as Oblivion. Everyone is going to have their favorite games of course so you can't cater to everyone but in limiting the number of games you test you also limit the number of people that truly gain anything from your benchmark numbers. If the card for example is faster in Crysis that means nothing to me if it's slower in Oblivion.


Oblivion is old as hell, and you should be able to max it out with any card worth it's salt. There are only a few GPU limited games out right now, the others are more CPU and RAM limited which end up clumping together anyway. The architecture is not all that new either, with ATi anyway. I would however like to see some lower resolution apples to apples here on [H], at least the popular 1680*1050.
 
Well the thing I like about Anandtech is that they go into more detail about the architecture itself outside of just performance numbers so that you can really understand how the new architecture works. They also test several different games which I would prefer to see more of instead of just thorough testing in only 3 games since a GPU architecture can perform wildly different from game to game. I also look for a review that covers the games I'm personally interested in even playing such as Oblivion. Everyone is going to have their favorite games of course so you can't cater to everyone but in limiting the number of games you test you also limit the number of people that truly gain anything from your benchmark numbers. If the card for example is faster in Crysis that means nothing to me if it's slower in Oblivion.


Yes, they do go into more detail, but we find that less and less people truly want that level of detail. Looking at page views prove this to us. We are focusing on what our readers want more of. Regurgitated specs are not what H readers are asking for.

Again, I will suggest that the three games we used to base our 4800 conclusions on worked out just fine. Do you suggest that more games would have changed our conclusion on the 4800 series?

And yes, you are right, if all you play is Oblivion. We did start covering Oblivion in 2006. I don't think suggesting that focusing on games from 2006 is going to help us determine usages that are important today.

http://www.hardocp.com/article.html?art=MTE4Myw2LCxoZW50aHVzaWFzdA==

The fact is that a 9600 GT is going to play Oblivion on just about any screen except 2560 resolution. And if that is your setup, then I suggest spending more money on your video card for your 30" monitor.

So all in all, I see your statements not meaning much. We did not show Quake 3 either, but the fact is that it is irrelevant when looking at this generation of GPUs.
 
For a guy like me who is average/slightly above in terms of GPU knowledge & understanding, Anandtech's 5 pages of architecture dissertation were pointless. It means nothing to me really, until they tell me about a new functionality or performance optimization that the architecture allows.
R600 is exhibit A. Huge on paper. (relatively) Small on gaming perfomance.
 
Back
Top