The beyond3d review of the nv40 was good, but questions

[Fast_3dcard]

Hi there, I am new and I finished reading the nv40 review earlier today. Where do I find the 3dmark 2001 and 3dmark 2003 benchmarks, along with the details of each game test? I hear rumors of 12k in 2003 and over 30k in 2001 with the nv40 but cant find it in the review nor anywhere to confirm!

 

[991060]

Simply put, there's no 3dmark bench in B3D's review for a long time and we are quite happy with that.

 

[christoph]

plenty of other sites offer those numbers already, like here
http://www.xbitlabs.com/articles/video/display/nv40_44.html

 

[Natoma]

Hi there, I am new and I finished reading the nv40 review earlier today. Where do I find the 3dmark 2001 and 3dmark 2003 benchmarks, along with the details of each game test? I hear rumors of 12k in 2003 and over 30k in 2001 with the nv40 but cant find it in the review nor anywhere to confirm!

One of the reasons it's not used here I believe is because of the cheats that Nvidia implemented in its prior drivers which skewed the results of 3DMark03 too much. The other reason being that the overall "score" is not necessarily a good indicator when you're trying to dissect particular portions of a card's architecture. However, certain portions of the test such as GT4 or the Elephant Shader test can stress particular portions of the architecture in order to come to a conclusion.

 

[Guden Oden]

I can't make heads nor tails of the graphs in the B3D preview; there's no labelling on the X axis, and while one can assume it's supposed to be resolutions, the lowest fpses are listed on the left-hand side and flatten out towards the right, which is opposite of how it typically looks in review graphs...

 

[Bambers]

Actually they're fillrate graphs (as is labled on the y axis)


the x axis is the total number of pixels on the screen.


640x480 typically has lower fillrate than 1600x1200 because the card is limited by the cpu.

A graph with a linear increase of fillrate vs total pixels is a game that i completely cpu limited.

A graph that is horizontal shows a game that is limited by the graphics cards pixel rendering.

The fillrate graph can also point out oddities in a card behaviour. For example some cards (some of the radeons do this) actually drop fillrate for the highest resolution. This actually shows where the Hierachical Z gets disabled becuase the card is running out of vid memory.

 

[Fast_3dcard]

so the 12k in 2003 is a cheat score and nowhere near as high? I am still curious about 2001, cheat or not. I know its not an accurate indicator but wouldnt it be something insane like 30k?

 

[991060]

Futuremark approved the driver used in the test, so far so good. You can compare the delta between NV40 and NV38 in 3dmark and games, they're quite in line with eachother mostly.

 

[Colourless]

Beyond3D uses fillrate graphs. The X axis is Resolution and the Y axis is the Fillrate achieved (FPS * Resolution). What does does is let you easily see if you have hit both CPU and GPU limited scenarios by looking at the slope and shape of the line. If the slope of the line is 0 then the benchmark was GPU (fillrate) limited. If the line is straight and diagonal then it's likely CPU or vertex limited. A curved line implies a mix between the 2.

While a normal Resolution vs FPS graph lets you see if you are CPU/vertex limited it's difficult to tell if you are fillrate limited.

 

[Bambers]

so the 12k in 2003 is a cheat score and nowhere near as high? I am still curious about 2001, cheat or not. I know its not an accurate indicator but wouldnt it be something insane like 30k?

nope it wouldn't. nv40 gets about 22000-24000 on 2001 (the latter on a very fast system). This is because 2001 is very cpu/system limited now and only the nature test tends to show any real gains with a faster graphics card.

 

[Fast_3dcard]

then what would that thing get in nature stock?

 

[ANova]

rthdribl is a decent program that can test both the performance of the card and it's implementation of DX9 shaders through the use of HDR.

At this point in time it seems rather obvious though that the NV40 does indeed deliver on support for PS and VS2.0/3.0 shaders unlike the NV30, and it's performance is definitely an improvement on the whole as well.

 

[Reverend]

Hey Nick, how close are you to finishing up my benchmark article (that'll explain just wtf our graphs mean)?

 

[Pete]

Hey Nick, how close are you to finishing up my benchmark article (that'll explain just wtf our graphs mean)?

Am I missing something here, or is Rev asking Nick to finish an article to be published under Rev's name? Pssst, Rev, on the internet, everyone can hear you post.

J/K! I'm sure this is related to editing or some such exciting activity.

 

[Simon F]

I can't make heads nor tails of the graphs in the B3D preview; there's no labelling on the X axis, and while one can assume it's supposed to be resolutions, the lowest fpses are listed on the left-hand side and flatten out towards the right, which is opposite of how it typically looks in review graphs...

Those style of graphs were my invention: the idea is to show the transistion from being CPU/Triangle setup limited to being fill-rate limited.

 

[Guden Oden]

Beyond3D uses fillrate graphs.

Thanks for the explanations, but why isn't the X axis labelled, and why aren't these graphs explained in the text itself (well, I didn't SEE an explanation, and I don't think I glossed over the article either, but I might have missed it).

Also, the term 'swizzling' isn't explained at all really when it comes up (in one of the first couple pages), the article just says it has something to do with how stuff is arranged in memory without going into any kind of detail. I don't think it's right to assume people reading B3D articles to know these sort of things already...