What does the 3DMark Score Represent?

I have no statistical data about the upgrade cycle of the average 3DMark user, but I assume 2/3 years would be rather the exception than the rule.

So all other things aside, if 3DMark could give an realistic outlook of the performance of future games it would be appreciated by a lot of people.

 

The Radeon 7500 says it has VS1.1 in its caps. Even if they are a software implementation, they are a lot faster than running in DX9 software mode (up to about 10 times faster for the simple shaders in the DX9 samples). This would explain the speed difference. The multitexture fill rate, as measured by 3DMark2001 is the same for the Ti with two textures and the 7500 with three. For single texturing and dual texturing, the 7500 is significantly slower.

 

2/3 years = 2 to 3 years.

Knowing how often do 3DMark03 users upgrade their PCs would indeed, be usefull to see how valuable 3DMark03 is for them.

The younger will prolly like to say 'my 3DMark03 score is bigger than yours', but that's not enough to make it valuable

 

Apologies for the misunderstanding.

 

What does the 3dMark score represent? Penis size.

 

I think it's quite the contrary.

People showing off with their 3dMark score are certainly compensating...

 

Heheheh. . .

 

Surely, we can agree what FutureMark is trying to do is weigh scores by overall "goodness"- but the debate then focuses on- By who's measuring stick?.

The problems arise when a deemed feature (be it by 7% or 33% or whatever) is given weight in "goodness" yet pans out to have 0% "goodness" for all of eternity.

A good example is 3dmark2000, where "goodness" quotient runneth over with fixed function HW T&L, which to this day still does not attribute performance, especially in light of how the benchmark was coded.

How the benchmark is coded also brings to the forefront the effort and time required to implement a feature in a sane and reasonable way for a game engine versus a very fixed and static benchmark demo. Obviously, if a programmer is determined to spend 2 years coding a static path of 2-3 minutes of a scene, they may be more inclined to use a feature in such a way that would be unrealistic for a game engine, from which all graphics coding, AI, content, sound/music and artistry is only on a 6 to 14 month schedule. You have 10x the amount of detail, dynamics and other conditions that are totally different from the needs of writing a static, fixed path demo. We see this, again, with 3dmark2000 (and to some degree 3dmark2001) where fixed function, 2-vertex straining is performed for 95% of the scene geometry as the benchmark is static/small and this is sane for the time frame, where such an equivalent effort would take 5+ years for a dynamic game engine.

There is nothing wrong with a benchmark being "forward looking"- but it has to do so within sane and legal boundries. It is not "forward looking" to expect a game developer to spend 5+ years coding a dynamic game engine to utilize features to the extent a fixed-path, static demo may afford. It's also not "forward looking" if the techniques used differ so dramatically from techniques used in dynamic game engines or if features used stand little to no chance of ever being implemented. I believe these are just a few issues with 3DMark compared to the game market.

 

And as Ingenu said, the problem is that game test 2 and 3 are supposed to be somewhat similiar to Doom3 type of rendering. Yet runs 5-10 times slower. (I have the leaked alpha and even in that early state, it runs WAY better then the game tests).

Imo, that's pretty misleading and not really good for the consumers in any way.

 

Well we can find out that in future game the R9700 is 100 times faster than the GF2

 

I'll move some of my thoughts from yesterday into this thread, I guess...I think they are pertinent to the discussion:

 

I'm going to quote myself again, not because I'm trying to insinuate that you didn't read something, but because I'm lazy. Italic for words added for clarity, and some bold emphasis added for addressing your comments:

 

Well, i am a bit lazy (just ask my girlfriend). Didn't feel like quoting your entire post here either.

Anyway, am i right in that what you're saying is that it's ok with 3-5 or more times lower performance then Doom3 since the proportion between the different cards is (might be perhaps) correct ?

 

Hmm...I'm more saying that I'd give two examples of how to do an apples to apples benchmark between, for example, the 8500 and 9700:

• Write code to run acceptably on the 8500 using common functionality and let the 9700 perform really well without breaking a sweat.
• Write code to run acceptably on the 9700 using common functionality and have the 8500 show its limitations.

I'm saying the goal of Doom3 (and other games) is different in that it is not trying to do an apples to apples comparison between different levels of functionality, it is trying to get acceptable performance on each and will reduce the workload for the 8500 where applicable to achieve this. (I'm also indicating that this is why the performance is so low on "DX 8" level cards that are out).

Further, it is my opinion this benchmark behavior is successfully accomplished in 3dmark03 because of the focus on shaders and their evolution going forward, using effects that can't be reasonably done otherwise, while 3dmark 2001 's accomplishment of this goal was much inferior due to various factors.
See my comments I quoted just before my reply to you for a list of those factors, especially the responses to nvidia's commentary which recognize that we might need more information to determine whether 3dmark03 might still have similar problems to 3dmark 2001 as nvidia is stating (currently, it just strikes me as a smear campaign).

 

There's all the reason in the world to use vertex shaders with DX7 texturing. Most per pixel effects need the texture coordinates set in specific ways -- per vertex. Other effects possible with vertex shaders and standard texturing include glow and fur. Even particle systems are most easily handled with vertex shaders.

Since most cards sold today come with reasonably high performance vertex shaders, either hardware or software, doing things using vertex shaders is a practical way for developers to go.

 

Though not strictly on topic can someone tell me why on the tests if you cannot use pixel shader 1.4 and therefore you have to use 1.1-- THEN WHY DO YOU HAVE TO DO MORE POLYGONS as well ?

I just can't grasp that .. is their a technical reason for it ? If that was in a game the game developer would include less polygons to speed it up, no ?

Maybe I am being slow.

I like 3DM03 as a pure benchmarking competition program, but they even manage to stuff that up by putting different cards in the same bracket, ie 9500 and 9700. Poor sods with 9500's!

3dmark2001SE was very good for seeing if your entire system was on par with everybody elses or something was sick with yours and let you correct it. This just sees if your video card is sick...

If games start to use a lot of pixel shaders in the next two years and video cards get advanced enough so that cpu limitations come into play then it will be judged a success. If games go off at a tangent to 3DM03 then it will become irrelevant.

Regards


Andy

 

Ok, well, i agree with you with regards to making a "correct" benchmark. But, as correct as it may be, what will it tell us about performance in future Doom 3 type of games ?