A bad trend?

Well that changes from fairies.....

 

It would be because I didn't notice it till you mentioned it

 

The synthetic benchmarks and test showed the truth back in the days of the 9700PRO arival. All that bash the synthetic test need to take a new look at it and see from bak then to now and see how true the test were.

 

The synthetic tests systematic bashing started with 3dmark 2003 dismiss by nVIDIA if I remember correctly ? Even though other tests already showed similar figures. I mean, we knew even before 3dmark2003 launch that nV30 had problems. The HDR demo, rthdribl, had already shown poor image quality and poor performance on nV30.

 

What would be interesting to learn (after reading that thread Brent linked) is what nVidia's developer support tells game developers to code for when using FX boards. What PS/VS model they recommend for that particular generation.

 

Clearly, this would be interresting. For the nV3x, they recommended FX12 and PS 1.1/1.4.... If I remember correctly....

 

SW adoption always lags HW adoption. Once PS/VS3.0 cards become standard, we'll see heavy use of PS/VS2.0. And we probably won't see heavy use of PS/VS3.0 until DX10 cards are here (all IMO).

 

tbh I remember that synthetics being bashed by 3dfx fans when it was 3dmark2000 and Kyro fans when it was 3dmark2001 etc..

 

That's true, I was among the 3dfx fans at the time

 

mabye you can discuss this with your boss and get him to change his stance on the value of synthetic tests such as 3dmark. granted it would be compleatly uncharacteristic of him to change his stance on anything regardless of how much evedence is in his face, but it is still a noble cause.

 

That was for a different reason, though - 3dfx and PowerVR were hurt badly in synthetics by the total absense of hardware T&L, but it really didn't show in actual games. (the Voodoo5 5500 would often score lower than GeForce256 in 3DMark2000)

 

ya, considering software t&l was a perfectly viable alternative there was really no excuse for the way futuremark handled that.

 

Absence of "L" yes (with rare expeptions like MDK2 f.e.), "T" no

Was 3dmark2001 for instance really aimed to predict in relative terms performance for games in 2001? It really comes down to as to how someone really wants to interpret the results of a synthetic application that aims to predict relative game performance ahead of it's time.

Is software T&L on the other hand really a viable sollution for UT2k3? Apart from any possible objections it is a game with pure T&L optimized code.

 

Silly... ^^

No, software T&L destroys performance for UT2003... but keep in mind, that when we say software T&L was a viable alternative, we're pretty much referring to the 2000-2001 timeframe, where the only game in which it truly made a difference was MDK2...

 

well i can't agree with you there Tagrineth as futuremark's goal to be repersentitive of future gameing performace, hence the name and all.


that said, a voodoo5 runs ut2003 reasonably well for as old as a card as it is and i doubt a geforce2gts does a whole lot better. however, 3dmark2000 required hardware t&l just to run some tests and therby negating the fact that t&l can be accompished though software.

 

Isn't 3dmark03 sort of unfair to none ps2.0 capable cards then? Many of the tests can't be run without a ps2.0 capable card, and it puts a lot more weight on shader performance than any game available when 03 was new, and possibly now as well.(or at least I don't think a radeon 9700 pro is 4 or 5x as fast a geforce 4 ti 4600 outside of pixel shading)

 

i don't see why. if they made it where a chip wouldn't be aloud to run a test simply because it ofloads vertex shaders onto the cpu, now that would be unfair.

 

Actually, they always said that 3DMark03 isn't a fair judge of pre-DX9 hardware, and that 3DMark2001 should be used instead for older cards.

 

[quote="Tagrineth]Actually, they always said that 3DMark03 isn't a fair judge of pre-DX9 hardware, and that 3DMark2001 should be used instead for older cards. [/quote]

so its Ok to for 3dmark03 to not run tests on pre DX9 hardware and score it badly, but it wasnt ok to score an effectivley DX6 card (the V5) badly on a Dx7 test?

oh and the only test not run on a V5 in 3dmark2000 was the environmental bumpmapping test, which is nothing to do with hardware t&l, it was a a pure lack of hardware support.

this horse is sooo dead its fossilised