3DMark03 - damage control for the GeForce FX?

[ChrisW]

Originally posted by SamuraiCatJB
or tests that use specific math in the shaders that is known to work better on the FX.... there are lots of ways he could leverage the data towards the FX... I hope he doesn't though.

True. They could use Cg compiled shaders that use instructions not supported by the 9700. Then the 9700 would have to make these calculations via the cpu which would also slow it down.

[SamuraiCatJB]

don't get me wrong... there are some valid reasons for changing a benchmark to change dominance, but it should not be hardware related. IF a larger portion of games are using the more advanced shaders, 128bit color, etc, this is the reason to put it in the benchmark. Doom is coming out with the newest generation using many of these things that give the FX dominance.... so there are some valid reasons... but even dreaming that game takes 10% of the nations sales, it's methods should not get more than 10% dominance in the scoring. Even assuming the new Doom engine spreads out to many other games providing 25% market potential... it should not get more than that potential in the bench. Yes those numbers are high, I'm kind of exaggerating a bit, but the point is those are valid reasons to up the bench specifically to one side, but they should not extend beyond the market potential of those reasons. There are very strong DirectX 9 games that use the new shaders (half between Gf4 and Gf-FX), that should also have techniques demonstrated. Those, arguably, will cover most of the advanced gaming market BECAUSE they are not card specific.

Steppy, i think it's been obvious to nvidia since they saw the r9700Pro reviews that their 128bit bus isn't going to cut it. they knew by August 2002 that NV30 would fall short of r9700 for sure. hence the delays and the dustbuster. they definitely know that a 128bit bus will never stand up to R350 or R400.

[UDawg]

3DMark is a waiste of time. I test and bench in games and compare to the hardware I had before on the same set up. I do like to look at 3DMark but that is abou it.

[Steppy]

Originally posted by StealthHawk
Steppy, i think it's been obvious to nvidia since they saw the r9700Pro reviews that their 128bit bus isn't going to cut it. they knew by August 2002 that NV30 would fall short of r9700 for sure. hence the delays and the dustbuster. they definitely know that a 128bit bus will never stand up to R350 or R400.

I'm not saying they didn't know, but being able to switch that pretty late into the dev. cycle, AND release it 4 months earlier than they planned to seems a tad unlikely to me. Especially when an NVx5 core is usually pretty fundamentally the same as a NVx0 core is replaces. If they WERE having so many problems with ramping the .13 NV30 core high, they're suddenly gonna ramp an even more complex .13 core just a few months later? Could it be done, sure and Nvidia's got a better chance than most. Saying it and doing it are two different things though. It just doesn't seem a snap of the fingers and "relatively" guaranteed like so many seem to think.

[Chalnoth]

Originally posted by creedamd
Who wants to bet that the GFFX will kick the 9700 in the new 3dmark! This will be unbelievable if it does! Discuss!!!

Before bothing to read the rest of the replies, here are my two cents:

If FSAA and anisotropic filtering are not used, it only makes sense that the GeForce FX will beat out the Radeon 9700 Pro anyway, since it does so in any current benchmark when neither is used.

The only current question is whether or not the FX's drivers will be improved enough to do this with regards to shader performance.

Originally posted by Steppy
If they WERE having so many problems with ramping the .13 NV30 core high, they're suddenly gonna ramp an even more complex .13 core just a few months later? Could it be done, sure and Nvidia's got a better chance than most. Saying it and doing it are two different things though. It just doesn't seem a snap of the fingers and "relatively" guaranteed like so many seem to think.

i thought NV35 was using the low-k dielectric? oh well, i'm kind of drifting right now, through all the rumors. i thought that was one of the "confirmed" rumors, but maybe i'm smoking some of the good stuff

i think it's pretty clear they either need low-k for high clockspeed, or 256bit bus for more bandwidth, or both, in order to compete.

[Steppy]

I agree, and am sure the NV35 will have those...it's just a matter of when.

[Uttar]

I just love saying again and again what my guess for the NV35 is
This guess is based on combining several rumors, and when any contradictory information existed, the most reliable & recent information was taken.

Also note that this is the high-end specs, and a low-end NV35 ( probably around $299 or $349 ) will probably be released at the same time.

- 625Mhz core using TSMC's 0.13 Low K
- 500Mhz+ GDDR-II with a 256-bit memory bus
- bug fixes ( most bugs gone ) and optimizations ( resulting in very slightly increased per-clock efficiency )

And here's a few things which are really nothing more than guesses, and pretty much come straight from my wistlist:

- CineFX 2.0. ( including *static* branching in the PS, thus there's only Dynamic Branching left to support to get PS3.0. support )
- Intellisampling 2.0. ( including AA Gamma Correction, better AA patterns and an intermediary ways to do Aniso )


The NV35 will crush the R350, but be in stores about 2 months after it, too. The R400 should also come a few months after the NV35. It's expected to retake the performance crown, but nothing is certain just yet.


Uttar

[CherryPopper]

Also note that this is the high-end specs, and a low-end NV35 ( probably around $299 or $349 ) will probably be released at the same time.

- 625Mhz core using TSMC's 0.13 Low K
- 500Mhz+ GDDR-II with a 256-bit memory bus
- bug fixes ( most bugs gone ) and optimizations ( resulting in very slightly increased per-clock efficiency )

personally i'll be very surprized if Nvidia is able to get a workable 256 bit bus for the NV35. I think it's more likely this will be seen in the NV 40.

[digitalwanderer]

Originally posted by Uttar
The NV35 will crush the R350, but be in stores about 2 months after it, too.

That's if nVidia gets it out the door on time...which I ain't counting on.

I think ATI is gonna keep the crown 'til at least Q1-Q2 2004.

[Uttar]

Two things:

1. Even if the NV35 is not on time, it'll still be slightly before the R400
2. The 256-bit memory interface is one the most reliable piece of information we got on the NV35, another one which we can also be sure of is it's with Low K. The NV35 *will* be on a 256-bit memory interface: it's a fact

BTW, the less certain piece of information I said about the NV35 is the 625Mhz core. It could be easily less ( even significantly less, who knows... ) , but not more.

Uttar