Is this the end of Synthetic Benchmarks?

[StealthHawk]

Originally posted by Hanners
Actually, if I remember correctly the entire R300 pipeline supports FP32 (128-bit), apart from the Pixel Shaders which are FP24 (96-bit).

Something like that.

[Morrow]

Originally posted by Hanners
At the end of the day, nobody is cheating in the sense of what precision they are using, all that's happened is that both IHVs have chosen different strategies for their architectures, both of which have their good and bad points.

Actually that is exactly the same thing I'm thinking about the current "which-card-is-the-best debacle".

We arrived at a point where both leading architectures (R3x0 and nv3x) are no longer anyhow comparable. They seem to do the same thing in the end but they achieve it in two completely different ways much like Intel Itanium's 64-bit architecture compared to the AMD 64-bit Hammer technology.

It's no longer possible to compare them in a fair manner (same thing for Athlons an P4s). Both architectures have there advantages and disadvantages and it all depends on the task whether one or the other architecture will perform better.

This leaves us now with two conclusions.
* 3dmark03 can no longer be used to objectively measure 3d performance
* everybody needs to choose their hardware according to how well the games/programs they want to run perform on it, which basically leaves us with wait for Doom3, DX2 and HL2 to decide which card you want to buy...

I believe this is a very favourable situation for us customers, which didn't exist for a long time. We now have 2 strong performer on the market which will fight each other for the benefit of the buyer.

[StealthHawk]

Before you make the argument that ATI is "cheating" you will have to show me the IQ differences between FP32 and FP24.

Pictures (http://www.beyond3d.com/forum/viewtopic.php?t=6042). There you can see FP16 on NV35, FP32 on NV35, and FP24 on R350. Whereas FP16 stands out, FP24 and FP32 look the same to me. The water level is different, that is the only change I see. I don't see any more graininess or banding in FP24 than in FP32, do you?


As a side note, are you really trying to say that FP24 is out of spec for DX9? That's utterly ridiculous. There have been discussion about what the minimum spec is, I'll look later. Regardless, this has been defined by Microsoft, I'm sure if you look for it, you will find where it says FP24 minimum.

[Morrow]

Originally posted by StealthHawk
As a side note, are you really trying to say that FP24 is out of spec for DX9? That's utterly ridiculous.

Actually it was not me who said that but it was JC, I only quoted him...

And no, I'm not calling ATI cheaters, I was just point out that they are currently using a lower precision than nvidia. Anyway, there's no doubt about the higher IQ of FP32 but it will be difficult to demonstrate it by screenshots, much like 32-bit colors only offer superior IQ in scenes which a smooth color gradient (like a sky) or fog scenes, compared to 16-bit colors...

[StealthHawk]

Originally posted by Morrow
Actually it was not me who said that but it was JC, I only quoted him...

Well he's wrong. No one is perfect.

And no, I'm not calling ATI cheaters, I was just point out that they are currently using a lower precision than nvidia. Anyway, there's no doubt about the higher IQ of FP32 but it will be difficult to demonstrate it by screenshots, much like 32-bit colors only offer superior IQ in scenes which a smooth color gradient (like a sky) or fog scenes, compared to 16-bit colors...

People have posted shots from the UT2003 menu of all things that showed vast differences between 16bit and 32bit color. Like I've said, 32bit color in today's games is pretty easy to distinguish from 16bit color, now that all art is made with 32bit color in mind. Fire up any recent game, and most effects show an improvement in 32bit color over 16bit color.