Futuremark: 3DMark06

[Hanners]

uhmmm So its different then the 03 and 05 on that point?

Yep - Whereas 3DMark03 and 05 used the CPU tests to generate a separate score, away from the overall 3DMark total, 3DMark06 incorporates it into the final calculation of your overall score.

 

[digitalwanderer]

1972 onions for Bubbles; SM2.0 1040, CPU 842

Dang it Hanners, you broke EB again!

 

[Hubert]

Question

Do I get it right ? 3DMark 2006 has a SM 3.0 test, namely the Shader particle test, wich makes use of vertex texture fetch (nvidia only, so it is skipped by X1000 cards), but the result of that are not calculeted in the final score ? (Elite Bastards review.)

 

[Hanners]

Do I get it right ? 3DMark 2006 has a SM 3.0 test, namely the Shader particle test, wich makes use of vertex texture fetch (nvidia only, so it is skipped by X1000 cards), but the result of those are not calculeted in the final score ? (Elite Bastards review.)

Yes, that's right, it's one of the additional feature tests that doesn't count towards any of the final scores.

 

[Joe DeFuria]

Do any of the test make use of significant dynamic branching in the pixel shaders? (Whether or not they contribute to the final score?)

 

[Hubert]

FP16 emulation in shader means about 100 marks loss for 7800/6800 GT ... can we assume that X1000 cards have the same loss ?

 

[Hanners]

Do any of the test make use of significant dynamic branching in the pixel shaders? (Whether or not they contribute to the final score?)

Not that I'm aware of, I don't remember it being mentioned at all.

 

[Jawed]

Oh what a surprise.

Jawed

 

[m4trix]

Oh what a surprise.

Jawed

what do you mean?

 

[Hubert]

Do any of the test make use of significant dynamic branching in the pixel shaders? (Whether or not they contribute to the final score?)

Dunno, I just read the Elite Bastards review, a very nice one btw. In the beginning they have a small list of the new features used :

"• vPos Register
• Derivative Instructions
• Dynamic Flow Control
• Large number of interpolators
• Large number of constants
• Large number of instruction slots
• Texture instructions with explicit LODVertex
• Vertex Texture Fetch "

Dynamic Flow control is one of them, isn't that dynamic branching ?

Removed rant meaningful only for myself

 

[Joe DeFuria]

Dynamic Flow control is one of them, isn't that dynamic branching ?

Yes, but is it in pixel or vertex processing?

 

[Dave Baumann]

3DMark06 has multivendor "DST" and "PCF" support.

Can you be more specific about that Nick (I'd like to know for some testing I'm doing at the moment). DST just means "depth stencil texture" is used, whereas PCF means "NVIDIA can do 4 samples in one cycle" on a single format surface; Fetch4 enables ATI to do 4 samples in one cycle from a DST surface as well - do you actually use Fetch4 in the tests where NVIDIA use PCF and if so can they both be turned off?

 

[Geo]

Must give props to www.bjorn3d.com No registration, no rate-limited server, 720KB/s+ the whole way, 576MB in less than 15 minutes. Getting to the download page was a bit slow, but once there it smoked.

Nice, guys.

 

[Joe DeFuria]

This bodes pretty well for the X1900:

http://www.firingsquad.com/hardware/3dmark_06/page7.asp

The x1600XT does very well with the SM 3.0 tests. At a cursory look, the tests do indeed look shader (rather than TMU) bound. Dunno if that will still be the case for the high-end cards, but if so, X1900 should do very well.

I also noticed that X-Fire seems to scale better (much closer to theoretical 2X performance) than SLI.

Would be nice to see Quad SLI scores though.

 

[Rys]

There is branching in the pixelshader in 06 (I can see branch instructions in the shader dump), but (and I'm guessing here without poking it too much) not to improve performance to any reasonable degree.

 

[Hanners]

Can you be more specific about that Nick (I'd like to know for some testing I'm doing at the moment). DST just means "depth stencil texture" is used, whereas PCF means "NVIDIA can do 4 samples in one cycle" on a single format surface; Fetch4 enables ATI to do 4 samples in one cycle from a DST surface as well - do you actually use Fetch4 in the tests where NVIDIA use PCF and if so can they both be turned off?

According to the documentation, either PCF or FETCH4 can be used. I believe the 'Disable Hardware Shadow Mapping' feature turns off both of these if selected.

 

[digitalwanderer]

Did Hanners just very politely tell Dave to RTFM?

 

[Tim]

Note the OC cards

Also seems like the GT and the GTX are about only 120 or so points apart, does that seem right? As I was expecting to see a larger detla between the two...

They are pretty close, but the fact that the CPU-score is a part of the overall score makes them look even closer. Also remember that both the GT and GTX are running at 450Mhz in this review.

 

[Mariner]

From the Firingsquad numbers, the X1600 (which supports Fetch4) seems to have relatively impressive performance compared to competing products. The X1800 (which doesn't support Fetch4) is less impressive in comparison to its competitors.

I wonder how important this support (or lack of it) will be to the 3DMark score - and is this why EB says one of the IHVs isn't happy with the way the software has been written?

 

[dizietsma]

The water effects in Canyon now officially rate an Alan Partridge style "Sooperb!".