Question about PS1.4 hardware

Any card that also supports Pixel Shader 2.0. In other words the current list is:

8500
9000
9100
9500
9500 Pro
9700
9700 Pro

Geforce FX
 
So other then ATI the only other card that can support PS1.4 is GeForceFX via PS2.0? What about Matrox, VIA/S3 or 3DLabs?

huh? Now I have two more questions then.
1) Why is 3DMark03 not using the PS1.4 support via PS2.0 in its tests.
2) Why are they even using PS1.4 since only one hardware company supports it?

Not trying to start a fight or anything just trying to understand how a benchmark can be "neutral" when it is written only for one brand of card.
 
James said:
So other then ATI the only other card that can support PS1.4 is GeForceFX via PS2.0? What about Matrox, VIA/S3 or 3DLabs?

huh? Now I have two more questions then.
1) Why is 3DMark03 not using the PS1.4 support via PS2.0 in its tests.
2) Why are they even using PS1.4 since only one hardware company supports it?

Not trying to start a fight or anything just trying to understand how a benchmark can be "neutral" when it is written only for one brand of card.

1) Only the GFFX is would receive benefit from this and it does IIRC as does the Radeon 9500/9700 series
2) Only one company supported Pixel Shaders when 3DMark 2001 was out, FutureMark expects other manufacturers to join the bandwagon with PS2.0 giving compatibility to PS1.4. In essence all DX9 cards will support it and 3DMark03 is meant to be a forward looking benchmark.

Simply put PS1.4 is very good and does what people like JC wanted the original PS to do. The fact that only ATI was forward looking enough to implement this feature is a moot point in the grand scale.

P.S. I do not think 3DMark03 is without faults but this [PS1.4 support] is not IMHO a fault of 3DMark03.

Edit: clarification
 
Why do you think "a benchmark" (undoubtedly you think of 3dmark 2003 here) is written for only one brand of card? For starters, we already concluded PS1.4 is not limited to just one brand of cards.

Furthermore, ANYONE is free to implement PS1.4 in hardware. It's part of directx specs. That Nvidia decided not to do so with the GF4 was their choice (and loss). Not sure about S3 or SiS, which PS 1.x version they support.

Finally, your assumption the benchmark is written for only one brand is more like an ass-umption, since 3dmark features a completely identical ps1.1 codepath in addition to the ps1.4 path. They provide the same output, ps1.4 simply gets the job done in fewer passes.

What possible benefit to anyone would it be to not support ps1.4? Your reasoning seems strangely backwards to me...


*G*

Edit: clarification here as well. :)
 
What about Matrox, VIA/S3 or 3DLabs?

What about them? Only two of them have shader capable parts, and as the recent Peddie report highlights they account for less than half 1% of sales over the last quarter. When you talk about only one manufacturer supporting them, effectively there is only two manufacturers that have any significant marketshare. Considering NVIDIA's majority volume product has no shader capabilities at all, you are probably looking at a reasonable proportion of all shader capable boards supporting PS1.4, which will only increase as more PS2.0 parts become available in the market, and those OEM's keep shipping 9000's out.

FYI, Your IP screams at me what firewall you are posting through... :)
 
James said:
1) Why is 3DMark03 not using the PS1.4 support via PS2.0 in its tests.
2) Why are they even using PS1.4 since only one hardware company supports it?
1) PS1.4 support via PS2.0? What do you mean? Either you write a PS1.4 shader or a PS2.0 shader. And the driver makes sure it is executed according to the specs.

2) When you write a shader, you try to model a visual effect. If the effect you want to create is possible via PS1.4, and you don't need the floating point precision that is guaranteed by PS2.0, then you use PS1.4. No need to use PS2.0 and leave PS1.4 cards out.
 
Tahir said:
James said:
huh? Now I have two more questions then.
1) Why is 3DMark03 not using the PS1.4 support via PS2.0 in its tests.
2) Why are they even using PS1.4 since only one hardware company supports it?

1) Only the GFFX is would receive benefit from this and it does IIRC as does the Radeon 9500/9700 series
Wouldn't any DX9 card benefit from this?

Tahir said:
2) Only one company supported Pixel Shaders when 3DMark 2001 was out, FutureMark expects other manufacturers to join the bandwagon with PS2.0 giving compatibility to PS1.4. In essence all DX9 cards will support it and 3DMark03 is meant to be a forward looking benchmark.
Maybe I am missing something here. How can other manufacturers join the bandwagon via PS2.0 when 3DMark03 will not use PS1.4 via PS2.0?

Dont get me wrong I think PS1.4 great and very useful. But since PS2.0 is here what good is writing for just PS1.4?
 
James said:
Tahir said:
Maybe I am missing something here. How can other manufacturers join the bandwagon via PS2.0 when 3DMark03 will not use PS1.4 via PS2.0?


Uhm... not sure what you mean...
Any PS2.0 support card HAS to support PS1.4. It's a DX requirement that PS compliance requires full backwards compatibility to execute lower-level shaders.

Your second question was already answered.
 
Xmas said:
1) PS1.4 support via PS2.0? What do you mean? Either you write a PS1.4 shader or a PS2.0 shader. And the driver makes sure it is executed according to the specs.
Hmm then why is the GeForceFX ending up using PS1.1 in the 3DMark03 tests? Is it a driver bug?

Xmas said:
2) When you write a shader, you try to model a visual effect. If the effect you want to create is possible via PS1.4, and you don't need the floating point precision that is guaranteed by PS2.0, then you use PS1.4. No need to use PS2.0 and leave PS1.4 cards out.
Ahh ok, I am new to using Shaders and to DirectX as well.
 
Ichneumon said:
Uhm... not sure what you mean...
Any PS2.0 support card HAS to support PS1.4. It's a DX requirement that PS compliance requires full backwards compatibility to execute lower-level shaders.

Yeah thats what I thought too. Makes me wonder why GeForceFX is falling back to PS1.1 in 3DMark03. In any case I think I have my answer now.

Thanks for the help.
 
Hmm then why is the GeForceFX ending up using PS1.1 in the 3DMark03 tests? Is it a driver bug?

A - does it? You have confirmation of this?

B - If it does do it, I would presume it would be for speed. AFAIK the integer path of GeForce FX is only capable up to PS1.3, it may only be able to support PS1.4 on the floating point hardware. However, there are fewer FP calculating units in GFFX than there are integer so the increased shader execution speed via PS1.1 on the integer hardware may offset the performance degredation of multipassing.
 
James said:
Yeah thats what I thought too. Makes me wonder why GeForceFX is falling back to PS1.1 in 3DMark03.
It doesn't. Maybe your confusing this with the optimizations Nvidia has done to their drivers, they stated (maybe on an interview on extreme-tech? Can't remember) that they execute some shaders as PS 1.1 and some as PS 2.0 (they probably can't run PS 1.4 through their fixed-point register-combiner path which is fast but inflexible and thus would need the flexible, but slow float path). I have no idea how they do this, maybe they're telling 3dmark03 the GFFX is only PS 1.1 depending what the benchmark wants to do.
(IMHO the large speedup with the latest nvidia driver for the GFFX is probably because of exactly this optimization, at least it would make sense after reading this thread http://www.beyond3d.com/forum/viewtopic.php?t=4252 unless I misunderstood something).
 
James said:
So other then ATI the only other card that can support PS1.4 is GeForceFX via PS2.0? What about Matrox, VIA/S3 or 3DLabs?

huh? Now I have two more questions then.
1) Why is 3DMark03 not using the PS1.4 support via PS2.0 in its tests.
2) Why are they even using PS1.4 since only one hardware company supports it?

Not trying to start a fight or anything just trying to understand how a benchmark can be "neutral" when it is written only for one brand of card.

GForceFX supports it. Also NV31, NV33, NV34,NV35,NV36 unless nVidia does not really fully support DX9.
 
mczak said:
maybe they're telling 3dmark03 the GFFX is only PS 1.1 depending what the benchmark wants to do.
I don't believe this is possible. At CAPs reporting time, you don't know anything about the application calling you. Either you report PS 2.0 or you don't. Since 3D Mark 2003 is using the DX9 interface on GT2,3,4 (not sure about GT1 at the moment), PS 2.0 should be reported.

MS has made it so you can't tell the name of the application calling you, so I don't see how you could change your behavior without any hints.
 
Back
Top