PDA

View Full Version : Pixel shader 2.0 problem on FX???


Pages : [1] 2 3 4 5

Dragunov
08-26-03, 09:21 AM
Can someone tell me of this problem is a driver-issue or a hardware-issue

If it is a driver-issue, when are they going to fix this?

If it is a hardware-issue, what is Nvidia going to do about this?

Does anyone has some more info/links

http://www.theinquirer.net/?article=10602 (here Nvidia says it is a driver-issue)

http://www.driverheaven.net/dhinterviews/nvidia/ (here Derek Perez says this:

Zardon: Is there any validity to the claims made by the INQ that nVidias PS2.0 part of the detonators is faulty at the minute?

Derek: No this is not true.

What is true??

THanks

Dragunov

The Baron
08-26-03, 09:23 AM
IIRC, it was a driver problem that was fixed in 44.67 or 44.71 or 44.90. ONE of those. Don't know too much about it, though.

Dragunov
08-26-03, 09:25 AM
But why is there such a big gap between the performance of an ATI and a Nvidia-card in those DirectX 9.0-game (Tomb raider)

THanks

Dragunov

Hanners
08-26-03, 09:37 AM
Originally posted by Dragunov
But why is there such a big gap between the performance of an ATI and a Nvidia-card in those DirectX 9.0-game (Tomb raider)

THanks

Dragunov

That part is almost entirely down to the hardware - It just isn't built for anything approaching good Pixel Shader 2.0 performance.

The Baron
08-26-03, 09:37 AM
Originally posted by Dragunov
But why is there such a big gap between the performance of an ATI and a Nvidia-card in those DirectX 9.0-game (Tomb raider)

THanks

Dragunov
That's a hardware problem. Plenty of other people could explain it better than I can, but basically, NVIDIA put a weak PS unit in the FX series because it probably thought that PS2.0 wouldn't be used until at least 2004.

Now, anyone want to give the Ultra Technical Explanation?

991060
08-26-03, 10:50 AM
nvidia's having trouble when dealing with PS 2.0 at present.

they don't expose the support of FP16/32 RTF(render target format) in the current drivers. Someone said such support can be given through the 50.XX det,but I'm not quite sure about it.

I dont think it's a big deal that FX's having inferior PS 2.0 performance than R3XX since PS 2.0 is not widely used now. we're playing DX8 games most of time,right?

I just hope NV40 could be a competitive product....

Hellbinder
08-26-03, 10:52 AM
Now, anyone want to give the Ultra Technical Explanation?

Sure...

The Fx....

its... *Slower*...

There you can all relax your brains now, I know that was a tough one.

:cool:

Hellbinder
08-26-03, 10:59 AM
I dont think it's a big deal that FX's having inferior PS 2.0 performance than R3XX since PS 2.0 is not widely used now. we're playing DX8 games most of times,right?

Earth to man from Pluto....

*Half Life 2* only the biggest game of all time. Comming in mere weeks. fully supporting PS and VS 2.0

If you think having inferior PS performance is not going to have an impact..

:eek:


they don't expose the support of FP16/32 RTF(render target format) in the current drivers. Someone said such support can be given through the 50.XX det,but I'm not quite sure about it.

That is not the major cause of the performance issues with PS 2.0. They are not currently exposing a whole ream of floating point functions. Like Floating point Cube maps, Floating point volumetric textures etc. Still the issues causing their lack luster DX9 performance are *Hardware* related. Not software. Except where they have not had the opportunity to specifically hack the game code to support their Architecture.

991060
08-26-03, 11:14 AM
HL2 won't use PS2.0 everywhere,otherwise even the faster card available now will struggle with it. (I didn't see anything need to be done by PS2.0 in HL2 from the E3 demo).

And I'm not saying the lack of FP RTF support is what gave FX such bad PS performance. there's just not enough FP unit in the GPU,it's hardware related:(

Dragunov
08-26-03, 11:22 AM
If you look to the specs of the NV 35:

- PS 2.0+
- VS 2.0+

So , Does Nvidia lie to us, I don't thinks so, can they post this if it is not true, if they use PS 2.0+ and VS 2.0+, they go ever farther than directX 9.0

Does this problem excist in the NV35??

Dragunov

991060
08-26-03, 11:33 AM
it's *mainly* the more instructions FX supports per pass that where the "+" comes from.(I could be wrong,checking DX spec now). most of the shaders can be executed on current hardware with decent speed won't exceed the length of 100 instructions. So the "+" doesn't really mean sth useful.(again,I could be wrong because I don't have a FX card,don't really know what magic it can do.)

the lack of FP RTF doesn't mean you can't use FP in single pass shaders.
you just can't write data to the memory in FP format. if everything is done within the GPU,FX is ok.

reever2
08-26-03, 11:34 AM
The FX's have less pixel shading units than the radeons, thats all i know...

Dragunov
08-26-03, 11:35 AM
Originally posted by 991060
it's *mainly* the more instructions FX supports per pass that where the "+" comes from.(I could be wrong,checking DX spec now). most of the shaders can be executed on current hardware with decent speed won't exceed the length of 100 instructions. So the "+" doesn't really mean sth useful.(again,I could be wrong because I don't have a FX card,don't really know what magic it can do.)

the lack of FP RTF doesn't mean you can't use FP in single pass shaders.
you just can't write data to the memory in FP format. if everything is done within the GPU,FX is ok.

So there is no big problems and FX has PS2.0 and VS 2.0? But did you saw that gab between Radeon and GeForce whey they tested the cards on PS2.0 with Angel of Darkness? Enormous

Dragunov

991060
08-26-03, 11:41 AM
of course I did.

you have to understand that supporting some feature and speeding up such feature are completely different things....

FX5900 and 9800P follow the same spec,but that doesn't mean they can run the test equally fast.

Dragunov
08-26-03, 11:47 AM
But maybe if the Chip supports this really, the drivers are maybe not as good to support this too, if it is aboard the chip than the chip must be able to calculate it, and The FX has a higher bandwidth and clockspeed, so he must be able to calculate this faster (geforce Fx 5900 Ultra) than the Radeon 9800PRO, but he can't, so can it not be driver-issue??

Dragunov

991060
08-26-03, 11:55 AM
I think nvidia is doing their best to improve FX's PS performance. But *personally* I'm not expecting a big headroom for nvidia to improve FX's PS performance through driver upgrade.

DX9 spec is co-designed by M$ and ati, there's no wonder ati has some advantages beyond nvidia.

it's just like you can't run an ati demo faser on nvidia's card.

DX9 spec is based on R300,that's all why ati has better performance.

And higher clock/mem rate doesn't mean everything. R300/350 can output 8 pixels/clock when PS2.0 is used while NV30 can only give you 4 with the same situation. I remember NV35 can do 6(not quite sure about it).

Malfunction
08-26-03, 12:03 PM
I actually have a question: ;)

Since these GPU's/VPU's are programmable, does anyone know if they are re-programmable?

Could a BIOS update add better instructions to the GPU/VPU. Could it take something away? I mean they are programmable, does that mean only during the manufacturing process?

Peace,

:confused:

NickSpolec
08-26-03, 12:03 PM
Dear Lord...

Nvidia's flagship, the FX 5900, is nearly beaten out by the *lowly* 9500!

That is embaressing.

991060
08-26-03, 12:09 PM
Originally posted by Malfunction
I actually have a question: ;)

Since these GPU's/VPU's are programmable, does anyone know if they are re-programmable?

Could a BIOS update add better instructions to the GPU/VPU. Could it take something away? I mean they are programmable, does that mean only during the manufacturing process?

Peace,

:confused:

the programmability of the pipeline doesn't mean you can move the transistor within the GPU. with the old T&L pipeline,all that you can do is to input some parameters into the chip,setting some render state,etc.

and with programmable pipeline,you can do whatever you like with the provided instructions. you can define your own lighting/shading mode,writing vertex shader to animate your characters,even do non-graphics-related calculation within the GPU(like physics calculation).

Malfunction
08-26-03, 12:43 PM
Originally posted by 991060
the programmability of the pipeline doesn't mean you can move the transistor within the GPU. with the old T&L pipeline,all that you can do is to input some parameters into the chip,setting some render state,etc.

and with programmable pipeline,you can do whatever you like with the provided instructions. you can define your own lighting/shading mode,writing vertex shader to animate your characters,even do non-graphics-related calculation within the GPU(like physics calculation).

Thank you.

:)

Spotch
08-26-03, 12:46 PM
Originally posted by NickSpolec
Dear Lord...

Nvidia's flagship, the FX 5900, is nearly beaten out by the *lowly* 9500!

That is embaressing.

Actually it may actually lose to the 9600 Pro if things were equal. The ATI card is actually using more effects and features of the game that simply arent enabled with the FX 5900 Ultra like PS 2.0 DOF and 32 bit PS 2.0 shadows.

It would almost certainly lose with a nice overclock. It is arguable that 450-500 Mhz core is virtually guaranteed with these cards.

:D

The Baron
08-26-03, 02:33 PM
Originally posted by Spotch
Actually it may actually lose to the 9600 Pro if things were equal. The ATI card is actually using more effects and features of the game that simply arent enabled with the FX 5900 Ultra like PS 2.0 DOF and 32 bit PS 2.0 shadows.

It would almost certainly lose with a nice overclock. It is arguable that 450-500 Mhz core is virtually guaranteed with these cards.

:D
Check out the B3D Tomb Raider article.

It *does* lose to a stock 9600 Pro.

SmuvMoney
08-26-03, 02:43 PM
Originally posted by 991060
I think nvidia is doing their best to improve FX's PS performance. But *personally* I'm not expecting a big headroom for nvidia to improve FX's PS performance through driver upgrade.

DX9 spec is co-designed by M$ and ati, there's no wonder ati has some advantages beyond nvidia.

it's just like you can't run an ati demo faser on nvidia's card.

DX9 spec is based on R300,that's all why ati has better performance.

And higher clock/mem rate doesn't mean everything. R300/350 can output 8 pixels/clock when PS2.0 is used while NV30 can only give you 4 with the same situation. I remember NV35 can do 6(not quite sure about it).

What evidence do you have that nVidia was left out of the discussions regarding the specifications and creation of DirectX 9? I believe MS went to all major IHVs as well as other major players in the industry to develop DX9.

Skuzzy
08-26-03, 02:56 PM
Originally posted by 991060
I think nvidia is doing their best to improve FX's PS performance. But *personally* I'm not expecting a big headroom for nvidia to improve FX's PS performance through driver upgrade.

DX9 spec is co-designed by M$ and ati, there's no wonder ati has some advantages beyond nvidia.

it's just like you can't run an ati demo faser on nvidia's card.

DX9 spec is based on R300,that's all why ati has better performance.

And higher clock/mem rate doesn't mean everything. R300/350 can output 8 pixels/clock when PS2.0 is used while NV30 can only give you 4 with the same situation. I remember NV35 can do 6(not quite sure about it).

This is just way wrong. NVidia was just as involved in the DX9 specification as ATI was. The R300 was based on what MS decided DX9 would be and NVidia had the same specification in their hands as well, but NVidia did not want to go the direction MS wanted to go.
Both companies had a chance to lobby for things in DX9.

What you have suggested is so far from the truth, I am surprised the mods have not yanked the post from the board. Talk about mis-information! OY!

Hanners
08-26-03, 03:01 PM
Originally posted by Dragunov
But maybe if the Chip supports this really, the drivers are maybe not as good to support this too, if it is aboard the chip than the chip must be able to calculate it, and The FX has a higher bandwidth and clockspeed, so he must be able to calculate this faster (geforce Fx 5900 Ultra) than the Radeon 9800PRO, but he can't, so can it not be driver-issue??

Dragunov

No, it can't be a driver issue, and here's why:

When it comes to Pixel Shader 2.0 processing, bandwidth and fillrate take a back seat (relatively speaking) and the number of pixel shader units and how many instructions they can process per pass becomes all important.

This is where one of the big differences between the R3x0 and NV3x cards comes into play - R3x0 cards have more pixel shader units per pipeline. These pixel shader units are also dedicated to only pixel shading on R3x0, whereas the NV3x's units are also used to do other things.