R420 Performance boost

[RingWraith]

Considering the R3xx handled T&L through the Vertex shader, couldn't they boost the performance of the R420 by adding a dedicated T&L unit?

 

[Pete]

That would go against the R300's less is more philosophy, and I don't think games that use T&L need much further boosting. Transistors would be better budgeted toward shaders or other forward-looking tech.

 

[K.I.L.E.R]

Popular games that use a hardwired T&L unit are nearly dead with maybe some small exceptions.

 

[Ostsol]

Popular games that use a hardwired T&L unit are nearly dead with maybe some small exceptions.

Well, not really. . . but in any case, the speed boost would be negligible when compared to the already huge increase of speed the games will get from the new card's raw power.

 

[RingWraith]

T&L is still being used, because EVERYONE has T&L capable cards. I wasn't referring to a boost in T&L performance, I was referring to the increased VS performance for not having to emulate a T&L unit.

The rule that seems to be universal is the fact that a dedicated unit is much faster than a multi-pupose unit.

 

[KimB]

T&L is still being used, because EVERYONE has T&L capable cards. I wasn't referring to a boost in T&L performance, I was referring to the increased VS performance for not having to emulate a T&L unit.

The rule that seems to be universal is the fact that a dedicated unit is much faster than a multi-pupose unit.

I don't think you would have to worry about this. GPU's are fast because they can work on many parallel streams of data, and nearly all processing can be broken down into a few simple mathematical operations that are done repeatedly. The GPU programming algorithms enforce this independence of vertex and pixel processing. Any special hardware that must be added to the vertex shader for fixed-function T&L would not slow the vertex shader down, because it is still working on the same type of data. Still, I'm pretty sure that today's vertex shaders are more than general enough to fully emulate fixed-function T&L without any extra hardware dedicated to fixed function.

 

[trinibwoy]

Is the T&L implementation in hardware transparent to the application? Do you have to specifically code for a vertex shader implementation? I thought it was just a driver/hardware consideration.

 

[AndrewM]

Is the T&L implementation in hardware transparent to the application? Do you have to specifically code for a vertex shader implementation? I thought it was just a driver/hardware consideration.

You have to specifically program vertex shaders. HW T&L is transparent.

 

[trinibwoy]

Is the T&L implementation in hardware transparent to the application? Do you have to specifically code for a vertex shader implementation? I thought it was just a driver/hardware consideration.

You have to specifically program vertex shaders. HW T&L is transparent.

I don't think you understood my question. For those cards that emulate T&L in the vertex shader, do you need to do any special programming. I'm not talking about vertex shader programs here.

EDIT:Coming to think of it, it's a dumb question since R3xx handles old games fine so no special code would be necessary. I'm assuming here that R3xx has no FF T&L unit.

 

[UPO]

[edit] nevermind [/edit]

 

[Dave B(TotalVR)]

It would be simple enough for the graphics card driver to detect when FF TnL is being expected in the application running and to upload its FF TnL simulation vertex shader program to the GPU. Not a probalem at all.

Dave

 

[tb]

T&L is still being used, because EVERYONE has T&L capable cards. I wasn't referring to a boost in T&L performance, I was referring to the increased VS performance for not having to emulate a T&L unit.

The rule that seems to be universal is the fact that a dedicated unit is much faster than a multi-pupose unit.

You can't use fixed T&L and Vertex Shaders at the same time or with the same drawing instruction in DirectX (Don't know about OpenGL). Beside this, how many games do stress the fixed T&L unit that much? You could do much more with a Vertex Shader....

Thomas

 

[Ostsol]

I believe the spec in OpenGL allows some T&L features such as fog to be used in combination with vertex shaders. Lighting, however, must implemented in vertex shaders, if they are being used. All transformations must also be performed in vertex shaders. Looks like it's not all that different from D3D.