ATI Radeon HD 4000 Series discussion

[kemist]

LoL. You mis-understand, really. TMU's are TEXTURE MEMORY UNITS...cache...

This post and like the next 3 posts after it are all wrong. First off TMU's are not Texture memory units, they are texture mapping units, though they do have large amounts of cache associated to them.

Um, TMU's do the texture processing? NO!!! The shaders do! So yes, that IS the case! More specifically, the inconnection speed between the TMUs and the shader core is much faster in nV's solution, hence the difference.

This too is wrong, why would you have TMUs if they are not doing the texture processing? Maybe you are confusing offloading AA work from RBE's into the shaders, but this is not occurring with TMUs yet.

Its been shown that there is nothing wrong with Ati's AA in and of itself, in fact the hit from just AA is often less than nvidias at the same setting. Its AF that tanks performance. The source below shows this, especially with HQAF which is default for Ati now. If you want old AF you need to use ati tray tools or other progs like that.

Source: http://www.computerbase.de/artikel/h...d_3870_rv670/5

[AliG]

This post and like the next 3 posts after it are all wrong. First off TMU's are not Texture memory units, they are texture mapping units, though they do have large amounts of cache associated to them.



This too is wrong, why would you have TMUs if they are not doing the texture processing? Maybe you are confusing offloading AA work from RBE's into the shaders, but this is not occurring with TMUs yet.

Its been shown that there is nothing wrong with Ati's AA in and of itself, in fact the hit from just AA is often less than nvidias at the same setting. Its AF that tanks performance. The source below shows this, especially with HQAF which is default for Ati now. If you want old AF you need to use ati tray tools or other progs like that.

Source: http://www.computerbase.de/artikel/h...d_3870_rv670/5

thank you for clearing that up, that's what I thought, but by the tone of his posts I thought I and everyone else in this discussion was wrong

[cadaveca]

thank you for clearing that up, that's what I thought, but by the tone of his posts I thought I and everyone else in this discussion was wrong

Um, actually, there is some confusion here over TMU's, and thier individual components. You guys are speaking in broad terms, I'm talking specifics.

Texture mappingm units include texture filters, texture memory, and a few other elements.

Texture mapping follows thusly:

1) Specify the texture. (R, G, B, A, mipmapping)


2) Indicate how the texture is to be applied to each


pixel. (decal, modulate, blend)


3) Enable texture mapping.


Before actual mapping can happen, certain work needs to happen first. I find the fault in a specific area, you guys are clueless spouting acronyms you aren't really aware of.

[kemist]

Um, actually, there is some confusion here over TMU's, and thier individual components. You guys are speaking in broad terms, I'm talking specifics. [SNIP] Before actual mapping can happen, certain work needs to happen first. I find the fault in a specific area, you guys are clueless spouting acronyms you aren't really aware of.

What specific area do you find at fault?

Your previous posts were citing weak AA performance not TMU deficiencies.

If you care to explain your reasoning more perhaps people wouldnt disagree with your assertions, but you are not going into very much detail.

edit: Are you talking about it needing more TFUs so that TAU:TFU is 1:1? This would double AF perf even though tmu count would technically stay the same.