Who has the better shader architecture?

N

no2censorship

Limp Gawd
Joined
Oct 20, 2008
Messages
393
To me it sounds like neither; ATI's is clearly more programmable, but nvidia's is clearly more transistor/thermal/performance efficient.

But, if I'm wrong let me know.

My reasoning:
taking 2 issues at a time makes it so transistor space is saved, so clock speeds can go higher; if ATI takes 5 it will take more transistors, not optimizing the number of shaders thermally possible on the pcb, nor the clock speed they can run at.

ATi's more programmable because they've always had better programmable shaders since the 9700 pro, or it seems that way by their willingness to remove more features from hardware and emulate them thru shaders; they got rid of the w-buffer first (which is said to be complex and take up a lot of transistors), for example; or at least since the x1k in 2005.

not that I'll buy an ati product until they make their colors look natural to me and not until their filtering is unoptimized trilinear+selected number of samples is what you get. Basically, if they made a 12x angle invariant mode that wouldn't look any worse than nvidia's near-invariant 16x mode. As for performance, I couldn't care less.

but, you've got to give credit where credit is due, rather than have blinders on like the people at rage3d and nvnews appear to=]
 
yeah, performance in terms of what?

IIRC ATI has 1 shader per back-end color channel (which is just wierd), so in actuallity 800 is more like 266
And I think CUDA puts any claims ATI has to "Our SP's are programmable!" to shame. Unless I've been left in the dark, ATI doesn't have, and doesn't plan to have, a C programming environment for their GPUs. To get at em, you still need to go through DX or OGL APIs.
 
you can use Brook+, a variant of ANSI-C with ATI's stream processors, so they are both just as programmable. (Brook+ is a variant of BrookGPU by stanford university that allows integer and double precision processing)
 
Danny Bui said:
Better in terms of what?
Click to expand...
better in terms of performance per transistor and programmability. Sorry for the ambiguity and thanks for replying=]

MrWizard6600 said:
yeah, performance in terms of what?

IIRC ATI has 1 shader per back-end color channel (which is just wierd), so in actuallity 800 is more like 266
And I think CUDA puts any claims ATI has to "Our SP's are programmable!" to shame. Unless I've been left in the dark, ATI doesn't have, and doesn't plan to have, a C programming environment for their GPUs. To get at em, you still need to go through DX or OGL APIs.
Click to expand...
Okay, makes sense. Hadn't thought about that. But what about OpenCL?

I don't know "jack or shit" about it, so you should school me, to help make me a less ignorant person.

THRESHIN said:
i think the OP was drunk
Click to expand...
Kind of close=]
 
ATi have without a doubt the most powerful ALU processor and therefore Shader architecture of the two (RV770 vs. GT200).

RV770 is around 140-200% faster at Shader Calculations than the GT200.

We've tested this out with various shaders using a whole slew of games. Beyond3D also have come up with the same conclusions in this thread: http://forum.beyond3d.com/showthread.php?t=49327&page=2

b3da018.png


What makes nVIDIA's GT200 faster is it's Texturing (TMU) as well as RBE (Render Back End or Pixel Throughput).

GT300 is going to look a lot more like RV770 than GT200.
 
ElMoIsEviL said:
ATi have without a doubt the most powerful ALU processor and therefore Shader architecture of the two (RV770 vs. GT200).

RV770 is around 140-200% faster at Shader Calculations than the GT200.

We've tested this out with various shaders using a whole slew of games. Beyond3D also have come up with the same conclusions in this thread: http://forum.beyond3d.com/showthread.php?t=49327&page=2

What makes nVIDIA's GT200 faster is it's Texturing (TMU) as well as RBE (Render Back End or Pixel Throughput).

GT300 is going to look a lot more like RV770 than GT200.
Click to expand...

Aye, and this is also why RV770 cards are the champs of furmark (when it's not crippled by the drivers), which is basically a pure shader benchmark. My 4870 scored over 6k - running my GTX260 @ 837/1674/1296 I'm just shy of 4400. Maximum framerate is within a few fps of the minimum framerate on the 4870.
Still, its all about gaming performance. In this respect, I'll take my 260 purely because of the overclock it will run (at stock speeds it was a wash).
 
You must log in or register to reply here.
Back
Top