graphic card pipelines VS Pixel Shader Processors

[junwang0808]

i don't get it whats the difference between these 2??? because
when i look at X850 and x1950 AGP


x1950's pipelines are not listed but have 36 shader processors then
x850 have 16 pipelines and 16 shader processors sooo

what is the difference between these 2 ?? pipeplines and pixel shader processors ??

if u have to choose x1950 and x850 AGP which one would you choose ?

thanks guys i am a noob


and what is Geometry Rate??? because x1950 have no geometry rate...

[saphalline]

Well, the X1950 is 1.5 generations beyond the X850, so that's a no-brainer to me! :p

Pixel shaders are the shader units responsible for operating on surfaces before the final rasterization. A "pipeline" is now an old term to describe the full rendering path in GPU's/VPU's (historically even including the ROP's).

Ummm... maybe now would be a good time to take a break. :p How much more info do you want? Trust me when I say that learning too much about hardware quickly turns into a black hole!

[junwang0808]

thx for responding i would like to learn everything about graphic card can you please teach me?? so i will know what graphic cards to buy later?? instead of just blindly selecting one that looks cool?

[saphalline]

Start here (http://www.pcguide.com/vb/showthread.php?t=39536). Let me know when you're ready to learn more. ;)

[junwang0808]

i have already read that before. there is something i still don't get.

what is the difference between DDR. DDR2 and DDR3 and DDR4
i know like 2 of faster than DDR and 3 is faster than 2 and 4 is faster than 3.

and for example a graphic card's RAM speed is 600MHz

then for it is DDR the true ram speed is : 600
then whats true ram speed for DDR2?
DDR3?
ddr4?


and is it possible to enable pipelines on a card?? cuz i have x800 pro AGP it has only 12 pipelines i want to enable all 16 is it possible?

thanks

[saphalline]

The various forms of DDR have increased pre-fetching capabilities, as well as better power consumption, faster theoretical speeds, smaller process designs, reduced minimum voltage requirements, and a few other assorted features. In addition, GDDR variants are specifically designed for graphics applications because of their ability to zero-out completely with a single strobe command. (This is not useful for system RAM. :p)

However, GDDR4 is not 4 times faster than DDR in effective clock speed. The base technology is still DDR, so 600MHz GDDR4 and 600MHz DDR are both effectively 1200MHz total. The convention in the RAM industry is that if another multiplier came out, such as 3 times, TDR would be used instead of DDR (for Triple Data Rate instead of Double Data Rate). So all DDR variants are still only doubled.

is it possible to enable pipelines on a card?? cuz i have x800 pro AGP it has only 12 pipelines i want to enable all 16 is it possible?It may be possible depending on your specific X800 vid card, but this is not something I endorse because of the extreme danger if done incorrectly. If you want to look into it further, you'll have to do a Google search on unlocking pipelines.


As to your previous question about pipelines vs shader processors, things have changed greatly in the graphics arena since programmability was added. This goes back to the early days when the first GeForce and Radeon vid cards were released. If you remember, hardware TnL (Transformation and Lighting) is where the second rendering stage is off-loaded from the CPU onto the vid card. This was the first step towards the rendering quality that we see today in games running in real-time. But even then, there was a distinctive pipeline that had to be followed within any GPU/VPU.

Fast forward to DirectX 8.0/8.1, when vertex shaders were first added with the GeForce 3 and Radeon 8500. This represented the step into fully programmable GPU's/VPU's because vertex processes could be looped and contain any data required for a certain effect. The concept of a pipeline became fuzzy during this time period. There was no longer a single path of rendering because the number of conventional pipelines and vertex shaders were different. In essence, it was possible for data to start in one pipeline and come out a different one, depending on its need for vertex processing and how long it took. But still the industry held onto the pipeline concept because they were mostly intact, except for the errant vertex shaders.

Stepping forward again to DirectX 9, we see the full decoupling of the conventional pipeline with the GeForce FX and Radeon 9500. With the introduction of pixel shaders and the acceleration of DirectX requirements, the early and mid DX9 days were full of confusion and constantly changing compliancy. In addition to that, pixel shaders more or less killed the concept of a conventional pipeline. New designs bred new interpretations of vertex and pixel shader processor ideas, with new versions coming out all the time. Vertex Shader 1.0, Vertex Shader 2.0, Pixel Shader 1.0, Pixel Shader 2.0, Pixel Shader 2.0b, etc. This is the era in which shaders became ever more important to the gaming experience because of course all the latest games were relying more heavily on them. Doom 3 and FarCry were early forays into the "shader experience", paving the way for later games, such as Half-Life 2. It became an age when gamers asked each other how many shader units they had, not if they had shader units.

In the final days of DirectX 9, we see the culmination of the final DX9 spec: DirectX 9.0c. Gone are individual vertex and pixel shader designations. We now simply refer to them as shaders. In keeping with that idea, vertex and pixel shader versions were promoted to 3.0 status, and merged into a single Shader Model 3.0. The idea of a Shader Model binds the vertex and pixel portions into one and helps to reduce complexity in the industry. ATI and NVidia move on from their confusing Radeon X800 and GeForce 6 days into a new era of X1000's and GF 7's. The conventional pipeline is now destroyed, as customized Dispatch Processors are integrated into the latest GPU/VPU designs. All graphics data now enters the vid card into a single port and is delegated from there. Texture units handle hardware TnL and basic texture fill-ins; Z-Compare units cull unseen surfaces and objects; Vertex shader units are responsible for 3D mesh manipulation; Pixel shader units handle advanced pre-raster effects on surfaces and volumes; and the ROP's finalize the scene and convert it into a 2D image suitable for viewing. There are now millions of possible paths that any given chunk of 3D data can take down the rendering path of modern GPU's/VPU's. It's all fully programmable and self-sustaining. They can respond dynamically to any situation while retaining near 100% efficiency.

All hail the modern graphics processor! The most advanced and complex programmable processor in the world!

[junwang0808]

wow thanks but i sitll got some questions

if you would look into x1900xtx and x1950 pro

x1900xtx have both 16 pipelines and 48 shader processors

but

x1950 pro only have 36 shader processors

from my understanding when programming is advanced pipeline is no longer needed but why does x1900xtx still have 16 pipelines?

does that means x1900xtx is faster than x1950 pro?

and also what is Pixel Fillrate and Geometry Rate*

and whats the difference between N/A ( none available) and N/A MTriangles (??? )
i saw those from the ATI product comparison charts often don't know what they are

thanks

[saphalline]

Yes, the X1900 XTX is more powerful than the X1950 Pro. It has more units overall, since the X1950 Pro is more of a high-end budget vid card and the X1900 XTX was the previous ATI top dog. There's only half a gen between these two, but both are waaaay beyond the X800 you mentioned before.

from my understanding when programming is advanced pipeline is no longer needed but why does x1900xtx still have 16 pipelines?It doesn't. It has 16 texture/TnL units, not 16 pipelines. But like I said before, the concept of a "pipeline" is still stuck in people's minds, so the marketing and resellers still call them pipelines. If you look on Newegg or any other online vendor, they still list them as pipelines, even though they aren't.

what is Pixel Fillrate and Geometry RateFillrate is how many raw pixels can be pumped out by the vid card, usually measured in hundreds of millions or billions per second. It's a direct relationship between the GPU/VPU clock speed and the number of ROP's. Not to say that the vid card will always output that many pixels per second (in terms of frame rate) but fillrate represents the theoretical maximum.

The geometry rate is actually a measure of how many triangles can be calculated by the GPU/VPU per second. All polygons and objects are represented in a 3D scene by individual triangles put together, or by other curved geometric surfaces, such as circles or spheres. Leaving curved surfaces out of the mix (because of their non-uniform nature) triangles per second are often measured as a method of determining the raw power of a graphics processor. However, if any of these triangles have special needs (ie, TnL or vertex shading) this can throw off the geometry rate. It is for this reason that geometry rate is used very little in today's world of shader-based games.

and whats the difference between N/A ( none available) and N/A MTriangles (??? )
i saw those from the ATI product comparison charts often don't know what they areWithout a context, I'm not sure what they are, either. Can you give a link or screenshot?

[junwang0808]

well if you would search

[ ati product comparison ] on google it should show up as the first one

it gives me all those stuff and i always go there. you know to compare their products

[saphalline]

Oh, you mean "N/A" and "N/A MTriangles" for both the X1900 XTX and X1950 Pro under "Geometry Rate"? (Yeah, it helps to have a context.)

That just means they didn't test the geometry rate for those vid cards. Like I said, for today's shader-based world, the raw triangle output is nearly useless. I want to see shader-based performance, not old DX7 cr@p. If I want to test my vid card's triangle output, I'll fire up something old, like Quake III, and crank up the res. If I want a modern visual experience in a game, I'll play Oblivion or Half-Life 2. After all, Oblivion looks more realistic at 800 x 600 than Quake III does at 1600 x 1200. Why? Shaders.

[junwang0808]

thanks buddy !! now i understand graphic cards better

even though i know R600 is about to come out and price for graphic cards are gonna drop i ordered my sapphire x1950 pro agp today !!

thanks !!!

i would never make up my mind without your help !!