CUDA/shader/Core

[amdinside]

I still cannot understand the very difference between nVidia's CUDA cores, the so called (graphics) "core" and the shaders, which operate at different frequency compared to the "core".

Could you possibly help me with the chaos in my mind and explain me the differnce between them. Thanks.

[STEvil]

CUDA cores is a fancy marketing name so you will buy their product.

[amdinside]

CUDA cores is a fancy marketing name so you will buy their product.

Are they renamed Shaders ?

[Chumbucket843]

afaik shaders were never an official name and to be fair that term is misleading because they do much more than shading. for some reason nvidia has trademarked parts of their architecture such as parallel data cache(tm), gigathread engine(tm), warps(tm), streaming multiprocessor(tm). idk why they would want to do this.

their are two main parts to gpu's: the core and the shader engine. the core is the ROPs (called render back ends by ATi)which are fixed function and do AA, AF and rasterization tasks. the shader is programmable and is used for shading, shadows, translating/rotating/transforming thousands of triangles, bump mapping, nad almost any cool graphics effect you see in games. it gets fairly complex trying to explain how texturing/tmu's are incorporated into this system but it's safe to leave it out.

different kinds of logic are used for rops and shaders. this means that they have different properties like clockspeed/performance/power consumption. to get the most out of the available transistors nvidia clocks the shaders 2x faster rather than running them at the same speed as rops. ATi optimized their rops to run at higher clockspeeds instead. to make up for slower rops nvidia adds more of them.

[STEvil]

afaik shaders were never an official name and to be fair that term is misleading because they do much more than shading. for some reason nvidia has trademarked parts of their architecture such as parallel data cache(tm), gigathread engine(tm), warps(tm), streaming multiprocessor(tm). idk why they would want to do this.

Fancy names make for fancy advertising

[WhiteFireDragon]

to answer OP's question in the most basic way, CUDA cores = shaders = stream processors.

[Sam_oslo]

In old days when GPUs were used only for gaming, developers could program "shaders" for all kind of grafical effects for games. Now days GPUs can do much more, including some useful parallel computing too, and that's where the CUDA ("Compute Unified Device Architecture", which is a parallel computing architecture developed by nVidia) Cores can be programmed by developers.

For those who doesn't need (or understand) CUDA, it sounds like a marketing name, but it is actually an architecture for parallel computing.

[amdinside]

Thank you very much.