Technical question on video cards and CPUs

[TheCommander]

Why can't processors do the job of video cards? Are the two components so different from each other that CPUs could not do the job of video cards?

 

[Snowknight26]

They can, just look at WARP10. However, GPUs are good at what they do because they are far better at doing floating point operations than CPUs, while CPUsf are far better at doing integral operations.

 

[evilsofa]

Intel has been working on a multi-core GPU/CPU hybrid thing called Larrabee. There's so much FUD and propaganda from both Intel and the GPU makers that I can't really sort out if this thing is going to work and how it will really compare to what nVidia and ATi are doing.

 

[Red Falcon]

Why can't processors do the job of video cards? Are the two components so different from each other that CPUs could not do the job of video cards?

Are you talking about computing graphics or doing stuff like (CUDA) encoding videos?

Back before the time of DirectX, CPUs did process the majority of the graphics (early 1990's and 1980's).

CPUs are not very good at processing graphics however and with any modern machine, OS, and games (last 10 years) CPUs won't process the graphics unless the program specifies it to do so.

 

[ivansoze1]

Why can't processors do the job of video cards? Are the two components so different from each other that CPUs could not do the job of video cards?

..GPU's,traditionally had functions/effects, that are specific to video output ie., what you see on the screen - a CPU would need the programming and then power/speed to process same effect - very time consuming. It is only lately, with the new GPU(SIMD based) processing architecture that all this CUDA ect stuff is possible and thus comparisons with the 'ol CPU are made..all v interesting

 

[Kadji.Kahn]

A general purpose CPU and a GPU are two very different types of processors. Generally a cpu is considered scalar; doing one thing at a time with mulitple instructions. GPU's being vector in design can perform mathematical operations on multiple data elements simultaneously. Streaming apps like video and the like work great on a GPU. However vector instruction units incur serious performance penalitys when doing out of order instructions.

Being complimentary, I think the two designs combined are the future. heterogenous multiprocessing seems like the next big thing for a single physical design.