2 Replies Latest reply on Aug 19, 2008 5:19 PM by Nikolay_Mikhalev

    Compatibility of processors


      Hello everyone!

      Not so long ago I have got HD 2600XT (RV630) and programming GPU with use CAL/IL became accessible to me. But at viewing of examples from CAL SDK I had some questions. Here one of them. The code of a kernel from an example "global_input_IL" ("global_output_IL" too) in ShaderAnalyzer was successfully compiled only on processor RV670. In the documentation there is a reservation what only processors RV670 can work with global memory but why there was no compilation on RV770, after all it is later processor? Certainly, it is a program complex it is intended for family R600, but unless here there should not be a compatibility? 

          • Compatibility of processors

            Thanks for the answer, ryta. It has a little cleared a situation. 
            Speaking more widely, me the compatibility problem excites. And use of global memory it is simple one of displays. It is logical to assume, that tomorrow there will be other difficulties with code carrying over. But it would be desirable to believe, that the code written for GPU now will work at least on third of those video processors that will be created in 3-4 years. 
            I for a long time am interested in programming in the field of a 3D-graphic (without DirectX and OpenGL). There are very many operating time of a code in this area. But, as always, it would be desirable that all worked faster, than is. Naturally, having learnt about CAL at once it wanted to put it into practice, but process of copying of existing algorithms and realisation new under GPU, in this case, - is very labour-consuming. Also it would be desirable to know, that as a result it will be not vain. Or all the same to leave all calculations for CPU where compatibility will be and in 10 years? Unfortunately, now I more tend to the last.

            P.S.: the doubt imposes still that CPU is quite good with it consults (Athlon64 X2 4200+): all code is written on the pure assembler under AMD64 with very heavy use SSE3 (further will be and SSE5)