| Popis: |
Graphics Processing Units are now powerful and flexible systems supporting 8, 16 or 32 bit floating point operations. Technical documentation provided by manufacturers (ATI and Nvidia) is not sufficient concerning the various floating point operators available in the vertex and the pixel shaders. However such detailed documentation is essential in applications involving numerical calculation which requires estimates of round-off error, floating-point reduction and compensated evaluation when implementing double, quadruple or multiple precision arithmetic. We present in this work a set of programs to discover the principal characteristics of floating point arithmetic of GPUs. We discuss of results obtained on two recent graphics cards: Nvidia 7800GTX and ATI RX1800XL. |
