The article shows the limits in the processing speed which
are achieved by the modern graphic cards of the NVIDIA
type. It also considers the level of the computational errors
that registered with FFT and inverse FFT at different plans for
their implementation and lack (the absence) of distortion type
jitter. A comparison of the speed of the spectral data analysis
using different plans and different drivers was carried out on
three types of graphic cards NVIDIA GeForce GTX 210,
660M, 680. Also the comparison of the speeds and the
processing errors was carried for the mentioned graphic cards
and CPU Intel Core I7. Some inconsistencies were found in
the recommended technical documentation, which is given on
the official NVIDIA website, with parameters of the really
(actually) working FFT and inverse FFT plans.
The aim of this work is to determine the possibility of
using NVIDIA graphic cards for spectral analysis and
processing of radio astronomy data in real time (''on the fly'').
Currently, the great demands are made to information
analysis in real time or ''on the fly''. The demands are do to the
fact that in the Moore's Law [1] saturation is not observed.
The achieved density of p-n junctions cause the increase of
both the clock frequency of digital devices and their capacity.
In line with these trends, and, according to the formula of
Shannon-Hartley, the throughput capacity of existing and
planned registration channels, transmission and processing of
the information grows [2, 3].