7 SETI Program at the Medicina INAF Radioastronomy …
55
Fig. 7.4 Block diagram of the spectrometer hardware design
raising the computing power by one or more many-core accelerator boards (see
Fig. 7.4).
Considering budget limits, as DAQ, we evaluated both internal boards and external
adapters typically used in electronics measure applications as fast oscilloscopes, that
due to high volume production, have a much affordable price.
We tested many boards, both older ones connected to the PC by PCI or HighSpeed Universal Serial Bus (USB) buses and newer ones connected by PCI Express
or Super-Speed USB buses. In the tests of these devices, we observed that external
adapters were much more resilient against RFI than the internal boards. In a more indepth test of the latter, made without input signals and with input connectors closed
on the characteristic impedance, we observed in the spectra many sharp and narrow
lines at specific frequencies related to the PC clock value.
This situation, although not critical in SETI applications because the postprocessing algorithms remove these artifacts, leads to problems in other applications
where it is advisable to perform an accurate calibration of the instrument before
measurement.
About the accelerator boards, we evaluated Nvidia GPUs, AMD GPUs, and Intel
Xeon Phi boards. Our performance evaluations were focused on many algorithms
but mainly on FFT, and they were carried out using the manufacturer’s free available
SDKs. The test results on FFT computations showed better performances of Nvidia
GPUs with CuFFT software library [23] than the comparable (by price, computing
power, and semiconductor technology) products of other manufacturers. Both in FFT
and other algorithms test, results were generally similar to others in the scientific literature [24–27]. The latter usually reports Nvidia GPU computation tests on General
Purpose GPU boards, namely Tesla product line. These products had (and also today
55
Fig. 7.4 Block diagram of the spectrometer hardware design
raising the computing power by one or more many-core accelerator boards (see
Fig. 7.4).
Considering budget limits, as DAQ, we evaluated both internal boards and external
adapters typically used in electronics measure applications as fast oscilloscopes, that
due to high volume production, have a much affordable price.
We tested many boards, both older ones connected to the PC by PCI or HighSpeed Universal Serial Bus (USB) buses and newer ones connected by PCI Express
or Super-Speed USB buses. In the tests of these devices, we observed that external
adapters were much more resilient against RFI than the internal boards. In a more indepth test of the latter, made without input signals and with input connectors closed
on the characteristic impedance, we observed in the spectra many sharp and narrow
lines at specific frequencies related to the PC clock value.
This situation, although not critical in SETI applications because the postprocessing algorithms remove these artifacts, leads to problems in other applications
where it is advisable to perform an accurate calibration of the instrument before
measurement.
About the accelerator boards, we evaluated Nvidia GPUs, AMD GPUs, and Intel
Xeon Phi boards. Our performance evaluations were focused on many algorithms
but mainly on FFT, and they were carried out using the manufacturer’s free available
SDKs. The test results on FFT computations showed better performances of Nvidia
GPUs with CuFFT software library [23] than the comparable (by price, computing
power, and semiconductor technology) products of other manufacturers. Both in FFT
and other algorithms test, results were generally similar to others in the scientific literature [24–27]. The latter usually reports Nvidia GPU computation tests on General
Purpose GPU boards, namely Tesla product line. These products had (and also today