What can be the maximum frequency of the processed signal on a desktop computer?

Question

Signal processing

Greetings,
I do not know how correct my question is, I will give specific characteristics. Having a regular desktop computer (with a core-i5 processor), and receiving signals via Ethernet (via UDP protocol), what is the maximum frequency of signals that can be processed (if it depends on the IDE and the programming language, then most likely QT or MVSC), I mean modulation /demodulation, filtering, FFT, etc.? Has anyone processed signals with a frequency of 30 MHz? And is it possible to somehow increase the processing speed through the use of CUDA?

Answer 1

The bottleneck here will be the network, the network driver, the network adapter. Things will not even come to CUDA, CUDA is redundant here.
But the server network adapter , which has its own hardware on board, and does not shift work to the operating system driver, unlike cheap consumer-grade analogues, makes sense.
And the quality of the switch.
Where to dance nginx.org/ru/docs/events.html
After you have solved the problem with the network, then it makes sense to optimize further:
CUDA, programming language, etc.
Maybe you have each package requires 10 minutes of thinking on CUDA. From the question, it's not clear.
Constant load or floating, whether it is possible to implement due to queues - it is not clear from the question.
Will you be able to load data on CUDA at such a speed?
As for whether it is possible to load on CUDA and receive an answer for many small pieces - it is doubtful.
MHz? 30 millihertz ? i.e. 0.03 times per second??? what is the problem?
if we are talking about MHz - megahertz - then write correctly.
To begin with, an educational program for you:
SI prefixes
As for UDP, the speed also depends on the size of the packet .

Answer 2

With 1 gigabyte ethernet, more than 90 MB per second cannot be squeezed out, this is approximately 40 MHz of the analog signal bandwidth with 8 bit din. range. If you start simultaneously on two ethernets, then you get 80 MHz of the signal bandwidth.
Such a stream can theoretically be processed in real time if processing meant - buffer FFT without overlap, DDC and decimation without frequency response correction. If in real time it is enough just to write to the screw, and then process it slowly, then you can do anything. Any FFT with any overlap, any filtering, etc.