1 Topic by Carsten Simon

  • Carsten Simon
  • Member
  • Offline

Topic: Perfomance increase in processing FTMS spectra by GPU?

Hi again,

I recently have encountered some performance problems of mmass. My main question is whether this can be solved by using mmass with the graphics board.

When it comes to processing of peak-rich/ full-scan FTMS spectra (in my case, stemming from an Orbitrap system), especially recalibration becomes tedious due to re-processing of the whole spectrum. It already helps  a bit to zoom into a nominal mass to minimize perfomance loss for spectrum display. Anyhow, when big peak lists are used and many (> 10) peaks have to be deselected, processing is slow.

The IT department at our institute suggested to use the GPU (graphical processing unit, i.e., the graphics board) instead of the CPU for calculations as it might perform much better in these kind of tasks.

The question is, can mmass be used that way? Is it compatible with GPU usage?

Thanks in advance for your help.
Carsten

2 Reply by RMaytum

  • RMaytum
  • Member
  • Offline

Re: Perfomance increase in processing FTMS spectra by GPU?

To get mmass to use GPU processing would require the numeric processing parts of the program to be re-written to use the GPU. Unfortunately this is not a simple process as the GPU uses a different specific programming language (CUDA in the case of Nvidia GPUs). As mmass is apparently largely written in Python, this is likely to take a considerable amount of work.

Having said that, as the source code is available, it would be entirely possible for somebody to look at this and re-write the appropriate numeric processing parts to take advantage of GPU acceleration.

Just needs somebody with both the skill and the time...

So the answer is, yes it's possible, but it would take somebody an amount of time and effort to do it. It's not just a matter of recompiling the current code with appropriate options selected.

Regards,

Robin