The gdal2tiles.py is very useful, but it is some slow when I want 10 or up levels. I think the bottleneck is I/O, so I want to how solve the problem by GPU?

  • 2
    GPU won't help with I/O, a GPU can only access data already in RAM via the PCIX bus (I assume, unless you're using a PCI or AGP graphics card), either way irrespective of GPU hardware, even if you could use it like a co-processor, it will not help I/O; if your disc access is your bottleneck then invest in a RAID or SSD, depending on your budget and data size, to speed up your cache process. – Michael Stimson Apr 4 '18 at 5:31
  • The gdal2tiles aims to generate map tiles of different levels,and it is about 20,000 files when 10 levels.It mainly includes raster pixel rezoom and rewrite to disk.Is there any solutions or technology to speed up it except for the hardware?I know there is a maptiler (maptiler.com) ,but it is not free – qxang Apr 4 '18 at 6:40
  • Please ask a new question about that IF there is not one already. Please be aware that there are multiple forks of gdal2tiles, eg for multiprocessing. – bugmenot123 Apr 4 '18 at 15:51
  • If your total data size is less than half your available RAM you could try a RAMdrive, it creates a physical drive inside your existing RAM and is very fast.. you could also turn off your pagefile if swaps are swamping I/O; Use Windows Resource Monitor to see what file(s) are being written, if pagefile.sys appears with excessive R/W bytes it could be swamping your process - this is a Windows parameter, not available in Linux that I'm aware of. Avoid processing over a network. If you use a RAMdrive or turn off pagefile it is important to have nothing else running. – Michael Stimson Apr 4 '18 at 21:31

As you say, the bottleneck for tiles generation is usually I/O. GPUs are meant for computationally expensive calculations with rather low I/O. So adding GPU processing will not help for things like this.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.