[R-sig-Geo] Graphic CPU usage in R

[Roger Bivand]

On Tue, 31 Dec 2019, Rich Shepard wrote:

> On Tue, 31 Dec 2019, maurizio marchi wrote:
>
>>  Anyway I was wondering whether any other ways were available. My question
>>  concern how to speed up old R codes involving GIS procedures and mainly
>>  using rgdal, raster, biomod2, dismo, sp or other Spatial packages.
>
> Maurizio,
>
> How many cores are in the CPU of your machine? AMD processors have two
> threads per core (e.g., the Ryzen7 in my desktop has 8 cores and 16
> threads). Programs need to be compiled to use multiple threads and you need
> libraries such as mesa or opengl to take advantage of that.
>
> Also, how much memory is installed on that system? More is always better.
>
>>  I was wondering if this could be used with R. In other words I would like
>>  to solve the issue from the beginning, opening an R session from terminal
>>  running on the GPU instead of on CPU(s).
>
> Something else for you to consider is that there are two types of video
> cards: those designed for gamers and those designed for technical work. An
> explanation of the differences (focused on nVidia's products) is here:
> <https://www.quora.com/What-is-the-different-between-gaming-GPU-vs-professional-graphics-programming-GPU>.
>
> There are multiple facturs involved so it's not a simple solution. Of
> course, if you have a long spatial model running you can start it using
> screen and it will continue running even after you log out as long as the
> computer is running.

GPU's are where the action was about ten years ago, but are not now. Many 
of the spatial packages that can benefit from multiple processors already 
facilitate their use, but often inter-process communication is the 
bottleneck, not per processor computation. A report from ten years ago is: 
https://papers.ssrn.com/sol3/Delivery.cfm/SSRN_ID1690584_code1391513.pdf?abstractid=1690584&mirid=1

Now, look to the stars and gdalcubes and many others, where the data are 
held in the cloud, and processing may be assigned to cloud nodes, with 
only target resolution output needing to be downloaded. The cloud nodes 
may actually be GPUs, but for the user this is transparent.

There are plenty of R packages accessing GPUs, described on the HPC task 
view: https://cran.r-project.org/view=HighPerformanceComputing.

Hope this clarifies,

Roger

>
> Hope this helps,
>
> Rich
>
> _______________________________________________
> R-sig-Geo mailing list
> R-sig-Geo using r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>

-- 
Roger Bivand
Department of Economics, Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: +47 55 95 93 55; e-mail: Roger.Bivand using nhh.no
https://orcid.org/0000-0003-2392-6140
https://scholar.google.no/citations?user=AWeghB0AAAAJ&hl=en