What hardware limits plotting speed in R?

Question

I would like to increase the speed of plotting, and I am happy with (and have lots of code requiring) the R graphics and ggplot packages - so I am only interested in knowing how I can configure my system to speed up plotting.

Specifically:

1. Is the speed of plotting in R limited by the processor, memory, graphics card?
2. Are there particular hardware components or configurations would increase plotting speed?

---

Update: Answers to questions in comments:

specs: Ubuntu 11.04, intel Core Duo, 8GB ram, but I am more generally interested in wether the graphical computation or the graphical rendering is limiting, and if so, how can I use this information.

My plots have lots of objects, but I have no idea what the computational costs of plotting is. I don't do any specific analyses while plotting (I am plotting after completing any required analyses), although I understand that some is done 'on the fly', as when plotting a smoothed line or even translating data into locations.

Answer 1

Unless you have computer-intensive single plots, a great way to speed up multiple plotting is with parallel processing. For example, suppose you have a dataframe and you want to break it down by a certain variable (or variables) and do plots for each partition.

There are many ways to register a parallel backend so I won't go into that. See, for example, this vignette: http://cran.r-project.org/web/packages/doSMP/vignettes/gettingstartedSMP.pdf

Then check out the function ddply in Hadley's plyr package and use the .parallel = TRUE option. That's basically it. Then just do plotting normally.

Here's a self-contained example:

#this is the particular library I chose to register a parallel backend. There are others. See the new "Parallel R" book for details.
library(doMC)
registerDoMC()
getDoParWorkers() #This lists how many workers you have (hopefully more than 1!)

library(ggplot2)
ddply(
        mtcars, .variables = "vs", .fun = function(x) {
        #do your plotting now 
        example_plot <- ggplot(x, aes(y = mpg, x = wt)) + geom_point() + geom_smooth(se = FALSE)
        #save your plot
        ggsave(paste(x$vs[1],".pdf",sep = ""), example_plot)
        },
        .parallel = TRUE
)

This will save two files, 0.pdf and 1.pdf, which are the levels (ie the unique values) of the vs variable of the mtcars dataframe. If you broke it down by a variable country name then the files saved would be the names of the countries. 0.pdf and 1.pdf are as below:

Answer 2

As @Xu Wang points out, you can use parallelization to draw several plots at once.

So hardware wise, a powerful fast multi-core machine with plenty of RAM would help a bit.

If you want to plot a single plot with, say, 1 million circles in an x-y plot (scatter plot), then graphics hardware acceleration would be very beneficial.

But a fast graphics card only helps if the graphics devices in R support hardware acceleration. Currently they do not - and as @hadley points out, ggplot uses the standard graphics devices.

The rgl package apparently uses OpenGL to do 3D-graphics. Haven't tried it though. You might be able to use it to draw some plots more efficiently...

I have some experience creating fast interactive hardware accelerated plots (2d and 3d), and it can be magnitudes faster. The 2d-plots are actually harder to accelerate than the 3d ones... Probably not an easy thing to plug into R's current graphics device concept though.

UPDATE I just tried rgl and its plot3d with 1 million points. It is fully interactive (small fractions of a second to update) on my (rather powerful) laptop.

library(rgl)
x <- sort(rnorm(1e6))
y <- rnorm(1e6)
z <- rnorm(1e6) + atan2(x,y)
plot3d(x, y, z, col=rainbow(1000))