Dear list,
According to this presentation (
http://www.rinfinance.com/agenda/2010/MarkSeligman_Tutorial.pdf), gpuHclust
in the gputools package should be about 20x faster than clust, but after
some testing, I find it actually 4x slower. Here is the code:
test.data <- function(dim, num, seed=17) {
set.seed(seed)
matrix(rnorm(dim * num), nrow=num)
}
m <- test.data(1920, 9500)
library(gputools)
d <- gpuDist(m) # Euclidean distance
system.time(gpuHclust(d))
user system elapsed
28.230 41.283 69.727
system.time(hclust(d))
user system elapsed
6.517 0.260 6.790
The gpu on my machine is Telsa C5020 from NVIDIA.
Best regards,
Kaiyin ZHONG
------------------------------
FMB, Erasmus MC
k.zhong@erasmusmc.nl
kindlychung@gmail.com
[[alternative HTML version deleted]]