Profile cover photo
Profile photo
Wolfgang Viechtbauer
94 followers
94 followers
About
Posts

Post has attachment
Comparing the time to fit a complex model (using the metafor package) with vanilla R vs MRO+MKL. Huge difference, even when using just one core. Note that this is really a comparison between the internal BLAS of R versus MKL (vanilla R versus MRO is not the issue here).

It makes sense that MKL would be more efficient, because most of the heavy lifting in the model fitting involves matrix algebra. What surprised me is that the largest gains come from switching to MKL, even if one just sticks to a single core. Throwing more cores at MKL does speed things up, but not nearly as much. And there doesn't seem to be any benefit from more than 7-8 cores.

In the end though, MKL with 8 cores was approximately 15 times faster than vanilla R. Quite a difference.
Photo

Post has attachment

Post has attachment
Interesting. I've tried R with the Intel MKL a while ago and the speedup was considerable.

Post has attachment
I'll try this when I run my next Monte Carlo simulation study.

Post has attachment
Just got an invitation (= spam) to submit an article to the Journal of International Research in ... Windings?!?
Photo

Post has attachment
That's mean =) (and just in case: don't do this!)

Post has attachment
Pretty neat!
Photo

Post has attachment
I don't know why, but I find it very funny that the English version of this article (https://en.wikipedia.org/wiki/Whipcracking) is quite a bit more comprehensive than the German one (https://de.wikipedia.org/wiki/Goa%C3%9Flschnalzen). But then again, there is the Bavarian one: https://bar.wikipedia.org/wiki/Goa%C3%9Flschnoizn.

Post has attachment
Depending on how much of a stats geek you are, this may eat up all the time you had planned in for writing that article/chapter/review/proposal/etc today. Don't say I didn't warn you!

Post has attachment
Wait while more posts are being loaded