Any benchmark of Julia v1.0 vs older versions

General Usage Performance
61

I would add that using the Batch Mode we can also create a nice optimization for Broadcasting over the third dimension with Matrix Multiplication (More advanced use case, Jitting Matrix Multiplication which happens in a loop).

@Seif_Shebl, The link you provided Speed Up Small-Matrix Multiplication using New Intel® Math Kernel Library Capabilities is gold!

I would summarize it with this:

image
image1041×756 170 KB

@Elrod, Maybe you can replicate some of the Compact API tricks of MKL in your library for small matrices.

@ChrisRackauckas, have a look at this:

image
image1033×761 162 KB

Maybe the reason MATLAB gets better performance out of MKL is due to configuration.
It seems you need configure it correctly and pay attention to small tips (Not sure this is the case, but better check the integration).

1 Like
62

The problem with Julia Pro is it seems to be still using Julia v0.6

What about the garbage collector. How good is it? Any improvement through the different Julia versions?

63

Not to the GC per se but Julia 1.0 is much better at not allocating things in the first place.

64

What about Julia v1.2?

65

It does not exist yet :wink:

66

I know it hasn’t been publicly released yet, but some people is already playing with it:

67

You are basically talking about master at this point. It is not only that it is not released yet, even what will be in 1.2 is not yet fixed (technically, as of course it is possible to have a pretty good idea).