3 ms·
Completely agree. I like Bret's stuff and love Julia but that section seemed very weird and almost preachy. He seems a little out of his element because a lot o
by emcq 11y ago
Completely agree. I like Bret's stuff and love Julia but that section seemed very weird and almost preachy. He seems a little out of his element because a lot of physicists doing earth modeling use compiled languages like Fortran and C/C++ for the big simulations and python for the smaller stuff. Julia is cool though and a great substitute for the python stuff. It won't implicitly solve distributed computing and the computational problems people have to create large models. Unfortunately being closer to the metal can give big wins in terms of communication (mpi), avoiding memory copies, using simd compiler intrinsics and gpus, etc. and will continue to do so for the foreseeable future.
- tavert 11y ago> It won't implicitly solve distributed computing Why not? It's out to explicitly address distributed computing - http://docs.julialang.org/en/latest/manual/parallel-computing/ http://docs.julialang.org/en/latest/manual/parallel-computin... Julia is performance-competitive with C++ and Fortran, and productivity-competitive with Python. It's out to do both. Julia code compiles to LLVM and is incredibly "close to the metal" including SIMD, ref http://docs.julialang.org/en/latest/manual/performance-tips/?highlight=simd#performance-annotations http://docs.julialang.org/en/latest/manual/performance-tips/... and https://software.intel.com/en-us/articles/vectorization-in-julia https://software.intel.com/en-us/articles/vectorization-in-j..., and GPUs http://blog.maleadt.net/2015/01/15/julia-cuda/ http://blog.maleadt.net/2015/01/15/julia-cuda/ Try using Julia a bit before you talk about what it "won't solve."