Re^5: Perl in data science: could a grant from Perl foundation be useful?

I think https://chrisarg.github.io/Killing-It-with-PERL/2024/07/07/The-Quest-For-Performance-Part-II-PerlVsPython.md.html et al make a powerful argument that Perl (especially with PDL) can be a powerful, and high-performance, adjunct to other programming languages or tools in data science, and other science.

I created complementary demonstrations for the article on the web, running on the CPU or GPU.

CPU results captured on AMD Threadripper 3970X.

$ perl pdl_cpu.pl
==================================================
In place in (Perl PDL - CPU multi-threaded)
==================================================
   33974.886 盜
   37286.043 盜
   36175.966 盜
   32786.131 盜
   34714.937 盜
   32924.175 盜
   33201.933 盜
   33097.029 盜
   33265.114 盜
   33904.076 盜

$ python numba_prange.py 
==================================================
In place in (Python Numba - CPU multi-threaded)
==================================================
   30696.392 盜
   37623.167 盜
   32857.656 盜
   26406.765 盜
   26831.150 盜
   26624.441 盜
   26728.153 盜
   26698.828 盜
   26748.657 盜
   26932.001 盜

$ python taichi_cpu.py 
[Taichi] version 1.8.0, llvm 15.0.4, commit 37a05638, linux, python 3.
+10.13
==================================================
In place in (Python Taichi - CPU multi-threaded)
==================================================
[Taichi] Starting on arch=x64
   76603.651 盜
   45389.175 盜
   37540.913 盜
   29339.075 盜
   28975.725 盜
   29037.952 盜
   30524.731 盜
   29897.928 盜
   30224.562 盜
   28925.419 盜
[download]

GPU results captured on NVIDIA GeForce RTX 3070.

$ python cupy_blas.py 
==================================================
In place in (Python CuPy - GPU cuBLAS)
==================================================
    6246.805 盜
      24.557 盜
      17.881 盜
      17.643 盜
      17.643 盜
      17.643 盜
      17.643 盜
      17.405 盜
      17.405 盜
      17.405 盜

$ python numba_cuda.py 
==================================================
In place in (Python Numba - GPU CUDA)
==================================================
     182.629 盜
      40.054 盜
      30.279 盜
      28.133 盜
      27.895 盜
      27.657 盜
      27.418 盜
      26.941 盜
      26.941 盜
      26.703 盜

$ python taichi_cuda.py 
[Taichi] version 1.8.0, llvm 15.0.4, commit 37a05638, linux, python 3.
+10.13
==================================================
In place in (Python Taichi - GPU CUDA)
==================================================
[Taichi] Starting on arch=cuda
   15263.557 盜
      51.975 盜
      28.133 盜
      25.034 盜
      24.080 盜
      26.464 盜
      26.226 盜
      24.557 盜
      24.080 盜
      28.133 盜
[download]

Comment on Re^5: Perl in data science: could a grant from Perl foundation be useful? Select or Download Code

Replies are listed 'Best First'.

Re^6: Perl in data science: could a grant from Perl foundation be useful?
by marioroy (Prior) on Aug 04, 2024 at 04:54 UTC

@AmesingFlank tried Taichi on Javascript targeting WebGPU. The Javascript demonstrations work well in Google Chrome. It requires the following arguments. Doing so will cause Chrome to emit a message, "You are using an unsupported command-line flag: --enable-unsafe-webgpu. Stability and security will suffer."

--enable-unsafe-webgpu --enable-features=Vulkan
[download]

Enabling WebGPU on Linux:

I enable WebGPU in Chrome DEV, and it still doesn't work

[reply]
[d/l]


Problems? Is your data what you think it is?
	PerlMonks