The Rosetta smallpt project: the same path tracer (modifications of Kevin Baeson's 99 line C++ path tracer) written in multiple languages.

Note: I deliberately chose for the same software design for all programming languages out of clarity and performance reasons (this can conflict with the nature of declarative/functional programming languages).

Current languages:

• C
  • single threaded
  • multi threaded (OpenMP)
• C++
  • single threaded
  • multi threaded (Windows Threads)
  • multi threaded (OpenMP)
• C#
  • single threaded
  • multi threaded
• CoffeeScript
• CUDA
• Erlang
• GLSL
• Haskell
• Java
• JavaScript
• J#
• Prolog
• Python 2
  • Python
  • Python + NumPy
• Python 3
  • Python
  • Python + NumPy
  • Python + Taichi
• Racket
• TypeScript

In order to give a very rough performance comparison, the wall clock time is measured of a single run to generate an image of resolution 1024x768 using 64spp. All single process - single threads experiments output the current progression to standard output. All multi process - multi threaded experiments do not output the current progression to standard output (this can change in the future).

System:

• OS: Windows 8.1 Pro 64 Bit
• CPU: Intel i7-4770K @ 3,50 GHz (TB @ 3,9 GHz) with 8GB DDR3 @ 1600 MHz
• GPU: NVIDIA GeForce GTX 970 with 4 GB GDDR5

Note: the random number generators used, differ between the languages and their implementations. It makes no sense to use the same random number generator in all implementations (i.e. C++ vs CUDA) and would introduce a huge burden to explicitly pass the state in languages such as Prolog.