What is this?
Taichi Lang is an open-source language for high-performance GPU programming embedded in Python: you write Python-like code and its JIT compiler turns it into efficient parallel GPU or CPU machine code.
Taichi Lang is an open-source language for high-performance GPU programming embedded in Python: you write Python-like code and its JIT compiler turns it into efficient parallel GPU or CPU machine code.Is it a Python library or its own language?Both, in a way: it's a separate language that lives inside Python — it shares almost the same syntax and integrates with NumPy and PyTorch, while the compiler itself is written in C++.
Both, in a way: it's a separate language that lives inside Python — it shares almost the same syntax and integrates with NumPy and PyTorch, while the compiler itself is written in C++.Do I need an NVIDIA GPU to run it?No — plain x64 and ARM CPUs work too. For GPU acceleration Taichi supports most mainstream APIs: CUDA, Vulkan, OpenGL 4.3+, and Apple Metal. WebAssembly support is experimental.
No — plain x64 and ARM CPUs work too. For GPU acceleration Taichi supports most mainstream APIs: CUDA, Vulkan, OpenGL 4.3+, and Apple Metal. WebAssembly support is experimental.Any gotchas before I try it?Python support stops at 3.10 (64-bit only), and the nightly package may crash because it is not fully tested. Quantized computation and WebAssembly are still experimental.
Python support stops at 3.10 (64-bit only), and the nightly package may crash because it is not fully tested. Quantized computation and WebAssembly are still experimental.Who is it for — is it a graphics thing or an AI thing?Mostly compute-heavy work: real-time physical simulation, numerical computation, augmented reality, AI, vision and robotics, and visual effects in films and games, plus general-purpose parallel computing.
Mostly compute-heavy work: real-time physical simulation, numerical computation, augmented reality, AI, vision and robotics, and visual effects in films and games, plus general-purpose parallel computing.What are SNodes and why would I care?SNodes are Taichi's generic data containers for composing hierarchical, multi-dimensional fields — the mechanism behind spatially sparse computing, which matters for large numerical simulations.
SNodes are Taichi's generic data containers for composing hierarchical, multi-dimensional fields — the mechanism behind spatially sparse computing, which matters for large numerical simulations.Does it support differentiable programming?Yes, it's one of the headline features: kernels can produce gradients, which DiffTaichi uses for differentiable physical simulation and optimization in machine learning research.
Yes, it's one of the headline features: kernels can produce gradients, which DiffTaichi uses for differentiable physical simulation and optimization in machine learning research.Where can I see examples or learn it?The README links Taichi Lang examples and advanced examples plus an Awesome Taichi list, and there are lectures: a SIGGRAPH 2020 course on YouTube and Bilibili and the GAMES 201 advanced physics engine course.
The README links Taichi Lang examples and advanced examples plus an Awesome Taichi list, and there are lectures: a SIGGRAPH 2020 course on YouTube and Bilibili and the GAMES 201 advanced physics engine course.Can Taichi code run outside Python?Yes — there are AOT (ahead-of-time) deployment demos and a tutorial for running compiled kernels beyond the Python runtime.
Yes — there are AOT (ahead-of-time) deployment demos and a tutorial for running compiled kernels beyond the Python runtime.Which topics is it in?parallel-computing, graphics-programming, vulkan-api, jit, gpu-programming, scientific-computing, opengl, computer-graphics, numpy, programming-language, simulation, gpu-acceleration
parallel-computing, graphics-programming, vulkan-api, jit, gpu-programming, scientific-computing, opengl, computer-graphics, numpy, programming-language, simulation, gpu-acceleration