NVIDIA (NVDA.US) has introduced CUDA 13.1 and CUDA Tile, with CEO Jensen Huang describing it as the most significant advancement for the platform since its launch nearly 20 years ago. Engineers Jonathan Bentz and Tony Scudiero stated in a blog post, "This exciting innovation introduces a virtual instruction set for module-based parallel programming, focusing on enabling higher-level algorithm development while abstracting hardware-specific details like tensor cores."

CUDA, developed by NVIDIA, is a parallel computing platform and programming model that allows developers to leverage GPU computing power to accelerate applications. The new tile-based programming option provides developers with "fine-grained control" over code execution, particularly across multiple GPU architectures.

The engineers explained, "Tile-based programming lets you write algorithms by defining data blocks (tiles) and specifying computations performed on them. Developers no longer need to manage execution at an element-by-element level—the compiler and runtime handle these details."

CUDA Tile is also available in Python, with plans to release a C++-compatible version in the future. NVIDIA CUDA architect Stephen Jones noted, "It’s now implemented in Python, which aligns perfectly with AI development frameworks. We intend to release CUDA Tile’s cutting-edge language as an open-source project."