Nvidia (NVDA) has launched CUDA 13.1 and CUDA Tile, which the Jensen Huang-led company said is the most substantial advancement to the platform since its release about 20 years ago.
“This exciting innovation introduces a virtual instruction set for tile-based parallel programming, focusing on the ability to write algorithms at a higher level and abstract away the details of specialized hardware, such as tensor cores,” said Nvidia engineers Jonathan Bentz and Tony Scudiero in a blog post.
CUDA is a parallel computing platform and programming model created by Nvidia. It helps developers increase the speed of their applications by harnessing the computing power of graphics processing units.
The new tile-based programming option allows developers to have “fine-grained control over” how their code is executed, particularly across multiple GPU architectures.
“Tile-based programming enables you to program your algorithm by specifying chunks of data, or tiles, and then defining the computations performed on those tiles,” Bentz and Scudiero noted. “You don’t need to set how your algorithm is executed at an element-by-element level: the compiler and runtime will handle that for you.”
CUDA Tile is available in Python. Nvidia plans to release a version compatible with C++ at some point in the future.
“It’s just come out in Python, and I think that maps beautifully to the AI development framework,” said Stephen Jones, Nvidia’s CUDA architect. “We are intentionally releasing the CUDA Tile frontier language as an open-source project.”