CUDA Tile Programming Now Available for BASIC!

2026-04-01 · Source: NVIDIA Technical Blog · Field: Technology & Digital — Software Development & Engineering, Artificial Intelligence & Machine Learning · Depth: Intermediate, medium

Summary

NVIDIA has released cuTile BASIC, an experimental implementation of the CUDA Tile programming model for the BASIC language, building on the CUDA Tile IR specification. This initiative aims to make fine-grained GPU parallelism accessible to a broader range of programming ecosystems, including legacy languages. cuTile BASIC allows developers to write tile kernels in BASIC, abstracting explicit thread management and simplifying GPU acceleration. It targets seasoned developers familiar with BASIC, enabling them to port legacy applications or explore AI and scientific computing on NVIDIA GPUs with compute capability 8.x, 10.x, 11.x, or 12.x, NVIDIA Driver R580+, CUDA Toolkit 13.1+, and Python 3.10+. The release includes examples for vector addition and matrix multiplication (GEMM), demonstrating how BASIC code can compile to cubin and execute on GPUs, achieving performance levels previously unimaginable for the language.

Key takeaway

For developers with existing BASIC codebases or those exploring GPU programming with a simplified syntax, cuTile BASIC offers a unique path to accelerate applications. You can now leverage NVIDIA GPUs for tasks like vector addition and matrix multiplication without deep CUDA C++ knowledge, potentially revitalizing legacy projects or enabling new AI explorations. Ensure your system meets the hardware and software requirements, including a compatible GPU and CUDA Toolkit 13.1+.

Key insights

cuTile BASIC brings GPU acceleration to the BASIC language via CUDA Tile, simplifying parallel programming for legacy applications.

Principles

• Tile-based programming simplifies GPU parallelism.
• Language openness expands GPU acceleration access.

Method

cuTile BASIC kernels define data tiling and mathematical operations; the system handles thread management, compiling BASIC to cubin for GPU execution.

In practice

• Port legacy BASIC applications to NVIDIA GPUs.
• Implement AI algorithms like GEMM in BASIC.
• Use `pip install git+https://github.com/nvidia/cuda-tile.git@basic-experimental`.

Topics

• CUDA Tile
• cuTile BASIC
• GPU Acceleration
• BASIC Programming
• Parallel Computing

Code references

• NVIDIA/cuda-tile
• nvidia/cuda-tile

Best for: Software Engineer, Machine Learning Engineer, AI Engineer

Related on AIssential

• cuTile.jl Brings NVIDIA CUDA Tile-Based Programming to Julia — CUDA Tile simplifies GPU programming by abstracting hardware details, enabling tile-based operations for high-performance kernels.
• Evaluating CUDA Tile for AI Workloads on Hopper and Blackwell GPUs — CuTile offers high productivity and peak performance on datacenter Blackwell, but lacks portability and consistent performance across other architectures.
• NVIDIA CUDA 13.3 Enhances GPU Development with Tile Programming in C++, Compiler Autotuning, and Python Updates — CUDA 13.3 streamlines GPU development with high-level abstractions and performance optimizations across C++ and Python ecosystems.
• Introducing hipThreads: A C++ - Style Concurrency Library for AMD GPUs — hipThreads enables incremental GPU acceleration of C++ `std::thread` code using familiar concurrency patterns and persistent multi-fiber execution.
• NVIDIA Foundational Technology Montage I GTC 2026 Edition — CUDA provides a unified architecture and extensive libraries for accelerated computing across diverse scientific and engineering fields.
• StitchCUDA: An Automated Multi-Agents End-to-End GPU Programing Framework with Rubric-based Agentic Reinforcement Learning — StitchCUDA automates end-to-end GPU programming using a multi-agent framework with rubric-based reinforcement learning to enhance code optimization.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by NVIDIA Technical Blog.