CUDA is a parallel computing and programming platform introduced by NVIDIA for general computing of GPUs. It helps developers speed up computing applications by leveraging the potential of GPUs.

The CUDA Toolkit has everything you need for creating GPU-accelerated applications in popular languages like C, C++, Python, MATLAB, Fortran, etc. The toolkit also consists of GPU-accelerated libraries, development tools, a compiler, and the CUDA runtime.

Features of CUDA 11 – the Latest Version

• Support for the NVIDIA Ampere GPU architecture
• Multi-Instance GPU (MIG) partitioning capability. It’s suitable for cloud service providers (CSPs) who want to improve GPU utilization.
• New third-generation Tensor Cores to speed up mixed-precision matrix operations on different data types, such as Bfloat16 and TF32.
• APIs for task graphs, fine-grained synchronization, L2 cache residency control, and asynchronous data movement.
• Performance optimizations for linear algebra, matrix multiplication, and FFTs.
• Updated Nsight product toolkit for tracing, debugging, and profiling CUDA applications.
• Full support on major CPU architectures.