CUDA 11 adds the below new features, which are essential to achieve the performance promised by the Ampere architecture.
- New third-generation Tensor Cores to accelerate mixed-precision matrix operations on different data types, including TF32 and Bfloat16
- Programming and API for task graphs, asynchronous data movement, fine-grained synchronization, L2 cache residency control
- Performance optimizations in CUDA libraries for linear algebra, FFTs, matrix multiplication, JPEG decoding, and more
- Support heterogeneous architectures with GPUs including X86_64, Arm64 server, and POWER architectures
- CUDA C++ enhancements:
- Compiler performance and usability improvements
- New link-time optimization capabilities
- Support for new host compilers and language standards including C++17
- Introducing Parallel C++ STL support using libcu++ and integration of CUB as a CUDA C++ core library in the Toolkit
- Operating System support updates
- Async-copy is offered as an experimental feature in CUDA 11
- Licenses free of charge
- Nsight Tools - Developer tools for tracing, debugging, analyzing, and profiling CUDA applications
- Nsight Systems 2020.3 & Nsight Compute 2020.1 - These releases are available now online, and with CUDA 11 on 7/8, and add support for the new NVIDIA Ampere GPU Architecture, and improved CPU feature parity for Power and ARM Server Base System Architecture.
- Nsight Systems 2020.3 - Including support for MPI, OpenACC, and OpenMP, as well as improvement in the CLI, and complex data mining capabilities
- Nsight Compute 2020.1 - Visualizations for Roofline Analysis, A100 memory system, and data compression, as well as theoretical peak (speed of light) metrics
- Cuda-gdb - Improved load times, debug information, and parallel cuda-gdb sessions
- New Compute Sanitizer - A functional correctness checking tool that helps you identify memory and threading errors in your CUDA code
- IDE integrations - NVIDIA® Nsight™ Visual Studio Edition and NVIDIA® Nsight™ Eclipse Edition and much more.
Try NVIDIA A100 GPU on E2E Cloud here
