Simulation / Modeling / Design

NVIDIA Announces CUDA Toolkit 11

AI-Generated Summary

  • CUDA 11 adds support for the NVIDIA A100 GPU based on the NVIDIA Ampere architecture, enabling accelerated scale-up and scale-out AI and HPC data centers.
  • The release introduces third-generation Tensor Cores for mixed-precision matrix operations with TF32 and Bfloat16 data types.
  • Multi-Instance GPU virtualization and partitioning capabilities improve GPU utilization across workloads.
  • Library performance optimizations cover linear algebra, FFTs, matrix multiplication, and JPEG decoding.
  • Programming and API improvements include task graphs, asynchronous data movement, fine-grained synchronization, and L2 cache residency control.
  • Support extends to heterogeneous architectures with GPUs including X86_64, Arm64 server, and POWER platforms.

Next Steps

Powered by NVIDIA Nemotron. AI-generated content may summarize information incompletely. Verify important information. Learn more

CUDA is the most powerful software development platform for building GPU-accelerated applications, providing all the components needed to develop applications targeting every GPU platform. CUDA 11 introduces support for the new NVIDIA A100 based on the NVIDIA Ampere architecture, Arm server processors, performance-optimized libraries, and new developer tools and improvements for A100. 

Read our new technical developer blog, “CUDA 11 Features Revealed” for a deeper dive on the breadth of software advances, and more specific details about support for the new NVIDIA Ampere GPU architecture.  

CUDA 11 delivers the following capabilities and much more:

  • Develop for the NVIDIA Ampere GPU architecture including:
    • The new NVIDIA A100 GPU for accelerated scale-up and scale-out AI and HPC data centers
    •  Multi-GPU systems based on A100 such as DGX A100 and HGX A100
  • New third generation Tensor Cores to accelerate mixed-precision matrix operations on different data types, including TF32 and Bfloat16
  • Multi-Instance GPU virtualization and GPU partitioning capabilities for improved GPU utilization
  • Library performance optimizations for linear algebra, FFTs, matrix multiplication, JPEG decoding, and more
  • Programming and API improvements for task graphs, asynchronous data movement, fine grained synchronization, L2 cache residency control 
  • Enhancements to the Nsight developer tools family for tracing, profiling, debugging, and roofline analysis
  • 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 
  • Operating System support updates
  • Introducing libcu++, which brings heterogeneous C++ Standard Library facilities to CUDA
  • CUB, the popular toolkit for CUDA C++ kernel authors, is now included in the Toolkit

There are a lot of innovations and improvements to the CUDA Toolkit 11 – to learn more, here are a few resources:

Read the CUDA Features Revealed Developer Blog  >>

Register for CUDA New Features and Beyond Webinar >>

Discuss (0)

Tags