Today NVIDIA announced CUDA-X HPC, a collection of libraries, tools, compilers and APIs that helps developers solve the world’s most challenging problems. Similar to CUDA-X AI announced earlier this year, CUDA-X HPC is built on top of CUDA, NVIDIA’s parallel computing platform and programming model.
CUDA-X HPC includes highly tuned kernels essential for high-performance computing (HPC). GPU-accelerated libraries for linear algebra, parallel algorithms, signal and image processing lay the foundation for compute-intensive applications in areas such as computational physics, chemistry, molecular dynamics, and seismic exploration.
The ever expanding list of CUDA-X HPC libraries are regularly extended and fine tuned to take advantage of new algorithmic innovations like mixed precision operations. Linear algebra libraries in the list include industry-best BLAS, Math, and SOLVER libraries that offer extensive functionality and flexibility for programmers. Libraries for optimized tensor primitives (cuTENSOR), fast fourier transforms (cuFFT), performance primitives for image and signal processing (NPP), parallel algorithms and data structures (Thrust), and multi-GPU scaling (NCCL) are some of the other components included in CUDA-X HPC.
From fluid dynamics and weather simulation, to computational chemistry and bioinformatics, HPC applications span across many domains. Developing these applications requires a robust programming environment with highly optimized domain specific libraries.
Also part of CUDA-X HPC are NVIDIA Nsight developer tools that provide class-leading GPU debugging and profiling. Nsight systems, a system-wide low-overhead performance analysis tool, helps developers identify system wide bottlenecks and Nsight Compute is an interactive kernel profiler for CUDA applications.
Compilers with support for popular languages such as C/C++, Python and FORTRAN make CUDA-X HPC the go to solution for HPC developers building a new application or accelerating existing ones.
Adoption and Availability
CUDA and CUDA-X HPC are used to accelerate over 600 HPC applications across a multitude of domains, on a variety of hardware solutions. Applications built on CUDA-X HPC can be deployed everywhere, including small IoT devices, desktops, data centers, cloud, and supercomputers. It is immediately available to over 1.3 million registered developers on developer.nvidia.com. Developers can also access CUDA-X HPC as containerized software stacks from the NVIDIA NGC software hub.
