In its drive to accelerate supercomputing and AI development via GPUs, Nvidia said this week it is expanding the number of application containers available on its GPU cloud.
The expansion seeks to combine cloud-based GPU processing with agile deployment of distributed applications on shared systems, the company noted in a blog post announcing nine new HPC containers. The GPU specialist said it now offers 35 containers for deep leaning, HPC and visualization applications.
Containers “have become a crucial tool in deploying applications on a shared cluster and speeding the work, especially for researchers and data scientists running AI workloads,” noted Chintan Patel, an Nvidia marketing manager. The company claims more than 27,000 users of its container registry.
Nvidia said it has added nine new HPC and visualization containers to its GPU cloud since last November’s Supercomputing Conference. One of them, PGI Compiler, is intended to help developers build HPC applications using multicore CPUs and Nvidia’s Tesla GPUs. Compilers and other tools in the HPC container can be used to develop what Patel called “performance-portable” HPC applications based on its CUDA Fortran parallel programming and other languages.
“Supercomputing has a dire need [for simplifying] the deployment of applications across all the segments,” Patel noted. “That’s because almost all supercomputing centers use environment modules to build, deploy, and launch applications.”
The GPU specialist is betting containers can help break the installation logjam. The elimination of installations means “users can pull the containers themselves and deploy an application in minutes compared to waiting for days for the advisory council to agree on an install and go through the actual process,” Patel added.
Nvidia said it tested it GPU cloud containers on its DGX servers along with Nvidia graphics processors offered by cloud providers, including Amazon Web Services and the Google and Oracle clouds.
Nvidia is among a growing list of vendors focused on harnessing maturing application containers for AI and HPC workloads. Among them is startup Sylabs Inc., which is seeking to bring the open-source HPC container technology Singularity to mainstream enterprises. The startup that emerged from stealth mode earlier this year cites the rapid adoption of Singularity by scientific users and the transition to AI and other data-driven workloads.