Sign up for our newsletter and get the latest HPC news and analysis.
Send me information from insideHPC:


OEMs Join Nvidia-Certified Program – Systems Pre-tested for AI Workloads

Dell Technologies, GIGABYTE, Hewlett Packard Enterprise (HPE), Inspur and Supermicro are among 11 systems makers engaged in an Nvidia certification program, announced this morning, designed to test hardware across a range of AI and data analytics workloads, including jobs that require multiple compute nodes and tasks that only need part of the power of one GPU.

The idea behind the program is to boost confidence among OEM customers that their systems can support use case needs such as deep learning training and inference, machine learning algorithms, video analytics and network and storage offload. The systems are optimized to run AI applications from the NGC catalog, Nvidia’s listing of GPU-optimized applications.

Nvidia said that to date, 14 servers from six systems makers have passed certification tests and that they are among nearly 70 systems from 11 system makers engaged in the program.

“Nvidia-Certified systems deliver the performance, programmability and secure throughput enterprise AI needs,” the company said in a blog today. “They combine the computing power of GPUs based on the Nvidia Ampere architecture with secure, high-speed Nvidia Mellanox networking.”

The company said the following systems, which use Nvidia A100 Tensor Core GPUs, have been certified:

  • Dell EMC PowerEdge R7525 and R740 rack servers
  • GIGABYTE R281-G30, R282-Z96, G242-Z11, G482-Z54, G492- Z51 systems
  • HPE Apollo 6500 Gen10 System and HPE ProLiant DL380 Gen10  Server
  • Inspur NF5488A5
  • Supermicro A+ Server AS -4124GS-TNR and AS -2124GQ-NART

Nvidia said OEMs certify the systems using Nvidia Mellanox cables, switches and network cards, such as ConnectX-6 InfiniBand or Ethernet adapters and BlueField-2 DPUs. Nvidia said the adapters support multiple layers of security from a “hardware root of trust  at boot time” to connection tracking for applications. Systems were certified using either an NVIDIA Mellanox 8700 HDR 200G InfiniBand switch or the Mellanox SN3700 Ethernet switch.

“AI models to sift through that data have grown in size by nearly 30,000x in just five years,” Nvidia product,” Nvidia NGC product manager Adel El Hallak said in the company’s blog, “driving the need for accelerated computing. And the diversity of models and workloads using them continues to expand, so businesses need the flexibility of GPUs. The rising tide of data and the expanding AI models to sift through it are spawning an exponential increase in network traffic both in the data  center and at the network’s edge. To cope, companies need a secure, reliable and high-speed infrastructure that scales efficiently.”

Leave a Comment

*

Resource Links: