Enterprise AI at-Scale with InfiniBand Storage – HPCwire

The world’s most advanced research and product development centers in manufacturing, life sciences, and financial services demand high performance storage platforms to execute their applications effectively. The pressure to maintain a competitive advantage across industries has increased with the addition of data-intensive applications brought on by AI and deep learning (DL).

NVIDIA® InfiniBand is the primary network of choice for high performance computing (HPC) and supercomputing, accelerating seven of the top ten supercomputers in the world. Infiniband is also the only network that can meet the data-intensive demands of HPC and AI workloads at scale. That’s why storage partners like Pavilion are leveraging NVIDIA InfiniBand and NVIDIA Magnum IO™ GPUDirect® Storage (GDS) technologies to meet the scale, bandwidth, and latency requirements needed to realize the potential of today’s storage solutions.

As the volume and speed of data creation for HPC workloads continue to accelerate, the ability for a storage solution to ingest data has become as critical as the need to serve it to the GPU. The Pavilion HyperParallel Data Platform™ ingests data with extreme performance and leverages GDS to deliver high throughput with ultra-low latency and low CPU utilization for both block and file workloads. With Pavilion, capacity and performance can scale across systems to support any size data set.

Organizations now have new opportunities to drive significant AI and machine learning (ML) value with the GA release of NVIDIA Magnum IO GDS. This exciting technology, when combined with ultra-low latency, high performance HDR 200Gb/s InfiniBand-based storage, enables customers to process data faster and get more value from their compute and I/O investments.

GDS takes advantage of GPUDirect and RDMA to provide a direct data path between storage and GPU memory, resulting in massive throughput at ultra-low latency—while simultaneously reducing the impact on system resources. This enables customers to process more data in less time than would otherwise be possible and extract more value from their data.

InfiniBand is an important technology for Pavilion and is fully supported by the Pavilion HyperParallel Data Platform. The combination of NVIDIA Magnum IO GDS and InfiniBand enables unprecedented IO performance and low latency with minimal impact on system resources. These technologies work together to ensure the best performance for AI, HPC, and IO-intensive applications.

To maximize performance, both the compute element, which includes NVIDIA DGX™ systems or other NVIDIA GPU-based platforms, and the storage solution, such as the Pavilion HyperParallel Data Platform, must support GDS. Pavilion is proud to be an NVIDIA partner providing full support for GDS and InfiniBand.

With up to 40 100Gb/s or 10 200Gb/s InfiniBand ports per HyperParallel Data Platform, Pavilion is unique as an NVIDIA partner supporting GDS for both block and file data on the first day of GDS general availability.

The GDS release is all about delivering the performance customers need to achieve faster insights from their data to drive business. With NVIDIA and Pavilion, customers can train their AI models faster, researchers can find better cures, law enforcement can detect more threats, scientists can make new discoveries, and more. What are you ready to achieve?

To learn more about how Pavilion works with NVIDIA Magnum IO GDS and InfiniBand, attend this webinar.

To schedule a technical discussion, click here.


Reggie Reynolds, Principal Product Marketing Manager, NVIDIA

Reggie Reynolds has over 25 years of experience in the data storage industry including 15 plus years in HPC storage serving engineering and management roles in Engineering, Quality Assurance and Technical Support. Reggie joined NVIDIA as Principal Product Marketing Manager in October 2020.

Keith Parker, Director of Product Marketing, Pavilion

Keith Parker has over 25 years experience in the data storage and networking industries in channel and product marketing roles. His deep industry knowledge spans topics including cloud, AI/ML, SAN, NAS, object storage, HCI, Fibre Channel, iSCSI, NFS, replication, and backup. Keith joined Pavilion in 2020.

Spread the love

Leave a Reply

Your email address will not be published.