Is your HPC networking solution able to meet your converged workload needs for today and tomorrow?
HPE Slingshot provides a modern, high-performance interconnect for HPC and AI clusters that delivers high-bandwidth and low-latency for HPC, ML, and analytics applications by bringing together the specialized requirements of HPC-optimized fabrics with the ubiquity of Ethernet. This delivers a converged infrastructure with high-performance on both HPC simulation codes and native IP applications, with efficient scalable access to data sources.
Building on Cray's specialized silicon, HPE Slingshot delivers consistent performance and low latency under load and at scale. This prepares you to efficiently service increasingly diverse users taking advantage of your HPC resources, and do so without overprovisioning bandwidth or deploying multiple systems to avoid congestion on your most demanding workloads.
SKU # R4K40A
HPE Slingshot switches provide 25.6 Tbps of bi-directional bandwidth with 64 ports each capable of 200 Gbps to deliver high bandwidth and low latency, overcoming traditional Ethernet approaches to buffering and error handling that create queueing latency and hinder low-latency RDMA applications.
Congestion management addresses the challenges of short-lived, small packet flows typical in HPC applications. Each switch detects congestion, identifies its causes, provides real-time feedback to its peers, and limits the injection rate from the congestion source until congestion clears.
Fine-grained adaptive routing, based on HPE Slingshot’s ability to track real-time information on load across each switch-to-switch path, can dynamically re-route traffic to balance loads (rather than sticking only to a predetermined path) to improve network performance and bandwidth utilization.
The HPE Slingshot interconnect supports highly configurable quality-of-service (QoS) that provides the ability to control how network bandwidth is allocated to different classes of traffic and applications.
The high-radix 64 port HPE Slingshot switches enable low-diameter topologies that reduce the quantity of HPC network equipment, cabling, and lower power and cooling costs, while scaling up to 250,000 endpoints with fewer than three switch-to-switch hops.
The Dragonfly topology delivers cost effective performance. HPC clusters using HPE Slingshot typically require substantially fewer optical cables without compromising performance. This reduces the cost and power required for an HPC interconnect solution.
HPE Slingshot’s fine-grained adaptive routing and congestion management technologies in silicon enable achieving sustained bandwidth at high utilization on demanding HPC applications without overprovisioning the network.
HPE Slingshot switches connect to the campus network using Ethernet, the standard for connectivity to data, without requiring gateway nodes. Very high scale connectivity can be achieved by aggregating multiple links from several HPE Slingshot switches to qualified edge routing switches.
HPE Slingshot NICs run standard IPv4 and IPv6 software using native software stacks, without encapsulation, alongside demanding high-performance RDMA that scales to large supercomputing scaled systems.
HPE Slingshot interconnect is offered with select third-party Ethernet RoCE NICs or the HPE Slingshot NIC that deliver 200 Gbps connectivity using HPE Slingshot Switches with very high performance RDMA and extensive hardware acceleration for MPI and SHMEM-based software.
The HPE Slingshot switch is offered as a direct liquid-cooled blade in the HPE Cray EX supercomputer, and a 1U top-of-rack form factor for HPE Apollo and ProLiant clusters, across a wide range of platforms with choice in CPU and GPU architecture.
HPE delivers full cluster solutions that reduce the complexity of building high performance HPC solutions. HPE Performance Cluster Manager software and HPE Cray Programming Environment software have built in capabilities to take advantage of HPE Slingshot performance and management features.
Hewlett Packard Enterprise provides factory integration and support services that make it easy for you to deliver high-performance clusters that are fully tested prior to deployment and supported as a cluster once deployed.
* Prices may vary based on local reseller.
