400G AI RoCE Data Center Networking
Ethernet Networking for HPC accelerates job completion times, maximize GPU use, and optimize IT resources.
400G AI RoCE Data Center Networking
This solution offers a perfect 400G interconnect solution for QSFP-DD switches and OSFP network cards, addressing the compatibility issues between different port encapsulations. And it is tailored according to the network topology of HPC architecture, including backend network, frontend network, out of band network and storage network, to meet various business needs.
Backend Network
Powered by NVIDIA® H100 GPUs and 400G Ethernet switches, the backend network offers higher bandwidth, lower latency and high throughput, enabling the construction of a high-performance network, which ensures high reliability and stability.
Frontend Network
The frontend fabric allows users to interact with the GPU and storage nodes to initiate training or inference workloads and to visualize their progress and results. It also provides an out-of-band path for NCCL (NVIDIA Collective Communications Library) collective communication.
Storage Network
Leveraging the speed, flexibility, and cost efficiencies of Ethernet, FS PicOS® switches provide a distributed storage network architecture with ideal scalability, optimized to deliver the highest levels of performance, industry-best latency and zero packet loss features.
Out of Band Network
With the advanced PicOS® software and AmpCon-DC management platform, FS switches empower customers to efficiently provision, monitor, manage, preventatively troubleshoot, and maintain the network while realizing higher utilization and reducing overall opex.
Product List
RoCE Network Delivers the Utmost Bandwidth with Extremely Low Latency
RoCE switches play a pivotal role in optimizing data center performance by leveraging RDMA technology to deliver high throughput, low latency, high stability and scalability benefits.
High Bandwidth
RoCE leverages the high speed and scalability of Ethernet networks to provide efficient data transfer. FS offers Ethernet switches from 100G to 800G, meeting high bandwidth needs and ensuring efficient and reliable data transfer.
Low Latency
RoCE networks use hardware acceleration and RDMA for low-latency Ethernet transmission. The latency of FS Ethernet switches can be as low as 0.4us, which reduces network congestion and improves bandwidth utilization.
High Stability
The solution adopts DLB redundant load technology, along with BGP GR, DLB FRR, and DLB Hash load balancing technology to ensure high network stability and reliability.
High Reliability
The switch supports PFC and ECN, which can significantly reduce network congestion and packet loss events, improve network reliability, and ensure business continuity and high availability of applications.
Cost-effective Solution for Simplifying Performance Complexity
Combined the benefits of Ethernet and IP, RoCE offers excellent interoperability and integrates seamlessly with existing Ethernet infrastructure, enabling efficient data transmission without significant hardware additions. This reduces construction and operational costs while simplifying data center management.
Interoperability
RoCE utilizes the existing Ethernet data center infrastructure, which enables efficient data transmission over existing Ethernet infrastructure without the need for extensive hardware additions or large-scale conversions, thereby reducing construction and operating costs.
Data Center Integration
RoCE can realize the integration of data center networks, integrating computing, storage and network in the same Ethernet infrastructure. This simplifies data center architecture and management, reducing operational complexity.
PicOS® Ensures Stable and Secure Network Operations
With PicOS®, deliver highly resilient, highly reliable, programmable networks that are leaner and more scalable than their monolithic predecessors.
Try PicOS-V free >
Full Interoperability
Supports NETCONF, SNMP, and gNMI protocols, provides API application programming interfaces, delivers interoperability with Cisco and other third-party infrastructure for progressive migration and upgrade.
Network Virtualization
Switches running PicOS® with both EVPN-VXLAN and MLAG can facilitate the establishment of networks that are nimbler, secure, scalable.
Hardened Security
Enables complex ACL without exhausting TCAM resources to flexibly control network traffic, ensuring data security without sacrificing hardware performance.
Network Reliability
PicOS® features a modular design for independent component operation, a dual-partition rollback for quick restoration, comprehensive L2/L3 features to enhance network reliability and availability.
AmpCon-DC Automated Configuration and Unified Management
AmpCon-DC management platform provides full Day 0 through Day 2+ capabilities to manage PicOS® data center switches, enabling provisioning, monitoring, troubleshooting, and maintenance for higher resource utilization and lower opex.
Free Trial Now >
Telemetry
Support telemetry for real-time network monitoring.
Topology Auto-Discover
Leverage topology auto-discover for physical network visual management.
Lossless Network Automation
Support lossless network automation for zero-packet-loss optimization.
Lossless Network O&M
Support lossless network O&M, monitoring the network in real time for proactive problem solving.
Overlay Auto Configuration
Support overlay-based auto configuration for fully automated and efficient business provisioning.
Underlay Auto Configuration
Support underlay-based auto configuration for simplified infrastructure management.
Resources
Featured Solutions
Ready to get started？
Have a question about purchasing a solution that you are interested in? Fill in the form and we'll respond in 1 working day.