ホーム/ イベント/ SC26 Supercomputing Conference
Industry Event 開催予定

SC26 Supercomputing Conference

Meet IP Infusion at the Broadcom booth, 4626, at SC26 in Chicago to discuss lossless 800G Ethernet AI fabrics built with OcNOS-DC.

Nov 17 - Nov 19, 2026
McCormick Place, Chicago, IL
Booth 4626 (Broadcom)
SC26

Visit us at the Broadcom booth

IP Infusion will be at the Broadcom booth, 4626, at SC26, the International Conference for High Performance Computing, Networking, Storage, and Analysis, at McCormick Place in Chicago. The conference runs from November 15 to 20, 2026, and the exhibit hall is open from November 17 to 19.

  • Tuesday, November 1710 am to 6 pm
  • Wednesday, November 1810 am to 6 pm
  • Thursday, November 1910 am to 3 pm

Sources: dates, venue and exhibit hours from the SC26 exhibits page; booth 4626 from the SC26 exhibitor list and floor plan.

What to talk to us about

Our engineers will be at the booth to discuss OcNOS AI Fabric: OcNOS-DC on qualified open switches for GPU clusters.

  • Lossless RoCEv2: PFC per priority, ECN marking for DCQCN-capable NICs, and PFC deadlock detection and recovery.
  • Dynamic load balancing moves flowlets off congested uplinks to distribute collective traffic across the fabric.
  • Rail-optimized designs on qualified 800G Broadcom Tomahawk 5 switches: the Edgecore AIS800-64D and the UfiSpace S9321-64E.
  • Sizing your cluster: leaf, spine and super-spine counts, with links, transceivers and cables.
  • Designs for clusters that span several buildings, with 800G ZR/ZR+ coherent optics in the spine ports of those switches.
Rail-optimized GPU fabric with OcNOS-DC Two GPU servers connect one NIC to each of four rail leaf switches; every leaf connects to two spine switches. OcNOS-DC runs on the leaf and spine switches with PFC, ECN marking, dynamic load balancing and PFC deadlock detection and recovery. DCQCN congestion control, the RDMA endpoint and the collective library run on the NICs and hosts. スパインスイッチ スパインスイッチ Rail 1 leaf Rail 2 leaf Rail 3 leaf Rail 4 leaf GPU サーバー one NIC per rail GPU サーバー one NIC per rail OcNOS-DC, switch side PFC per priority ECN マーキング 動的ロードバランシング PFCデッドロック検出 and recovery NIC and host DCQCN輻輳制御 RDMA endpoint, collective library
In a rail-optimized fabric, each GPU NIC connects to its own rail leaf. OcNOS-DC runs the switch side: PFC, ECN marking and dynamic load balancing. DCQCN congestion control runs on the NICs.

Why talk to us

  • Bought as OcNOS Systems, the switches ship with OcNOS-DC preloaded and qualified per switch model and ASIC stepping, with one PO, one support contract and one RMA path from IP Infusion.
  • Your NICs and GPUs stay your choice. OcNOS-DC runs the switch side, and the collective library and RDMA endpoint run on the NIC and host.
  • One OcNOS family runs from the cell site to the AI cluster.

Before the show

The VM is a lab image for learning the CLI and control-plane configuration. It does not show hardware features, scale or performance, so AI fabric features such as PFC and dynamic load balancing need qualified hardware.

See you at booth 4626 in Chicago

Discuss your cluster size and rail design with our engineers at the Broadcom booth, or book a time before the show.