Designed to unparreled performance and flexibility

Falcon II

  • PCIe Gen4 Box (16 GT/s)
  • Composable Rack-scale Solution
  • Support up to 16 PCIe accelerator GPU
  • Support up to 4 hosts
  • Overview
  • Specifications
  • Resources

Future-proof PCIe Gen4 Speed

As the surge of transitions from PCIe 3.0 to PCIe 4.0, Falcon II provides instant benefit by aggregating the traffic of multiple PCIe 3.0 accelerators and connecting to the server at full PCIe 4.0 speeds. It is ready for PCIe 4.0 accelerators as long as they become available.

FC_front_GPU
FC_front_GPU

Flexible resource sharing for Tailoring the Resource As Your needs

Up to 4 hosts to share 16 devices on demand. With Falcon II you can save up to a day in setup time alone. Just remove the resource from one host and add it to another host. The resource are re-arranged at right time at right place. For an IT administrator, that’s easy and intelligent. This cohesive hardware and software solution lets you spend more time focused on insight, instead of scrapping together components.

Unparalleled level of visibility across your AI Environment

H3 provides a control center for IT administrators to deploy resources quickly and monitor system health and usage at a glance. Starting from the dashboard, IT administrators could get real-time performance and usage. Within few clicks, the resource could be transferred from one idle server to a busy server. The one-pane view helps the user get needed information quickly and precisely. The simplified report delivers a detailed, easy diagrams and tables to management to track down the source of performance issues and preempt future service disruptions.

FC_front_GPU
FC_front_GPU

Optimize AI performance by moving data quickest ever

Falcon II integrates up to 16 GPUs as one powerful HPC solution and provide a revolutionary data movement mechanism, NVMe to GPU directly, to boost unparalleled GPU performance. When applications ask for data, the “NVMe to GPU” function will re-direct this request to NVMe DMA engine to move the data directly from NVMe to GPU, eliminating unnecessary memory copies, dramatically lowering CPU overhead, and reducing latency, resulting in significant performance improvements in data transfer times for applications.

Software Features

Features
  • Dynamical GPU allocation among multiple hosts
  • Device surprise add and remove
  • Device peer to peer (GPU p2p and NVMe to GPU)
  • Port type configuration (Upstream / downstream)
  • Support port bifurcation
  • Support multifunction PCIe devices
  • System Management
  • H3 Center for AI cluster management
  • Instant connection topology diagram
  • At a glance dashboard for quick info
  • Single pane resource management page
  • Predictive health monitor
  • Proactive performance and usage report
  • Authentication and access control
  • Management Interface

    Redfish®, RESTful API, GUI

    Supported Browsers

    Google Chrome, Microsoft Edge, Mozilla Firefox

    Hardware Specification

    Model Falcon II
    PCIe Device Slots Sixteen(16) PCIe G4 x16 FHFL slots (Support GPU cards)
    Host Interface Four(4) PCIe G4 x16 by host adapter and CDFP cable
    BMC/mCPU Aspeed AST2500
    PCIe Switch Broadcom PEX 88096 PCIe 4.0 switch
    Power Three(3) 3000W PSU, hot-swap, 2+1 redundant
    Fan Four(4) 80x80mm, hot-swap, redundant
    LEDs
  • System Power LED
  • System Fault LED
  • UID LED
  • Drawer Power LED
  • AIC Status LED
  • Dimension (mm) 4U; 174.8 (H) x 448 (W) x 835 (D) mm
    Weight 30 Kgw (empty chassis)

    Compatible Device

    Accelerator
  • Nvidia Tesla® P100, V100; Quadro® RTX6000, RTX8000
  • AMD Radeon™ Pro V340; AMD Radeon Instinct™ MI50, MI60
  • Storage
  • Intel® P4500, P4600, P4800X series
  • Samsung PM1725a, PM1725b series
  • Network Adapter
  • Mellanox ConnectX®-4, ConnectX®-5
  • Compatible Host OS

    RHEL / CentOS 7.2 ~ 7.6
    Ubuntu 16.04 LTS
    18.04 LTS
    Windows Windows 10 Pro (Build 1903)

    Resources

    Document