Document
Falcon 5012 is a 4U PCIe GPU expansion box that brings together a PCIe Gen5 Fabric, GPU Expansion, Multi-Host Connectivity, and Integrated Management to extend accelerator capacity beyond the host server. Built around a high-bandwidth PCIe Gen5 switching fabric, the system supports high-density GPU expansion, flexible multi-host connectivity, and integrated management through Redfish®, RESTful API, and GUI-based interfaces.
With configurable host-to-device topologies, Falcon 5012 gives infrastructure teams greater flexibility to scale GPU resources for growing generative AI and accelerated computing requirements.
Scale GPU capacity beyond the host server as generative AI workloads grow. Falcon 5012 provides a flexible way to add accelerator resources while preserving existing host infrastructure and supporting evolving compute requirements.
Connect host systems and accelerator resources through a high-bandwidth PCIe Gen5 switching fabric. The platform is designed to support efficient device connectivity and high-speed data movement across demanding GPU-based infrastructure.
Gain centralized visibility into GPU health, PCIe status, topology, and resource assignments. Integrated management interfaces help administrators monitor system conditions and manage connected accelerator resources through a unified operational view.
Explore how Falcon 5012 connects host systems and PCIe devices through a flexible PCIe Gen5 architecture. Select an element below to see how each part fits into the system.
A PCIe Gen5 switching fabric connects host systems with accelerator resources in the expansion chassis.
Establish a PCIe connection between the host and supported devices housed outside the server.
See how Falcon 5012 supports flexible configurations across host systems, GPUs, and other PCIe devices. Select a configuration below to view its topology.
Host and GPU count varies with the selected configuration.
Network card count varies with the selected configuration.
| System Specifications | Falcon 5012 |
|---|---|
| Form Factor | 4U, 19-inch rackmount |
| PCIe Fabric | Broadcom PEX 89144 PCIe 5.0 Switch |
| Supported Devices | GPU, FPGA, NIC, and standard PCIe add-in cards |
| GPU Form Factor | Dual-slot, up to 10.5-inch length |
| GPU Power Support | Up to 600 W per GPU |
| BMC | AST2500 |
| Management CPU | Intel Atom x86 CPU |
| Power Supply | 4 × 2100 W or 3200 W, hot-swappable |
| Power Redundancy | 2+2 or 3+1 |
| Cooling | 8 × 80 × 80 mm dual-rotor hot-swappable fans |
| Dimensions | 175 × 448 × 650 mm (H × W × D) |
| Operating Temperature | 10°C–35°C |
| Host Connectivity | Standard Mode | Advanced Mode |
|---|---|---|
| Host Interface | 1 × PCIe 5.0 x16 CDFP | Up to 4 × PCIe 5.0 x16 |
| Host Adapter | HBA required | HBA required |
| Cable | 1 m CDFP cable | 1 m CDFP cables |
| Configuration | Standard host-to-device topology | Flexible multi-host topology |
| Management | Capabilities |
|---|---|
| Management Interfaces | Redfish®, RESTful API, GUI |
| GPU Management | GPU resource assignment, hotplug, P2P |
| PCIe Fabric Management | Host/device port configuration and switch topology management |
| Monitoring | GPU health, utilization, PCIe link status, performance, and error monitoring |
| Topology Management | Host-to-device topology and assignment visibility |
| System Protection | Temperature monitoring, overheat protection, firmware secure boot |
| Centralized Management | H3 Management Center and system monitoring dashboard |
| Access Control | Role-based authentication and user management |
| Device Category | Supported Devices / Examples |
|---|---|
| NVIDIA Data Center GPUs | H200, H100 |
| NVIDIA RTX GPUs | RTX 6000 PRO, RTX A5000 |
| NVIDIA GeForce RTX | RTX 4090, RTX 4080, RTX 3090 (blower type) |
| FPGA / Other PCIe Devices | AMD Alveo U50 and standard PCIe add-in cards |
If you want to apply for any product display, please write a form and we will contact you after receiving the message.