The Falcon 5012 supports two scenarios: 10 GPUs or 8 GPUs + 4 Network cards, configurable in Standard Mode or Advanced Mode. Advanced Mode, requiring a Premium License, enables flexible, high-performance deployments.
| Application scenarios | Standard Mode | Advanced Mode |
| 10 GPUs | a. 2 Hosts + 10 GPUs | b. 4 Hosts + 10 GPUs |
| c. 6 Hosts + 8 GPUs |
| Application scenarios | Standard Mode | Advanced Mode |
| 8 GPUs + 4 Network cards | d. 2 Hosts + 8 GPUs + 4 Network cards | e. 4 Hosts + 8 GPUs + 4 Network cards |
| f. 6 Hosts + 8 GPUs + 2 Network cards | ||
| g. 8 Hosts + 8 GPUs |
| Configuration | Chassis-wide configuration | Per-switch configuration |
| Host port + Device port | 2 Hosts + 10 GPUs | 1 Host + 5 GPUs |
In the 10 GPUs scenario, each host can independently access the GPUs connected to its switch under Standard Mode.
Figure 1. Block diagram of Standard Mode architecture topology with 2 x16 hosts and 10 GPUs
In the 10 GPUs application scenario under Advanced Mode, two configurations are available.
| Configuration | Chassis-wide configuration | Per-switch configuration |
| Host port + Device port | 4 Hosts + 10 GPUs | 2 Hosts + 5 GPUs |
In Advanced Mode, each switch supports 2 hosts and 5 FHFL devices. With 2 switches, the system can connect 4 hosts and 10 FHFL devices.
Figure 2: Block diagram of Standard Mode architecture topology with 4 x16 hosts and 10 GPUs
| Configuration | Chassis-wide configuration | Per-switch configuration |
| Host port + Device port | 6 Hosts + 8 GPUs | 3 Hosts + 4 GPUs |
External expansion slots can be connected to retimers to support additional hosts. In this configuration, each switch supports 3 hosts and 4 FHFL devices, allowing 2 switches to connect a total of 6 hosts and 8 FHFL devices.
Figure 3. Block diagram of Advanced Mode architecture topology with 6 x16 hosts and 8 GPUs
| Configuration | Chassis-wide configuration | Per-switch configuration |
| Host port + Device port | 2 Hosts + 8 GPUs + 4 Network cards | 1 Host + 4 GPUs + 2 Network cards |
In the 8 GPUs + 4 Network cards scenario under Standard Mode, each host can independently access 4 GPUs and 2 Network cards on its connected switch.
Figure 4. Block diagram of Standard Mode architecture topology with 2 x16 hosts and 8 GPUs + 4 NICs
| Configuration | Chassis-wide configuration | Per-switch configuration |
| Host port + Device port | 4 Hosts + 8 GPUs + 4 Network cards | 2 Hosts + 4 GPUs + 2 Network cards |
In Advanced Mode, each switch at least supports 2 hosts, 4 FHFL devices, and 2 single-slot devices, enabling the system to connect 4 hosts, 8 GPUs, and 4 Network cards.
Figure 5. Block diagram of Advanced Mode architecture topology with 4 x16 hosts and 8 GPUs + 4 NICs
| Configuration | Chassis-wide configuration | Per-switch configuration |
| Host port + Device port | 6 Hosts + 8 GPUs + 2 Network cards | 3 Hosts + 4 GPUs + 1 Network card |
Users can designate one of the external expansion slots to connect additional hosts. In this configuration, each switch supports 3 hosts, 4 FHFL devices, and 1 NIC, enabling the chassis to support a total of 6 hosts, 8 FHFL devices, and 2 single-width devices.
Figure 6. Block diagram of Advanced Mode architecture topology with 6 x16 hosts and 8 GPUs + 2 NICs
| Configuration | Chassis-wide configuration | Per-switch configuration |
| Host port + Device port | 8 Hosts + 8 GPUs | 4 Hosts + 4 GPUs |
Users can designate external expansion slots to connect additional hosts. In this configuration, each switch supports 4 hosts, and 4 FHFL devices, enabling the chassis to support a total of 8 hosts, and 8 FHFL devices.
Figure 7. Block diagram of Advanced Mode architecture topology with 8 x16 hosts and 8 GPUs
For more information, please contact sales@h3platform.com for license purchase.
If you want to apply for any product display, please write a form and we will contact you after receiving the message.
