- GPU dynamic provisioning
- 4U 19” disaggregated GPU solution
- 8 dual slots for GPUs, 4 slots for NICs, and 2 slots for retimers
- 4 (x16) server host configuration
Features and Benefits
-
Composable infrastructure
- Compose and decompose GPU to unlock the trapped GPU and avoid overprovisioning.
- Flexible pools to rapidly meet the needs of variable application workloads.
-
Analytics to optimize utilization
- Provide IO performance and utilization rate for optimization of GPU.
- Optimize resource pool-- is it over/ under provisioned?
- Predict GPU failure based on AI.
-
Dynamically scale
- Pay-as-you-grow expansion, ensure maximum performance with scalable multi-chassis.
- Purchase and upgrade GPU independently without purchasing whole server.
-
Superior performance
- Support PCIe Gen4 GPU and backward compatible with PCIe Gen3.
- Protect your investment by future-proof PCIe Gen4 performance, 32GB/s, sub-micro latency.
Product
-
Falcon 5012
View Product -
Falcon 4205
- Four (4) GPU expansion box
- Support PCIe 4.0 x16 server host under standard mode
- Support up to 8 PCIe 4.0 x4 server hosts in advanced mode (beta version)
View Product -
Falcon 4210
- 5U, 8-Slot GPU expansion box
- Support up to 2 PCIe 4.0 x16 server hosts in standard mode
- Support up to 16 PCIe 4.0 x4 server hosts in advanced mode (beta version)
View Product -
Falcon 4005
- Tower, 4 GPU box
- Support 1 PCIe 4.0 x16 server hosts in standard mode
- Support up to 8 PCIe 4.0 x4 server hosts in advanced mode (beta version)
View Product -
Falcon 4010
- 5U, 8-Slot GPU box
- Support up to 2 PCIe 4.0 x16 server hosts in standard mode
- Support up to 16 PCIe 4.0 x4 server hosts in advanced mode (beta version)
View Product -
Falcon 4016
- 4U, 16-Slot GPU box
- Support up to 4 PCIe 4.0 x16 server hosts in standard mode
- Support up to 6 PCIe 4.0 x16 server hosts in advanced mode (beta version)
View Product
Software
-
Dynamically provision and scale PCIe devices with H3 Falcon series
Falcon series chassis creates shared pools of PCIe devices in disaggregated GPU chassis by PCIe cables. Users can draw GPUs needed and return PCIe resources into the pools when no longer needed without shutting down the GPU chassis or servers.
-
Increase effectiveness and accelerate operations with H3 Center
Embedded with Falcon series, H3 Center management UI or API automatically discovers devices that are added, moved, or removed from the system; adds them to its inventory; and alerting, dashboard, utilization, performance and health status.
-
Intelligence for resource optimal utilization and failure prediction
Built-in artificial intelligence, H3 Center provides a centralized analytics such as GPU utilization and performance; and a fault summary by continuous monitoring the errors for faster problem resolution and improved reliability.
