The Falcon 6048 utilizes a GPU-accelerated memory/storage architecture to expand effective GPU capacity through high-IOPS NVMe SSD integration. By orchestrating seamless data movement between GPU HBM and NVMe storage via a high-speed PCIe fabric, the system maximizes compute utilization and maintains ultra-low latency for massive-scale AI and HPC workloads.
Powered by 2 Broadcom PCIe Gen6 switches and 44 E1.S NVMe SSDs, Falcon 6048 achieves up to 200 million IOPS, delivering ultra-low latency and massive parallel throughput ideal for AI training, inference, prediction and large-scale data analytics.
Optimized for NVIDIA cuVS and the WholeGraph framework, the Falcon 6048 delivers significant performance gains for vector database operations and GNN workloads. It provides native GPU acceleration for the entire embedding-to-retrieval pipeline. This synergy, combined with support for large-scale GNN training, allows developers to orchestrate highly responsive AI systems capable of handling massive, unstructured datasets.
| Management Interface | Redfish®, RESTful API, GUI and CLI |
|---|---|
| System Management |
|
| Model Name | Falcon 6048 |
|---|---|
| BMC | Aspeed AST2600 Advanced PCIe Graphics & Remote Management Processor |
| CPU |
|
| PCIe Switch | Broadcom PEX 90144 PCIe 6.0 Switch |
| Memory |
|
| NVMe SSD |
|
| PCIe Slot |
|
| Power | 3200W (3+1 redundant) power supply 80+ Titanium |
| Fan |
Ten (10) 60x56mm fans Hot swap |
| LAN |
|
| Environmental Spec. |
|
| Dimension | 3U; 129.6 (H) x 438(W) x 850(D) mm |
| Accelerator |
|
|---|---|
| Network Cards | NVIDIA ConnectX-7, ConnectX-8, ConnectX-9 |
With the PCIe 6.0 switch at its core, Falcon 6048 delivers exceptional throughput and scalability. It supports up to 2 NVIDIA H200 GPUs, 4 NVIDIA ConnectX-8 NICs, and 48 NVMe SSDs (depending on available PCIe slots), along with 1 CPU and 8 DDR5 memory modules in the gpu-accelerated server.
This architecture is purpose-built for ultra-high IOPS performance and ultra-low-latency small-granularity data access, enabling efficient processing of massive datasets dominated by small, random I/O patterns—commonly seen in AI training, inference, real-time prediction, and data-intensive workloads.
If you want to apply for any product display, please write a form and we will contact you after receiving the message.
