8-GPU AI Training Server
8-GPU air-cooled training platform for LLM and vision workloads.
- 8× RTX 5090
- GPU
- 32× DDR5, PCIe 5.0
- Memory
- 104.8 TFLOPS FP16
- Compute
| GPU | 8× RTX 5090 |
|---|---|
| Memory | 32× DDR5, PCIe 5.0 |
| Compute | 104.8 TFLOPS FP16 |
8-GPU air-cooled AI training server with 32× DDR5 memory and PCIe 5.0 host interfaces, delivering 104.8 TFLOPS FP16 dense compute per node.
The G6400 packages eight top-bin consumer-class GPUs with server-grade DDR5 memory and redundant power in a 6U chassis, aimed at cost-sensitive training clusters where per-GPU TCO matters more than single-node peak.
Full mechanical, environmental and ordering documentation for this model is available on request. The specification below consolidates the published catalog data; contact the Foredge team for a complete datasheet package and region-specific ordering codes.
8× RTX 5090 GPUs
Dense FP16 training throughput per rack.
32× DDR5, PCIe 5.0
Memory bandwidth and host I/O for large model checkpoints.
Air-cooled 6U
Standard datacenter air handling — no liquid loop required.