8× H100
NVIDIA HGX H100 8-GPU Platform
Complete NVIDIA HGX H100 board with 8× H100 SXM5 80GB GPUs, fully connected via NVLink / NVSwitch. 640GB aggregate HBM3, enterprise BMC, and InfiniBand-ready I/O — the reference platform for multi-node LLM pretraining and large-batch fine-tuning.
Specifications
- GPUs
- 8× NVIDIA H100 SXM5 80GB
- Total GPU Memory
- 640GB HBM3
- Interconnect
- NVLink 4.0 + NVSwitch
- Aggregate FP8
- Up to ~32 PFLOPS sparse
- Networking
- InfiniBand / Ethernet ready
- Form Factor
- HGX baseboard (OEM chassis)
- Power
- ~10.2 kW GPU domain
- Target
- Multi-node training clusters
Related products
Frontier 4019
Higher-density Blackwell system for larger local workloads.
AED 153,400
Frontier 4018
Compact Blackwell workstation for local inference and fine-tuning.
AED 76,700
Frontier 4010
Entry Blackwell system with 64GB unified memory.
AED 31,500
H100 PCIe 80GB
Genuine Hopper PCIe accelerator for dual-slot servers.
AED 114,000