NVIDIA
NVIDIA H100
80GB HBM3 Hopper Architecture — the enterprise benchmark for large-scale training and inference with NVLink scale-out
Reserved and on-demand, deployed in Thailand
$2.85/hr per GPU
All prices in USD, ex-tax.
Deploys high-density NVIDIA H100 SXM5 8-GPU HGX nodes with NVLink 4 interconnect and full root access in Tier III regional infrastructure — optimized for enterprise foundation model training, distributed fine-tuning, and high-throughput inference.
Memory
80GB HBM3
3.35 TB/s bandwidth
Architecture
Hopper
Family · NVIDIA Hopper
Compute (FP16)
7.9 PFLOPS (8-GPU, FP16 tensor)
BF16: 7.9 PFLOPS (8-GPU, BF16 tensor) · FP8: 15.8 PFLOPS (8-GPU, FP8 tensor)
Interconnect
NVLink 4 (900 GB/s) / NDR InfiniBand
TDP · 700W
Highlights
- Enterprise HGX H100 SXM5 architecture with native NVSwitch all-to-all GPU mesh
- NVLink 900GB/s in-node connectivity for multi-GPU training
- Production-ready CUDA, NCCL, and TensorRT-LLM stack for distributed LLM workloads
Software stack
CUDA 12.x
PyTorch 2.x
Triton
vLLM
TensorRT-LLM
NCCL
Best-fit scenarios
- Large-scale LLM training and fine-tuning
- High-throughput inference (TensorRT-LLM / vLLM)
- Mixed-precision scientific computing
Pricing by configuration
All available bare-metal configurations for this GPU. Reserved and committed-use discounts available.
On-demand starting at $2.85 /hr per GPU
| Configuration | GPUs | VRAM total | CPU | Memory | Storage | Network | $/hr per GPU |
|---|---|---|---|---|---|---|---|
Bare Metal — 8× H100 SXM Popular h100-bm-8 | 8 | 640GB | 192 vCPU (dual Intel Xeon) | 2 TB DDR5 | 8× 3.84 TB NVMe | 8× 400G NDR InfiniBand | $2.85 |
All prices in USD, ex-tax.
Ready to deploy NVIDIA H100?
Spin up a dedicated server in days, or talk to our sales engineers about custom configurations.