Balanced tier · NVIDIA H100 80GB
NVIDIA H100 80GB
The workhorse of the entire industry. H100 pairs Hopper-class compute with clean NVLink fabric for production training, fine-tuning, and inference — the node your roadmap can count on.
ArchitectureHopper
Memory80 GB HBM3
Bandwidth3.35 TB/s
FP16 dense989 TFLOPS
Specifications
H100, by the numbers.
| Spec | NVIDIA H100 SXM (Hopper) |
|---|---|
| Architecture | NVIDIA Hopper (GH100) |
| Memory | 80 GB HBM3 |
| Memory bandwidth | 3.35 TB/s |
| FP8 (dense) | 1,979 TFLOPS |
| FP16 / BF16 (dense) | 989 TFLOPS |
| FP64 (dense) | 34 TFLOPS |
| NVLink | 900 GB/s per GPU |
| Form factor | 8-GPU HGX baseboard, 4U node |
| Power envelope | 700 W per accelerator |
| Boot & security | TPM 2.0, verified boot, encrypted RAM |
Use cases
Where H100 shines.
Production fine-tuning
Full-model fine-tunes with 7B–70B parameter ranges, comfortably inside a single node or a small fabric.
Self-hosted inference
Serve mid-size open models with room to spare for growing context windows and continuous batching.
Training at 8–32 GPUs
Data-parallel and FSDP workloads that need reliable bandwidth without paying flagship prices.
Ready to push the frontier?
Provision bare-metal GPUs in minutes, scale capacity on demand, and get a real engineer on call — not a ticket queue.