One hour
$2.69
1 GPU × 1h · compute only
NVIDIA Hopper cloud GPU
Rent an NVIDIA H100 GPU for training and inference.
Direct answer
NVIDIA H100 is available in two GPURento catalog configurations: H100 SXM with 80 GB HBM3 at $2.69 per GPU-hour and H100 NVL with 94 GB HBM3 at $2.59 per GPU-hour. Compare memory, form factor, interconnect and workload behavior before deploying from the funded dashboard in Paris or Frankfurt.
Reviewed by the GPURento infrastructure team · September 1, 2026
Cost scenarios
What one H100 SXM costs at the listed rate.
One day
$64.56
1 GPU × 24h · compute only
Seven days
$451.92
1 GPU × 168h · compute only
Average month · 730h
$1963.70
1 GPU × 730h · compute only
| Memory | 80 GB SXM · 94 GB NVL | Two Hopper configurations listed in the GPURento catalog. |
|---|---|---|
| Architecture | Hopper | Designed for transformer and accelerated computing workloads. |
| On-demand rate | $2.59 NVL · $2.69 SXM | Per GPU-hour; storage and optional networking are separate. |
| Available regions | Paris · Frankfurt | Both regions are exposed in the self-service dashboard. |
Best for
- Full or distributed model training
- FP8 inference where throughput matters
- Large fine-tuning jobs that need 80–94 GB VRAM
- CUDA workloads tuned for Hopper
Not the best fit when
- Small image jobs that fit a lower-cost RTX GPU
- Jobs that do not benefit from Hopper throughput
- Models that require more than 94 GB on one device without sharding
Decision method
What to verify before you deploy capacity.
The content below separates published specifications, GPURento catalog references and decisions that still require a workload benchmark.
01
Choose H100 when time-to-result matters
Hourly price alone does not determine job cost. A faster accelerator can cost less per completed training run when it shortens the wall-clock time enough. Compare the same model, precision, batch size, framework and checkpoint schedule before deciding.
- Estimate peak VRAM, not only model weights
- Measure samples or tokens per second on your own stack
- Include data loading, checkpoint and stopped-storage time
02
Know which H100 configuration you are comparing
H100 SXM provides 80 GB HBM3 in an HGX-oriented SXM form factor, while H100 NVL provides 94 GB HBM3 in a PCIe dual-slot form factor. NVIDIA specifies different memory bandwidth, power and NVLink characteristics. Choose from the complete server topology and software path, not memory alone.
03
Capacity and cost are explicit
Configure resources directly in the dashboard. Prepaid usage uses wallet credit; monthly rentals have a separate checkout.
Catalog shortlist
Relevant GPU options.
HopperAvailable
NVIDIA H100 SXM
80 GB HBM3 · $2.69/hr
Intensive training and FP8 inference
EU West · ParisEU Central · Frankfurt
HopperAvailable
NVIDIA H100 NVL
94 GB HBM3 · $2.59/hr
Large-model inference
EU West · ParisEU Central · Frankfurt
HopperAvailable
NVIDIA H200 SXM
141 GB HBM3e · $3.59/hr
Large-model inference and HPC
EU West · ParisEU Central · Frankfurt
AmpereAvailable
NVIDIA A100 80GB
80 GB HBM2e · $1.19/hr
Training, fine-tuning and HPC
EU West · ParisEU Central · Frankfurt
Questions
Clear answers, including the limits.
How much does it cost to rent an H100 on GPURento?+
The current catalog lists H100 NVL at $2.59 per GPU-hour and H100 SXM at $2.69 per GPU-hour. Storage, reserved IPs and optional networking can add cost.
What is the difference between H100 SXM and H100 NVL?+
The listed H100 SXM has 80 GB HBM3 in an SXM form factor, while H100 NVL has 94 GB HBM3 in a PCIe dual-slot form factor. Interconnect, bandwidth, power and system topology differ, so the better option depends on training or inference behavior.
Is H100 better than A100 for every workload?+
No. H100 is a stronger fit for recent transformer and FP8 workloads, but A100 can be more economical when a job already runs well on Ampere and finishes within the required deadline.
Can I deploy an H100 without contacting sales?+
Configure resources directly in the dashboard. Prepaid usage uses wallet credit; monthly rentals have a separate checkout.
Sources and scope
Last reviewed September 1, 2026. GPURento rates are current catalog references; provisioning state remains visible in the workspace, and manufacturer specifications do not substitute for workload benchmarks.
Continue the research
Deployment plan
Deploy H100 with the right constraints.
Create a workspace, fund the wallet and choose the region, image and storage configuration you actually need.