NVIDIA · Hopper Architecture
Rent NVIDIA H100 SXM in the Cloud
The flagship data center GPU for AI training. Industry standard for large language model training with 80GB HBM3 memory and NVLink interconnects.
VRAM
80 GB HBM3
Bandwidth
3,350 GB/s
FP16
990.0 TFLOPS
FP32
67.0 TFLOPS
TDP
700W
Architecture
Hopper
Cheapest On-Demand
$1.57/hr
Average On-Demand
$2.44/hr
Providers
7
Compare NVIDIA H100 SXM Cloud Pricing — 7 Providers
On-Demand
| Provider | Price / GPU / hr | Availability | Notes | Action |
|---|---|---|---|---|
Latitude.sh
Brazil
|
$1.57/hr CHEAPEST | Available | GPU VM, NVLink | Visit Provider |
Vultr
United States
|
$1.99/hr | Available | 24-month contract | Visit Provider |
Vast.ai
United States
|
$2.20/hr | Available | Marketplace avg | Visit Provider |
Massed Compute
United States
|
$2.35/hr | Available | PCIe | Visit Provider |
Novita AI
United States
|
$2.59/hr | Available | — | Visit Provider |
RunPod
United States
|
$2.99/hr | Available | Secure Cloud, SXM | Visit Provider |
DigitalOcean
United States
|
$3.39/hr | Available | HGX H100 | Visit Provider |
Spot / Preemptible
| Provider | Price / GPU / hr | Availability | Notes | Action |
|---|---|---|---|---|
|
Novita AI
United States
|
$1.49/hr CHEAPEST | Available | Spot, 1hr guaranteed | Visit Provider |
|
RunPod
United States
|
$2.69/hr | Available | Community Cloud | Visit Provider |
Reserved
| Provider | Price / GPU / hr | Availability | Notes | Action |
|---|---|---|---|---|
|
DigitalOcean
United States
|
$2.50/hr CHEAPEST | Available | 12-month, 8-GPU | Visit Provider |
Prices last verified: April 13, 2026
NVIDIA H100 SXM Technical Specifications
Manufacturer
NVIDIA
Architecture
Hopper
VRAM
80 GB HBM3
Memory Bandwidth
3,350 GB/s
FP16 (Tensor)
990.0 TFLOPS
FP32
67.0 TFLOPS
TDP
700W
Release Year
2023
Segment
Data center
Memory Type
HBM3
Best For
Compare with Other GPUs
See how NVIDIA H100 SXM stacks up against other popular cloud GPUs in specs, pricing, and availability.