Dedicated H100 & A100 Nodes
Single-tenant GPU nodes with no virtualisation overhead. Full hardware access, InfiniBand interconnects, and per-hour billing — spin up in under 2 minutes.
H100 Bare-Metal Node
GPU-NODE-07 • UK-SOUTH
GPU Memory
80 GB HBM3
Interconnect
3.2 Tbps IB
Built for Serious Workloads
Everything a training run or high-throughput inference fleet needs, with nothing between you and the silicon.
True Bare-Metal
Single-tenant nodes with direct hardware access. No hypervisor tax, no noisy neighbours — 100% of the GPU is yours.
InfiniBand Fabric
Ultra-low-latency interconnects between nodes for distributed training that actually scales across your cluster.
Local NVMe & Zero Egress
Fast local NVMe scratch space for datasets and checkpoints, plus free data transfer in and out. Move terabytes without a bill shock.
Simple Per-Hour Pricing
World-class hardware at transparent rates. Start and stop whenever you like.
| GPU Model | vCPUs | Memory | VRAM | Hourly Rate | |
|---|---|---|---|---|---|
NVIDIA H100 (80GB) Hopper Generation | 26 | 256 GB | 80 GB HBM3 | $2.85/hr | |
NVIDIA A100 (80GB) Ampere Generation | 12 | 96 GB | 80 GB HBM2e | $1.65/hr | |
NVIDIA L40S (48GB) Graphics & AI Generation | 8 | 64 GB | 48 GB GDDR6 | $0.95/hr |
Save up to an additional 45% with yearly reserved instance contracts.
Talk to Enterprise SalesGPU Cloud — FAQ
Are these GPUs shared or dedicated?
Dedicated. Every node is single-tenant, bare-metal — no hypervisor, no noisy neighbours. 100% of the GPU is yours for the duration you run it.
How much cheaper is GravityStack than hyperscaler GPU pricing?
On-demand H100 and A100 nodes are typically 45-77% below comparable hyperscaler on-demand rates, and reserved contracts add up to another 45% savings on top.
Can I cluster multiple GPU nodes for distributed training?
Yes. Nodes connect over InfiniBand for low-latency, high-throughput multi-node training. Multi-node clusters are available on request.
How fast can I launch a GPU node?
Under 2 minutes from request to a running node with local NVMe scratch space attached.
Your GPUs Are Waiting
Launch a dedicated H100 or A100 node in under 2 minutes and pay only for the hours you use.