NewH200 and MI355X clusters

GPU accelerated infrastructure for training, inference, and scale.

Reserve enterprise-grade NVIDIA and AMD compute by the hour, and deploy it in minutes.

tensor — deploy

$ tensor deploy --gpu h100 --count 8 --hours 72

· Reserving 8x H100 · 640GB HBM3

· Attaching 48 vCPU · 1TB NVMe

✓ Cluster ready in 94s

$ ssh root@203.0.113.24

GPU models

Max VRAM

288GB

Bandwidth

8 TB/s

From / hour

$2.50

Live in minutes

Pick a configuration, deploy, and get SSH access to a running cluster without a sales call.

Pay for the hours you reserve. No contracts, no minimum spend, no idle capacity to carry.

Hopper, Ampere, and CDNA accelerators side by side, so you can match silicon to workload.

Platforms

Both platforms share the same network, storage, and billing — pick the silicon that fits your stack.

Hopper and Ampere accelerators for LLM training and scale-out inference.

From $3.50 per hour

CDNA accelerators with up to 288GB of HBM3E memory per GPU.

From $2.50 per hour

Pricing

Rates shown are for the lowest vCPU and storage configuration. Every instance is billed per hour reserved.

Model	Memory	Architecture	Per hour
NVIDIA A100	80GB HBM2e	Ampere	$3.50
NVIDIA H100	80GB HBM3	Hopper	$4.75
NVIDIA H200	141GB HBM3e	Hopper	$5.50
AMD MI250X	128GB HBM3	CDNA3	$2.50
AMD MI300X	192GB HBM3	CDNA3	$3.75
AMD MI325X	256GB HBM3E	CDNA3	$4.50
AMD MI355X	288GB HBM3E	CDNA4	$6.00

Deploy your first cluster in minutes. Simple setup, no long-term commitments.