Dedicated GPU instances with one-click JupyterLab, ComfyUI, vLLM serving, and web terminal templates. Billed per second from allocation, including startup, until stopped, at rates from $0.25/hr.
請求はGPUが割り当てられた瞬間から、起動時を含めて停止または破壊されるまで、指定された時間単位で行われます。GPUをクリックすると、完全な仕様、在庫状況、ワンクリック展開オプションが表示されます。
Persistent volumes are $0.18 per GB-month, billed per second for as long as the volume exists, including while no GPU is running. Attach one at deploy to keep files in /workspace between sessions. A single volume holds up to 4 TB, and an account starts with 10 volumes and 10 TB of total capacity. Workspace volumes keep files between runs. Object buckets provide shared datasets mounted at /mnt/<name>. The size selected is the minimum billed; usage above it is billed in whole GB. Network volumes share files at /workspace across compatible GPU instances and cluster nodes.
A cluster is several whole machines wired to one high-speed interconnect and reserved together, for distributed training. Renting separate instances gives unrelated machines with no fast path between them. Choose how many nodes to take; the price shown is per node.
No clusters are available right now. Sizes and prices appear here as soon as one frees up.
Deploy a clusterBilling is per second at the listed hourly rate from the moment your GPU is allocated, including startup, until you stop or destroy the instance. The rate is locked in when you deploy, and stopping or destroying the instance stops the charge.
Spot instances usually cost less than on-demand instances and can be reclaimed at any time without notice. A reclaimed spot instance stops and loses the files on its disk, and it can start again only when spot capacity is available, which is not guaranteed; you can also deploy on demand. Spot rates are listed beside on-demand rates wherever spot capacity is available, and are billed per second like any other instance. To keep files, attach an object bucket or a network volume; workspace volumes attach to on-demand instances only. To deploy one, choose Spot when you deploy in the dashboard, or set spot to true in the API.
One-click templates cover JupyterLab notebooks, ComfyUI, vLLM model serving (bring a Hugging Face model id), and a browser web terminal. You connect through the authenticated EmpirioLabs connect endpoint or call the workload through /v1/gpu/connect/{instance_id}/{path} on the API.
Yes. Everything the dashboard does is also available through the API: deploy, stop, and destroy instances under /v1/gpu on api.empiriolabs.ai, and reach the running workload through the connect endpoint. The full reference is in the GPU Cloud docs.
Runtime storage is included in the listed GPU price. Select a GPU configuration to see its included storage. Files on the runtime disk are temporary.
Attach a persistent volume at deploy. Files you save in /workspace are kept when you stop or destroy the GPU and restored the next time you attach the same volume. Volumes are $0.18 per GB-month, billed per second while the volume exists, including while no GPU is running. Deleting the volume deletes its files and stops the charge. A single volume holds up to 4 TB, and an account starts with 10 volumes and 10 TB of total capacity. Workspace volumes keep files between runs. Object buckets provide shared datasets mounted at /mnt/<name>. The size selected is the minimum billed; usage above it is billed in whole GB. Network volumes share files at /workspace across compatible GPU instances and cluster nodes.
Create an EmpirioLabs account, open GPU Cloud in the dashboard, pick a GPU and template, and deploy. Billing is pay-as-you-go credits.
Check out our pricing or reach out if you want your own model deployed on our stack.