New: Dedicated Containers

GPU Containers:
Any Docker Image, on a GPU.

Pick a GPU, pick an image, and connect over SSH or in the browser. There is no server to set up or look after. Start from a template like PyTorch, vLLM or Ollama, or bring any public image from Docker Hub, GHCR or Quay.

GPU containers
From $0.41/hr
CPU containers
From $0.08/hr
Ready templates
16
Read the Docs
→
Dedicated ContainersvLLM server
Image
vllm/vllm-openai:latest
GPU
1x L40S, 48 GB
  1. Downloading the image...
  2. Starting the container...
  3. Ready
APIhttp://203.0.113.24:31402/v1

Billed by the minute from the moment it starts.

From Image to GPU in Four Steps

Pick a GPU

On Deploy GPUs, choose a GPU and how many, then Dedicated Containers as the instance type. CPU containers are on Deploy CPU.

Choose an image

Start from a template or type any public image. Spheron checks it as you type and tells you whether SSH will work on it.

Add what it needs

Open up to 9 ports, pass up to 50 environment variables, replace the start command if you want, and add your SSH key.

Connect

Most containers are ready within a few minutes, depending on the image size. SSH in, or open each port at its own public address.

What a Container Costs Right Now

Live prices for every GPU you can deploy as a container today, per GPU per hour. The 100 GB disk and the 500 GB /workspace volume are included.

The cheapest way inA CPU container with 2 vCPUs and 4 GB of memory$0.08/hr
  1. RTX2000ADA PCIe16 GB$0.41/hr
  2. A4500 PCIe20 GB$0.42/hr
  3. L4 PCIe24 GB$0.80/hr
  4. A40 PCIe48 GB$0.85/hr
  5. RTXPRO4500 PCIe32 GB$0.97/hr
  6. RTX4090 PCIe24 GB$1.23/hr
  7. A100 80G PCIe80 GB$2.30/hr
  8. A100 80G SXM480 GB$2.33/hr
  9. MI300X OAM192 GB$3.14/hr
  10. RTXPRO6000 PCIe96 GB$3.21/hr
  11. RTXPRO6000 WS PCIe96 GB$3.34/hr
  12. H100 SXM580 GB$5.09/hr
  13. H200 SXM5141 GB$6.77/hr
Sold out right nowBack on this board as soon as there is capacity
  • B300 SXM6288 GBSold out
  • B200 SXM6180 GBSold out
  • H200 NVL143 GBSold out
  • H100 NVL94 GBSold out
  • H100 PCIe80 GBSold out
  • A6000 PCIe48 GBSold out
  • L40 PCIe48 GBSold out
  • L40S PCIe48 GBSold out
  • RTX6000ADA PCIe48 GBSold out
  • RTX5090 PCIe32 GBSold out
  • A5000 PCIe24 GBSold out
  • RTX3090 PCIe24 GBSold out
  • RTXPRO4000 PCIe24 GBSold out
  • RTX4000ADA PCIe20 GBSold out
  • A4000 PCIe16 GBSold out

Prices come straight from the Spheron catalog and refresh every few minutes. Stock moves quickly, so the deploy page has the final word.

Start From a Template

Each template is a tested image with SSH ready where the image allows it. Or choose Custom image and type any public image.

On a GPU8 templates

  • PyTorch 2.11pytorch/pytorch:2.11.0-cuda12.8-cudnn9-runtimeSSH
  • CUDA 12.8 toolkitnvidia/cuda:12.8.1-cudnn-devel-ubuntu24.04SSH
  • vLLM servervllm/vllm-openai:latestAPI :8000SSH
  • SGLang serverlmsysorg/sglang:v0.5.22-runtimeAPI :30000SSH
  • Ollamaollama/ollama:latestAPI :11434SSH
  • Open WebUIghcr.io/open-webui/open-webui:ollamaBrowser :8080SSH
  • TensorFlow 2.21tensorflow/tensorflow:2.21.0-gpuSSH
  • Jupyter Lab with PyTorchquay.io/jupyter/pytorch-notebook:cuda12-latestBrowser :8888

On a CPU8 templates

  • Ubuntu 24.04ubuntu:24.04SSH
  • Ubuntu 22.04ubuntu:22.04SSH
  • Debian 12debian:12SSH
  • Python 3.13python:3.13-slim-bookwormSSH
  • Node.js 24node:24-bookworm-slimSSH
  • Go 1.27golang:1.27-bookwormSSH
  • Rustrust:1-bookwormSSH
  • Jupyter Labquay.io/jupyter/minimal-notebook:latestBrowser :8888
Storage

Keep Your Work in /workspace

Every GPU container comes with a 500 GB volume at /workspace that survives Stop and Restart. Stop a GPU container between sessions and you pay only for its storage.

  • Files in /workspace stay. Everything else resets on Stop and Restart.

  • A stopped GPU container bills its storage only.

  • Share one volume across several containers in the same region.

500 GB
/workspace volume on every GPU container, included in the price
$0.084
per GB per month for a shared volume that outlives any container
10 to 4,096 GB
for a shared volume, and it grows in place

Container or Virtual Machine?

FeatureContainerVirtual machine
You start fromA Docker image you pickA full operating system you set up
SetupNone: the image already has your stackInstall your stack, or use a startup script
AccessSSH on images that run as root, plus public portsSSH
Storage100 GB disk and a 500 GB /workspace volume per GPU containerThe machine's disk, plus volumes where the provider supports them
BillingHourly rate, billed by the minute, 20-minute minimumHourly rate, billed by the minute, 20-minute minimum
Best forNotebooks, model serving, short jobs and experimentsLong training runs and work that needs its own drivers or kernel

Compare Instance Types

Instance type

Dedicated

On-demand instances with a 99.99% SLA. Non-interruptible. Deploy instantly, pay by the minute.
Best for

Production inference, long training runs, and any job that can't afford a restart

Instance type

Spot

On-demand instances at a discount using idle capacity. Same hardware, up to 50% off, interruptible.
Best for

Batch jobs, evals, sweeps, and experiments that checkpoint and resume after a stop

Instance type

Reserved

Locked capacity with volume pricing. Custom clusters and dedicated support.
Best for

Multi-month training, large clusters, and teams that need capacity locked in early

Current

Containers

Any Docker image on a GPU or CPU, with SSH and public ports. Nothing to set up, billed by the minute.
Best for

Notebooks, model serving, short jobs, and anything that already ships as an image

FAQ / 06

GPU Container FAQ

What runs, how you connect, and how containers are billed.

A GPU container runs one Docker image on a GPU, with SSH and public ports, and nothing else to set up. You pick the GPU and the image, and Spheron starts it, installs your SSH key and gives each open port a public address. CPU containers work the same way, without a GPU.

Any public image from Docker Hub, GHCR, Quay and other registries, or one of 16 templates such as PyTorch, vLLM, Ollama and Jupyter Lab. Spheron checks the image as you type it, so a name that does not exist is caught before anything is created.

Over SSH on any image that runs as root: Spheron sets up SSH with your key when the container starts, even if the image has no SSH server of its own. Images that run as another user, like the Jupyter ones, are reached in the browser through their open ports instead.

Files in /workspace are kept. Everything outside /workspace is reset on Stop and Restart, and the whole container is deleted when you terminate it. For files that should outlive a container, attach a shared volume when you deploy.

At an hourly rate, billed by the minute from the moment the container starts, the image download included, with a 20-minute minimum. A stopped GPU container bills only its storage until you start it again.

View GPU pricing

Yes. List container offers with computeType=gpu-container or cpu-container on the offers endpoint, then deploy with a container block (image, command, ports and environment variables) on the same call you use for virtual machines.

Read the API reference
// GPU Containers

Your Image on a GPU in Minutes

Pick a template or bring any public image. SSH in, open your ports, and stop it when you're done.

View Pricing
→
Or talk to our team
Spheron GPU instance going live and linking data centers across the globe