CPU nodes

CPU Node is a CPU-only instance with no GPU attached. Use it for build steps, data preparation, schedulers, API workers, control planes, and any other job that never touches a GPU. Prices start at $0.09 per hour, well below the cost of renting a GPU you do not use.

CPU Node has its own page in the dashboard sidebar under Compute, next to Deploy GPUs. The page opens whether or not you are signed in.

When to use a CPU node

A CPU node runs no NVIDIA driver and reports no GPU. Commands such as nvidia-smi and torch.cuda.is_available() do not work. Deploy a GPU instance for anything that needs CUDA.

Available sizes

Every provider's CPU offer collapses onto a single CPU NODE entry, so there is no catalog to page through. The wizard lists sizes from 4 vCPU with 4 GB of memory up to 360 vCPU with 1440 GB, and shows the lowest available price for each size before you commit.

Availability varies by size and region. The wizard only offers sizes that the selected region can actually serve. Check the Deploy CPU page for live inventory and pricing.

Pricing

Spot CPU nodes on Verda start at $0.09/hr for 4 vCPU with 16 GB of memory. Dedicated rates start around $0.20/hr for the same size. Billing is per second, identical to GPU instances, and the order summary projects a full month from the hourly rate.

The 20-minute minimum runtime before manual termination applies to CPU nodes exactly as it does to GPU instances. See Instance Types for the full rule.

Spot on CPU

Verda CPU instances sell at spot rates as well as on demand. The spot and on-demand toggle behaves the same way it does for GPU offers: the provider can reclaim a spot node at any time, so checkpoint any long-running job to a persistent volume.

Deploy from the dashboard

Operating systems

CPU nodes list only images that make sense without a GPU. CUDA and other GPU driver images are excluded.

Connect to a CPU node

Connect over SSH exactly as you would to a GPU instance:

# Copy the SSH command from the instance details panel
ssh -i ~/.ssh/id_ed25519 root@<your-instance-ip>

The username varies by provider. See the SSH connection guide for details.

Verify the machine after connecting:

# vCPU count and model
lscpu | grep -E "^CPU\(s\)|Model name"
 
# Memory
free -h
 
# Disk
df -h

How CPU nodes appear in the dashboard

• Instance cards name the vCPU and memory where a GPU instance names its model, and carry a hardware icon so a CPU node and an accelerator are distinguishable at a glance.
• GPU count and video memory are omitted entirely rather than shown blank.
• The GPU catalog on Deploy GPUs lists GPUs only. CPU Node never appears there, so the result count and page count on that page cover GPU offers alone.

Deploy through the API

List CPU offers

Narrow the catalog to CPU with computeType=cpu:

curl -H "Authorization: Bearer $SPHERON_API_KEY" \
  "https://app.spheron.ai/api/gpu-offers?computeType=cpu"

The response contains a single row with gpuType: "CPU" and displayName: "CPU NODE". Every entry in its offers array reports gpuCount: 0 and carries the offerId a deployment takes, along with vcpus, memory, storage, price, os_options, and the region in clusters.

Create a CPU deployment

Send gpuType: "CPU" with gpuCount: 0:

curl -X POST "https://app.spheron.ai/api/deployments" \
  -H "Authorization: Bearer $SPHERON_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "provider": "data-crunch",
    "offerId": "CPU.4V.16G_FIN-01_200_spot",
    "gpuType": "CPU",
    "gpuCount": 0,
    "region": "Finland 1",
    "operatingSystem": "Ubuntu 22.04",
    "instanceType": "SPOT",
    "sshKeyId": "your_ssh_key_id"
  }'

Deployments created this way are stored as CPU nodes, so responses and usage records report a CPU node rather than an unnamed GPU.

See the API reference for the full endpoint contract.

Troubleshooting

Issue: No regions available for CPU NODE

Symptoms: The region list is empty and the wizard reports that the configuration is not available in any region.

Diagnosis: The selected instance size is not stocked in any region for that provider and instance type.

Resolution: Choose a different instance size, or switch between Spot and Dedicated. The wizard filters regions by the size you picked, not the other way around.

Issue: nvidia-smi is not found

Symptoms: nvidia-smi: command not found after connecting.

Diagnosis: This is expected. A CPU node has no GPU and ships no NVIDIA driver.

Resolution: Deploy a GPU instance from Deploy GPUs if your workload needs CUDA.

Issue: Deployment rejected with an invalid GPU type error

Symptoms: POST /api/deployments returns 400 naming a GPU type mismatch against a CPU offer.

Diagnosis: The request sent a GPU model name in gpuType for an offer with gpuCount: 0.

Resolution: Send "gpuType": "CPU" and "gpuCount": 0, or omit gpuType and let the platform resolve it.

What's next

• Instance types: Spot and Dedicated trade-offs across CPU and GPU
• Regions and providers: Provider capabilities and geographic coverage
• Cost optimization: Move non-GPU work off GPU instances
• Startup scripts: Automate setup on first boot
• API reference: Filter offers by compute type programmatically