Instance Types

Spheron instances are organized along three axes: compute type, interruptibility, and hardware isolation.

Compute type determines whether the instance carries an accelerator. GPU instances are deployed from Deploy GPUs and are the subject of most of this page. CPU nodes carry no GPU at all and are deployed from their own Deploy CPU page, from $0.09 per hour. See CPU Nodes for sizes, regions, and the CPU deploy flow.

Interruptibility determines whether a running instance can be reclaimed by the provider. Spot instances can be interrupted at any time. Dedicated instances carry a 99.95% SLA and are not reclaimed after deployment. Both apply to CPU nodes as well as GPU instances.

Hardware isolation determines how your workload accesses the underlying hardware. Spot always runs in a VM. Dedicated instances come in two sub-types:

• VM: Isolated virtual machine on shared physical hardware. This is the default for most GPU offers across all providers.
• Bare Metal: Full access to a physical server with no hypervisor layer. GPU count varies by offer and provider, from single-GPU up to multi-GPU servers. Identified by the BAREMETAL suffix in the GPU type name on the dashboard.

Decision Matrix

Spot

Spot instances are VM-based, typically 30 to 60% cheaper than Dedicated. The provider can reclaim them at any time.

• Experiments and prototyping
• Batch training jobs with checkpointing enabled
• Fault-tolerant workloads with checkpointing to a persistent volume

Save checkpoints to a persistent volume at regular intervals to resume training if the instance is reclaimed:

import torch
 
# Save checkpoint every N steps to a mounted volume
if step % checkpoint_interval == 0:
    torch.save({
        'step': step,
        'model_state_dict': model.state_dict(),
        'optimizer_state_dict': optimizer.state_dict(),
    }, '/checkpoints/checkpoint_latest.pt')

See the Volume Mounting guides to set up a persistent volume at /checkpoints.

Dedicated

Dedicated instances carry a 99.95% SLA and are not reclaimed by the provider after deployment. Two hardware sub-types are available under Dedicated.

VM

Dedicated VM is the standard virtualized offering. Your workload runs in an isolated VM on shared physical hardware. This is the most common instance type across providers and covers the majority of GPU offers on the platform.

• Production inference servers
• Interactive training sessions
• Demos and customer-facing workloads
• Any job requiring uninterrupted runtime without needing bare-metal performance

Bare Metal

Dedicated Bare Metal gives you full access to a physical server with no hypervisor or VM layer. You get the entire machine, which eliminates virtualization overhead and provides predictable hardware performance. GPU count varies by offer and provider; Bare Metal servers range from single-GPU configurations up to multi-GPU servers.

Bare Metal GPU offers display with the BAREMETAL suffix in the GPU type name on the dashboard. They appear alongside standard VM offers in the GPU offers list; there is no separate section or tab for them.

• Workloads sensitive to virtualization overhead
• Use cases that require direct hardware access for performance or compliance reasons

On the Dashboard

1. Open the Spheron dashboard and go to Deploy GPUs.
2. Browse the GPU offers list. Look for the BAREMETAL suffix in the GPU Type field of each offer card.
3. Select the matching offer, complete the deployment form (OS image, SSH key, optional startup script), and deploy.

Multi-GPU Bare Metal

Multi-GPU Bare Metal offers give a training job every GPU on a single host with no hypervisor layer. On SXM offers, the GPUs are connected by NVLink or NVSwitch, which provides the GPU-to-GPU bandwidth that all-reduce-heavy workloads depend on:

• Large-scale distributed training (PyTorch DDP, DeepSpeed ZeRO-3)
• Multi-GPU jobs requiring maximum GPU-to-GPU bandwidth (choose an SXM offer)

See the Distributed Training guide for setup details.

CPU Node

CPU Node is a CPU-only instance with no GPU attached. It is not a variant of a GPU offer: it has its own page in the dashboard sidebar under Compute, its own wizard, and no GPU count, cluster networking, or deployment type to choose.

Sizes start at 4 vCPU with 4 GB of memory. Verda sells CPU nodes at spot rates as well as on demand; Sesterce, Spheron AI, and Massed Compute offer them on demand only.

• Data preprocessing, ETL, and dataset sharding
• Build steps, CI jobs, and container image builds
• Schedulers, queue workers, and cron jobs
• API servers and control planes that front a GPU fleet

See CPU Nodes for the full size and region matrix, pricing, and the CPU deployment API.

Silicon vendor

The catalog is no longer NVIDIA only. Offers and instance cards carry a vendor mark so you can tell at a glance what you are renting:

An AMD or Intel instance needs a different software stack from an NVIDIA one. nvidia-smi and CUDA-only wheels do not work on them. Read ROCm and AMD Instinct before deploying an MI300X.

Stopping and restarting by instance type

What an instance can do after it is running depends on its type as well as its provider:

Spot machine types cannot be stopped, because the hardware is not held for you in the first place. See Instance lifecycle for the per-provider matrix and the billing rules.

Selecting Instance Type via Dashboard

When deploying from the Spheron dashboard:

1. Go to Deploy GPUs and browse the GPU offers list.
2. Each offer card shows the GPU model, vCPUs, RAM, storage, and price.
3. Identify the offering type by the offer details:
   • Spot: labeled as interruptible; typically shows a spot price alongside the regular price.
   • Dedicated VM: standard VM price with no interruption risk; the most common offering across providers.
   • Dedicated Bare Metal: identified by BAREMETAL in the GPU type name on the dashboard (e.g., H100_SXM5_BAREMETAL). For distributed training, check the GPU count on the offer card and prefer an SXM form factor.
4. Select the offer that matches your workload requirements and click Deploy.

What's next

• CPU Nodes: CPU-only instances for work that never touches a GPU
• Distributed Training guide: Multi-GPU bare-metal setup
• Regions and Providers: GPU tiers and capabilities per provider
• Volume Mounting: Persistent storage for checkpoints
• Instance Lifecycle: Stop, start, and restart support by type and provider
• ROCm and AMD Instinct: Running workloads on AMD MI300X GPUs
• Cost Optimization: Choosing the right instance type for your budget
• API Reference: Filter GPU offers by instance type programmatically