Skip to main content
Create a new Thunder Compute instance to start running GPU workloads. You can configure the mode, GPU type, CPU count, disk size, and template.

Interactive Mode

Run the create command to launch an interactive menu:
tnr create
This walks you through selecting your instance configuration step by step.

One-Line Creation

Specify all options in a single command:
tnr create --mode prototyping --gpu a6000 --vcpus 8 --template base --disk-size-gb 100

Configuration Options

FlagDescription
--modeprototyping or production
--gpua6000, a100, or h100 (prototyping), a100 or h100 (production)
--num-gpus1, 2, 4, or 8 (production only)
--vcpusCPU cores: 4, 8, 16, or 32 (prototyping only). RAM: 8GB per vCPU
--templatebase, comfy-ui, comfy-ui-wan, ollama, webui-forge, or a snapshot name
--disk-size-gb100-400 GB (prototyping), 100-1000 GB (production)

Mode Selection

Choose between optimized development pricing or full compatibility:
  • Prototyping (default): Lower cost with CUDA-level optimizations. Best for development.
  • Production: Standard VM with full compatibility. Best for long-running jobs and production workloads.
See Prototyping vs Production for details on each mode.

GPU Options

GPUVRAMAvailability
A600048GBPrototyping only
A10080GBBoth modes
H10080GBBoth modes

CPU and RAM

Prototyping mode: Configure vCPUs with 8GB RAM per vCPU.
  • 4 vCPUs = 32GB RAM
  • 8 vCPUs = 64GB RAM
  • 16 vCPUs = 128GB RAM
  • 32 vCPUs = 256GB RAM
Production mode: Fixed at 18 vCPUs and 90GB RAM per GPU.
  • 2 GPUs = 36 vCPUs, 180GB RAM
  • 4 GPUs = 72 vCPUs, 360GB RAM
  • 8 GPUs = 144 vCPUs, 720GB RAM
By default, 4 vCPUs and 32GB of memory are included with prototyping instances. Additional vCPUs are billed hourly at the rates shown on the pricing page.

Templates

Templates pre-configure your instance for common AI workflows:
TemplateDescription
baseUbuntu with PyTorch + CUDA
ollamaOllama server environment
comfy-uiComfyUI for AI image generation
comfy-ui-wanComfyUI with Wan2.1 pre-installed
webui-forgeWebUI Forge for Stable Diffusion
After connecting to an instance with a template, start the server with:
start-ollama      # or start-comfy-ui, etc.
See Using Instance Templates for more details.

Restore from Snapshot

Your snapshots appear alongside templates in the selection menu. Select a snapshot to create an instance with your previously saved state.
tnr create
# Select your snapshot in the "Templates" step