Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.sfcompute.com/llms.txt

Use this file to discover all available pages before exploring further.

San Francisco Compute is a market for buying time on GPU clusters. You can buy any number of individual VM nodes for any duration at any start time provided there is capacity available.

​
Install the CLI

curl -fsSL https://sfcompute.com/cli/install | bash
Source your shell profile to add the sf command to your PATH: 
source ~/.bashrc # For Bash
source ~/.zshrc  # For Zsh
Log in to your account: 
sf login

​
Buy VM nodes

After installing the CLI, you can buy reserved nodes varying the number of nodes, duration, and start time as you like. 
# buy 8 H100 VM nodes (64 GPUs) for 24 hours starting tomorrow
sf nodes create -n 8 -d "24h" -s "tomorrow at 9am"
To access your nodes by SSH, you must add your keys in a startup script or configure your nodes using a cloud-init user-data file.
You can buy and get access to nodes on-demand right now by not specifying a start time: 
# buy 2 H100 VM nodes (16 GPUs) for 1 hour at $20/node/hr with your SSH key configuration
sf nodes create cuda-crunch -n 2 --zone golden-gate --duration 1h --max-price 20.00 -U ./startup.sh

sf nodes ssh root@cuda-crunch
You can also buy nodes in the dashboard buy page. For more advanced usage such as custom VM images or auto reserved nodes, see Nodes.

​
Using the API

Generate an API token from the command line: 
sf tokens create
Pass the token as a Bearer token in the Authorization header: 
curl --request GET \
  --url https://api.sfcompute.com/v0/orders \
  --header 'Authorization: Bearer <token>'

​
CLI options

​
Telemetry

The SF Compute CLI collects some usage data. This is enabled by default. To opt out, set the SF_CLI_TELEMETRY_OPTOUT environment variable to 1 or true.

​
Auto-upgrades

The CLI automatically upgrades if there’s a new patch version. To opt out, set the SF_CLI_DISABLE_AUTO_UPGRADE environment variable to 1 or true.