GPU Workloads

Varity TeamCore Contributors

Varity Cloud offers two GPU execution classes. Their inventory and prices are live supply, not a documentation rate card.

Execution class	You receive	Create with	Access
virtual_machine	A GPU virtual machine	POST /api/machines	SSH when ready
bare_metal	A physical GPU host	POST /api/machines	SSH when ready

No documentation price is current

Read the authenticated catalog, choose the complete workload configuration, and request a fresh quote. The quote’s price, resource fingerprint, and valid_until deadline bind the create request. If a quote cannot be issued, the price is unavailable; do not fall back to a cached rate.

GPU has no monthly maximum

GPU billing follows the accepted hourly quote. A continuous-use total is an estimate, not an enforced ceiling. Delete the workload to release the reservation and stop future supplier usage.

GPU VM or Bare-Metal Machine

Read the matching catalog first:

curl "https://varity.app/api/deployment-profiles?workload=gpu&execution_class=virtual_machine" \
  -H "Authorization: Bearer $VARITY_API_KEY"

Use execution_class=bare_metal for a physical host. Region, accelerator, host resources, and available OS images belong to the selected profile; they are not independent assumptions.

1. Quote the exact profile and image

   curl -X POST https://varity.app/api/pricing/machine-quote \
     -H "Authorization: Bearer $VARITY_API_KEY" \
     -H "Content-Type: application/json" \
     -d '{
       "profile_id": "<profile id from the current catalog>",
       "execution_class": "virtual_machine",
       "os_image": "<image id from that profile>"
     }'

2. Create before valid_until

   curl -X POST https://varity.app/api/machines \
     -H "Authorization: Bearer $VARITY_API_KEY" \
     -H "Idempotency-Key: gpu-machine-001" \
     -H "Content-Type: application/json" \
     -d '{
       "name": "trainer-01",
       "profile_id": "<profile id from the current catalog>",
       "execution_class": "virtual_machine",
       "os_image": "<image id from that profile>",
       "ssh_public_key": "ssh-ed25519 AAAA...",
       "accelerator_quote_token": "<quote token>"
     }'

   Use execution_class: "bare_metal" only with a bare-metal profile and a quote for that same configuration.

3. Poll for access

   curl https://varity.app/api/machines/$MACHINE_ID \
     -H "Authorization: Bearer $VARITY_API_KEY"

   access is null until ready, then carries the SSH host, port, and user.

Billing and Lifecycle

• The binding price is the quote accepted by the create request, not a catalog snapshot or a value copied from this page.
• GPU machines carry the documented minimum authorization shown by the current quote flow.
• Additional storage is not accepted for GPU machines.
• Restart and soft reboot preserve the reservation and billing.

curl -X DELETE https://varity.app/api/machines/$MACHINE_ID \
  -H "Authorization: Bearer $VARITY_API_KEY"

Machine deletion is asynchronous. Acceptance is not closure proof: read the machine until lifecycle.cleanup_state is complete. Deletion destroys the machine disk, so copy required data first.

Related

Machine CatalogsChoose a live catalog and quote path.

CPU Virtual MachinesCPU VM sizing, quote, access, billing, and lifecycle.

API ReferenceAuthentication and the public resource interface.