Bare metal
The physical machine, single-tenant, no hypervisor. Direct PCIe to the GPUs, your choice of OS, and on-call engineers who can walk the aisle. Used when a training run or inference service cannot share a host.
Pick a SKU, a mode and a run length
Illustrative USD before tax. Spot is typical, not a guarantee. Reserved discounts are the published 18 / 24 / 28 / 31% schedule.
How this actually runs
A rental on Sundancæ is a physical GPU at one of our sites, billed by the second on a rate card you can read without a sales engineer. On-demand is the general pool: you start a node, you stop it, you pay for the seconds it ran with a sixty-second minimum. There is no reservation fee and no penalty for leaving early. If the SKU is free, median time to a schedulable card is under two minutes.
Reserved takes those same cards out of the pool for a term. The discount is published - 18% at one month, 24% at three, 28% at six, 31% at twelve - not a private SKU. After the first month you can cancel with thirty days’ written notice. Monitoring is included. Six- and twelve-month reserved and all bare metal carry a 99.9% uptime SLA and 24/7 on-call.
Bare metal is the whole machine, single-tenant, no hypervisor. You install the OS. Direct PCIe to the GPUs, with InfiniBand or 100 GbE depending on the SKU. This is what teams use when a training run or a production inference service cannot share a host. A caged rack exists for the few workloads that need physical isolation beyond a tenant boundary; it is quoted, not a checkbox.
Spot is leftover capacity after reserved and on-demand have taken what they need. Typical saving is around 70% off on-demand, with a five-minute interruption notice. If the job checkpoints - PyTorch, JAX, farm tiles with retry, overnight ETL - spot is usually the honest price. If it is a stateful inference frontend, do not use it. Reclaim is real; treat the notice as a kill.
One account covers both sites. Data stays in the site you provision it in. In-region transfer is included. Cross-site copies are explicit. We do not resell hyperscaler capacity and we do not quietly place you on someone else’s cloud when our pool is full - we tell you we cannot schedule you.
Compute used inside an engineering or advertising retainer still invoices at these rates. People are a flat monthly line; GPUs are à la carte. If the job can run on your cloud instead, that line is zero. Volume does not unlock a secret price. The card on Pricing is the card we bill.
Shared-nothing by default
The guest is the host. CUDA, drivers and fabric behave like a box under your desk, except the power contract is ours.
Install Ubuntu, a custom AMI-style image, or a locked-down inference stack. We do not restrict the software as long as it stays within acceptable use.
For workloads that need physical isolation beyond a single tenant, we can cage a rack. Quoted separately.
Illustrative week · GPU-hours consumed
What changes when you change the commercial model
| On-demand | Reserved | Spot | Bare metal | |
|---|---|---|---|---|
| Term | None | 1–12 months | None | 1–12 months |
| Price | Public $/hr | 18–31% off | ~70% off typical | Reserved + isolation |
| Kill / reclaim | You stop it | You stop it | 5-minute notice | You stop it |
| SLA | Best effort | 99.9% on 6–12 mo | None | 99.9% + 24/7 |
| Isolation | Hypervisor | Hypervisor | Hypervisor | Single tenant |
| Good for | Burst, tests | Steady baseline | Checkpoint batch | Train / prod inference |
Published SKUs, both sites
Full table including reserved discount math lives on Pricing. Spot is typical. Fabric is InfiniBand at InfiniBand fabric, or 100 GbE (25 GbE on smaller cards) depending on the SKU.
| Instance | Site | VRAM | CPU / RAM | Fabric | On-demand | Spot typ. |
|---|---|---|---|---|---|---|
| H100 PCIe 80GB | InfiniBand | 80GB HBM3 | 24 / 200 GB | InfiniBand | $2.19 | $0.66 |
| A100 PCIe 80GB | InfiniBand | 80GB HBM2e | 16 / 180 GB | InfiniBand | $1.24 | $0.37 |
| L40S 48GB | 100 GbE | 48GB GDDR6 | 12 / 128 GB | 100 GbE | $0.86 | $0.26 |
| RTX 4090 24GB | 25 GbE | 24GB GDDR6X | 8 / 64 GB | 25 GbE | $0.41 | $0.12 |
| RTX A6000 48GB | Multi-site | 48GB GDDR6 | 12 / 96 GB | 25–100 GbE | $0.95 | $0.29 |
| CPU render node | Multi-site | 64 / 256 GB | 25 GbE | $0.68 | $0.20 |