TensorDock pricing: GPU rental rates

Updated July 2026 · day 14 of the ledger · verified 2026-07-18

The cheapest GPU on TensorDock right now is the RTX A4000 at $0.10/hr (on-demand, verified 2026-07-18). TensorDock is a GPU marketplace that lists machines from independent hosts in Tier 3/4 data centers. Its lineup page publishes a typical hourly rate per GPU model, and rates vary by host. The GPU rate excludes vCPU, RAM and storage, which are billed a la carte per second, so the all-in instance cost is higher than the GPU-only figure. We track the published per-GPU rates daily; the live table above shows current values.

Live TensorDock prices (2026-07-18)

Cheapest listed per-GPU-hour rates verified at the source. Before storage, egress and networking.

Pricing model

TensorDock prices GPUs by the hour with per-second billing on resources. The published GPU rate is the GPU component only: vCPU, RAM and NVMe storage are added a la carte. Because it is a marketplace, hosts list at staggered prices and the site notes that availability at the lowest listed rate is not guaranteed. Billing is prepaid: you deposit funds and provision servers against your balance.

Capacity tiers

Three ways to buy: on-demand hourly, spot instances where you bid at or above a minimum bid and pricing fluctuates with market conditions, and reserved terms (monthly, annual, multi-year) at lower effective hourly rates. The H100 model page documents both the on-demand rate and the minimum spot bid; we publish both tiers where the site does. Reserved-term rates are commitments, not hourly rentals, so we do not publish them as live prices.

GPU lineup

The current lineup table lists H100 SXM5, A100 SXM4 and PCIe 80GB, V100 SXM2, L40, RTX 4090, RTX 3090, RTX 6000 Ada, RTX A6000 and RTX A4000. TensorDock says its on-demand H100 fleet is offered in partnership with Voltage Park, with 1-8 GPU virtual machines deployable on demand. KVM virtualization with root access and dedicated GPU passthrough is documented for on-demand customers.

Storage and hidden costs

The main cost gotcha is the a-la-carte model: the GPU-only rate looks cheaper than instance-priced clouds until vCPU, RAM and storage are added. On the plus side, stopped on-demand servers with unreserved GPUs are billed at a lower storage-only rate, and you can request an export of your VM disk image. A public bandwidth or egress fee schedule was not found on the pages reviewed. Not publicly documented.

Who should pick TensorDock

Pick TensorDock if you want marketplace rates on datacenter and workstation GPUs, are comfortable sizing vCPU and RAM yourself, and can tolerate host-to-host variability. Spot bidding suits interruption-tolerant workloads chasing the lowest rates. For strict SLA or compliance needs, verify the specific host and terms first: a formal public uptime SLA for marketplace rentals was not found in the pages reviewed. Not publicly documented.

FAQ

Profile verified 2026-07-18. Sources: www.tensordock.com · www.tensordock.com · www.tensordock.com