Provider record · updated 2026-09-10
TensorDock — GPU pricing & deliverability
5 tracked offerings · grade mix A:0 · B:0 · C:3 · D:2 · U:0. Marketplace of independent hosts; site explicitly states the same GPU has multiple prices by location/redundancy. The H100 page publishes hostnode composition (8x SXM5, 640GB VRAM) and '10 Gbps' network connectivity, but NO NVLink topology and NO InfiniBand/cluster fabric details; nothing published about multi-node contiguous capacity. No B200/H200/MI300X/RTX 5090 pricing found on public pages. Dedicated /pricing URL returns 404; live per-host prices are in the dashboard.
Tracked offerings
| GPU | SKU | $/GPU-hr | Terms | Grade | Conf | Fabric (as published) | Src |
|---|---|---|---|---|---|---|---|
| RTX4090 | RTX 4090 24GB | from $0.35* | on-demand | D | mark | not published | ↗vendor |
| A100-SXM | A100 SXM4 80GB | from $1.65* | on-demand | C | mark | not published | ↗vendor |
| H100-SXM | H100 SXM5 spot Evoque Dallas data center | from $1.91* | spot | C | publ | not published | ↗vendor |
| H100-PCIe | H100 PCIe 80GB | $1.99 | on-demand | D | mark | not published | ↗vendor |
| H100-SXM | H100 SXM5 on-demand Evoque Dallas data center | from $2.25* | on-demand | C | publ | not published | ↗vendor |
Grades measure what sellers publish — deliverability of the written claim, not measured performance. No fleet holds a measured (Verified) badge yet; that tier arrives with the verification harness. Source classes: API = provider's live pricing API · vendor = provider's published page · mkt API = marketplace API · 3rd-party = aggregator (provider publishes no price on its own pages) · quote = quote-gated or "starting at", not a walk-up price. Quote-class prices are shown in tables but excluded from headline ranges, spreads, and fair-price bands.
Terms & trust (as published, 2026-09-14)
| Uptime SLA | 99.99% marketed as a host standard ('TensorDock holds hosts to a 99.99% uptime standard'), not a customer contract — the downtime-compensation policy specifies no percentage target, and the ToS disclaims 'availability, timeliness, security or reliability of the Platform' — measured on: Downtime measured by a quorum of 'at least four servers located in different regions' agreeing a GPU hostnode is offline |
| SLA remedy | credits (account deductions) per published Downtime Compensation policy: Top Hosts = 10x hourly rate x downtime credited; Standard Hosts = 5x; minimum $5, capped at 100% of previous month's spending; unexpected data loss = credits at 40x hourly rate. Policy explicitly subordinate to ToS, which caps liability at greater of $100 or prepaid amounts. |
| Capacity guarantee | None published. On-demand has fixed pricing; spot instances (beta) are interruptible/outbid-able; long-term reserved capacity is 'reach out' only — no self-serve reservation guarantee documented. |
| Egress fees | 'no ingress/egress fees' (homepage claim) |
| Billing granularity | per-hour rates with continuous balance deduction ('we deduct balance continuously after a server is deployed'); exact proration granularity not published; storage billed at standard rate even when a spot instance is outbid/stopped |
| Self-serve signup | yes for on-demand (deposit funds, deploy from console); partial for reserved (requires contacting them) |
| Security certifications | not published |
Notes: Marketplace. No SOC 2 / ISO 27001 claims for TensorDock itself found — only 'Most hardware on our platform is now hosted in certified data centers' (certifications unnamed). Tier distinction published in the Downtime Compensation policy: 'Top Hosts' (demonstrated reliability, GPUs in data centers, 10x compensation) vs 'Standard Hosts' (possibly 'residential or business facilities' with 'little or no redundancy', 5x compensation); homepage markets 'Tier 3/4 Data Centers' alongside 'Converted mining rigs'. Only one of the three marketplaces with a published formula-based downtime compensation schedule.
Sources: ↗ ↗ ↗ ↗ ↗ · Fields marked "not published" are exactly that — the seller does not state them anywhere we could find. Absence is a finding.
Reliability record
TensorDock publishes no machine-readable incident history. We track every provider's public status feed weekly; a seller you cannot audit for downtime is a seller whose SLA you cannot price. Absence is a finding.
What TensorDock does not publish
- intra_node NVLink topology
- inter_node_fabric
- max_contiguous_gpus beyond one 8-GPU hostnode
- regions for non-H100 SKUs
- notice_period
- B200/H200/MI300X/RTX-5090 availability
Unpublished fields are recorded as unknown and grade conservatively. Sellers improve their grades by disclosing — that is the point.
Weekly context on every provider
The Delivered Compute Report tracks how these prices and grades move.
Subscribe free