TensorDock user reports 2-day outage, failed GPU deployments, zero support

A developer using TensorDock for corporate GPU benchmarking has shared a detailed complaint highlighting critical reliability and support failures. The user initially enjoyed a six-hour flawless session on an RTX 4090 cloud PC, but after stopping the VM to save costs, was unable to restart it. Over the next 10 hours, four separate deployment attempts in three different node locations all failed to initialize the GPU. The service consistently showed RTX 4090 availability, yet every deployment ended with a non-functional desktop VM that required deletion.

The situation worsened with an RTX 5090 deployment. The user set up this higher-tier machine for $10/day, intending it as an always-on workstation. However, after initial operation, the VM became unpingable and inaccessible. Attempts to contact TensorDock customer support over two days went unanswered, leaving the user with a paid but unusable cloud PC and a custom Windows image they cannot recover. This incident underscores the risks of relying on cloud GPU services for mission-critical development work, especially when availability indicators don't match actual provisioning capability and support is nonexistent.