GPUy.dev
TensorDock logo

TensorDock Architecture & Platform Gotchas

Independent AuditTier 3 Datacenter

Technical teardown of host infrastructure, networking, and engineering risks

Live RatesCompare RivalsOfficial Portal
Watch Out For (Known Quirks & Pitfalls)
  • •Host reliability differs between Core Tier (datacenters) and Community hosts; select Core locations for long unattended training jobs.
  • •Direct port mapping must be requested or opened during pod creation to expose custom WebUI ports without reverse tunnels.
  • •Stopped instances continue incurring disk storage fees until the associated NVMe volume is explicitly deleted.
Pro Tips & Optimization Hacks
  • ✓Take advantage of TensorDock's transparent REST API and CLI to automate provisioning ephemeral RTX 4090 and A100 test rigs.
  • ✓Direct SSH on native port 22 with assigned public IPv4 makes setting up VS Code Remote and PyCharm frictionless.
  • ✓Consistently provides some of the lowest on-demand rates worldwide for single RTX 3090 ($0.16/hr) and RTX 4090 ($0.39/hr) with zero waitlists.
Hidden Cost Radar
Stopped Storage:$0.10/GB/month for attached NVMe storage
Egress Bandwidth:Free / Unmetered inbound and outbound datacenter bandwidth
Billing Increment:Per Second
Min Deposit:$5.00 via Credit Card or Crypto
Interconnect & Fabric
NVSwitch Bandwidth:900 GB/s on A100 SXM4 / PCIe 4.0 on GeForce desktop hosts
Cluster Fabric:RoCE v2 on enterprise Core datacenter nodes
P2P Reliability:GOOD
Datacenter:Tier 3 Datacenter (Core) & Host Marketplace (Community)
DevOps & Automation
Direct SSH (Port 22):✓ Native TCP
CLI & Python SDK:✓ Official SDK
SkyPilot & Terraform:✓ SkyPilot Ready
Spot Notice Buffer:30s termination notice buffer on interruptible marketplace tiers
Recommended Workloads:Budget single-card and dual-card fine-tuning (LoRA / QLoRA) • Developer prototyping and batch media rendering pipelines • Cost-optimized inference endpoints on dedicated hardware
Avoid Using For:Distributed multi-node training clusters (>16 GPUs) requiring high-radix InfiniBand fat-trees
Host Infrastructure Model

TensorDock maintains standardized enterprise Tier 3 or Tier 4 colocation facilities with redundant N+1 UPS power, dual fiber network paths, and strict environmental controls.

Tier Rating: Tier 3 Datacenter
Key Platform Strengths
  • ✓Extremely aggressive pricing with substantial spot discounts compared to legacy hyperscalers.
  • ✓Full direct root SSH access without restrictive security groups or mandatory IAM policy setups.
  • ✓Cryptocurrency billing supported with zero mandatory KYC verification.
Gotchas & Traps to Watch For
  • !Stopped Storage Billing: Persistent volume disk space is billed 24/7 ($0.08/GB/month) until deleted.
  • !Bandwidth Variance: Individual host download speed can vary. Test HuggingFace download throughput before starting huge batch training.

Network Throughput & Disk Volume Architecture

Independent telemetry metrics verified across real machine instances.

HuggingFace Download Speed
1.1 GB/s (~9 Gbps)
Model cache loading performance
Egress Bandwidth Policy
1 TB Free / Month
No hidden outbound gigabyte charges
Public IP & Port Forwarding
Public IP Supported
SSH & Web UI connectivity model

Security, Privacy & Data Isolation Audit

Container Isolation & Root Permissions

Instances on TensorDock provide direct root container access. All running processes exist within a dedicated Docker namespace with NVIDIA container runtime GPU passthrough. Remember to use private SSH keypairs rather than basic passwords to protect public endpoint ports.

KYC Policies & Billing Privacy

TensorDock allows developers to deploy instances without mandatory government ID verification. Cryptocurrency deposits (USDT/USDC) can be used to fund accounts anonymously.