GPUy.dev
SaladCloud logo

SaladCloud Architecture & Platform Gotchas

Independent AuditDistributed Consumer GPU Grid

Technical teardown of host infrastructure, networking, and engineering risks

Live RatesCompare RivalsOfficial Portal
Watch Out For (Known Quirks & Pitfalls)
  • •Nodes run on distributed consumer gaming PCs: nodes have dynamic residential IPs, not fixed datacenter IPs.
  • •Not suitable for stateful SSH training: nodes can disconnect if the consumer gamer reclaims their PC.
  • •Strictly containerized via SaladCloud API; you must build your application as a self-recovering Docker image.
Pro Tips & Optimization Hacks
  • ✓Unbeatable cost for batch Whisper transcription, FLUX/SDXL image rendering, and async model inference ($0.18/hr for RTX 4090).
  • ✓Implement queue workers (e.g. SQS, Redis BullMQ) where tasks take <60s each, so node drops have zero user impact.
  • ✓Pre-bake model weights into container layers or pull from cloud storage with aggressive multi-part streaming.
Hidden Cost Radar
Stopped Storage:$0.00 (Stateless container caching only)
Egress Bandwidth:Free / Included in low hourly rate
Billing Increment:Per Second
Min Deposit:$10.00 initial credit
Interconnect & Fabric
NVSwitch Bandwidth:N/A (Consumer single-card desktop topology)
Cluster Fabric:None (Residential broadband)
P2P Reliability:LIMITED
Datacenter:Decentralized Edge Residential Nodes (100k+ global consumer PCs)
DevOps & Automation
Direct SSH (Port 22):✕ Web Terminal
CLI & Python SDK:✓ Official SDK
SkyPilot & Terraform:Manual Orchestration
Spot Notice Buffer:Instant drop if host gamer reclaims PC (queue retry architecture required)
Recommended Workloads:Asynchronous batch AI generation (Whisper, FLUX, Stable Diffusion) • Ultra-low-cost video processing and upscaling pipelines • Stateless API workers with automatic task retries
Avoid Using For:Long-running multi-hour stateful training jobs • Interactive Jupyter notebooks or live debugging
Host Infrastructure Model

SaladCloud operates as a decentralized compute marketplace. Physical hardware is hosted across independent commercial colocation facilities and vetted node operators worldwide, enabling unbeatable spot rates.

Tier Rating: Distributed Consumer GPU Grid
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 (Included in Container Rate) until deleted.
  • !Spot Preemption: Spot instances can be interrupted if a higher bidder arrives. Write periodic model checkpoints.
  • !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
450 MB/s (~4 Gbps)
Model cache loading performance
Egress Bandwidth Policy
Free / Included
No hidden outbound gigabyte charges
Public IP & Port Forwarding
NAT Port Forwarding
SSH & Web UI connectivity model

Security, Privacy & Data Isolation Audit

Container Isolation & Root Permissions

Instances on SaladCloud 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

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