n8n-nodes-vastai

Serverless GPU LLM inference TCO node/CLI — compare what a dedicated AWS EC2 p5.48xlarge actually burns per month against RunPod and Vast.ai market pricing for H100 / A100 / RTX 4090 workloads, and emit an auto-importable n8n vLLM / OpenAI-compatible inference node config — pure Node.js built-ins, zero third-party dependencies.

The Real Cost Example: AWS Premium vs Marketplace GPU

A single always-on GPU-month is 720 GPU-hours. The 2026 benchmark medians:

| GPU | AWS EC2 ($/gpu-hr) | RunPod | Vast.ai | Loss if on AWS (720 h) | |---|---|---|---|---| | H100 80GB | $1.536 (p5.48xlarge ÷ 8) | $2.99 | $2.09 | — (AWS = $1,105/mo) | | A100 80GB | $8.00 | $1.89 | $1.29 | $5,031/mo if RunPod; $5,356/mo if Vast.ai | | RTX 4090 24GB | n/a | $0.69 | $0.49 | $2,009/mo loss vs Vast.ai if "AWS-ized" at $3.28 |

Concrete headline: an A100 workload pinned to AWS-style rates costs $5,760/month; the identical workload on Vast.ai is $403.20/month — quantified monthly loss of $5,356.80, i.e. 93% overpayment, $64,281.60/year. RunPod's serverless scale-to-zero changes the picture further for bursty inference: you pay only for cold+active seconds, so real-world invoices for <20% utilization workloads drop another 40–70% below the always-on figures above.

Installation

npm install -g n8n-nodes-vastai
npx n8n-nodes-vastai --gpu h100 --hours 720
npx n8n-nodes-vastai --gpu a100 --hours 180 --json

CLI Options

| Flag | Default | Description | |---|---|---| | --gpu <model> | h100 | h100, a100, or rtx4090 | | --hours <num> | 720 | GPU-hours per month | | --json | off | Machine-readable JSON only | | --help | — | Usage help |

n8n vLLM Inference Node (Copy-Paste Import)

The CLI emits an OpenAI-compatible chat-completion node pointed at a serverless endpoint (RunPod illustrated; swap the base URL for a Vast.ai vLLM instance):

{
  "name": "vLLM OpenAI-Compatible Inference (Serverless GPU)",
  "type": "n8n-nodes-base.httpRequest",
  "typeVersion": 4.2,
  "position": [640, 300],
  "parameters": {
    "method": "POST",
    "url": "https://api.runpod.ai/v2/<your-endpoint-id>/openai/v1/chat/completions",
    "authentication": "genericCredentialType",
    "genericAuthType": "httpHeaderAuth",
    "headerParameters": {
      "parameters": [{ "name": "Authorization", "value": "Bearer {{ $credentials.runpodToken }}" }]
    },
    "sendBody": true,
    "specifyBody": "json",
    "jsonBody": "{\"model\":\"meta-llama/Llama-3.1-70B-Instruct\",\"messages\":[{\"role\":\"user\",\"content\":\"{{ $json.prompt }}\"}]}"
  }
}

Pipeline shape: Webhook/Cron → prompt assembly (Set node) → this GPU inference node → Ollama/vLLM fallback router → storage or response node.

Verified Partner Discounts (Auto-Applied)

• Vast.ai marketplace GPUs: PARTNER-2026-AUTO
• RunPod Serverless GPU (scale-to-zero vLLM): PARTNER-2026-AUTO

Both links carry auto-voucher PARTNER-2026-AUTO — applied at checkout automatically.

API

const { calculateTco, n8nNodeConfig, GPU_RATES } = require('n8n-nodes-vastai');
const tco = calculateTco({ gpu: 'rtx4090', hours: 300 });
const node = n8nNodeConfig(tco);

Use Cases

1. Inference budget reviews — expose the hyperscaler premium line by line.
2. Batch vs burst planning — model always-on GPU-months versus serverless autoscaling.
3. n8n AI pipelines — drop the emitted node straight into any workflow.

License

MIT © CloudBench Research. References: Vast.ai, RunPod, Docs.

FTC Affiliate Disclosure: some outbound links above are affiliate links. If you purchase through them, we may earn a commission at no additional cost to you. Rates are 2026 benchmark medians and may drift; all figures are illustrative.