Skip to content
vastai-cost-tuning logo

Vast.ai Cost Tuning

vastai-cost-tuning

Optimize Vast.ai GPU cloud costs through smart instance selection and

SKILL.md

Full skill instructions

Vast.ai Cost Tuning

Overview

Minimize Vast.ai GPU cloud costs by choosing the right GPU for your workload, leveraging interruptible (spot) instances, eliminating idle compute, and implementing auto-destroy safeguards. Vast.ai pricing is dynamic and varies significantly: RTX 4090 ($0.15-0.30/​hr), A100 80GB ($1.00-2.00/​hr), H100 SXM ($2.50-4.00/​hr).

Prerequisites

  • Vast.ai account with billing history
  • Understanding of your workload's GPU requirements
  • vastai CLI installed

Instructions

Step 1: GPU Selection by Cost-Efficiency

# Compare cost-per-TFLOP across GPU types
GPU_SPECS = {
    "RTX_4090":  {"fp16_tflops": 82.6,  "vram": 24},
    "A100":      {"fp16_tflops": 77.97, "vram": 80},
    "H100_SXM":  {"fp16_tflops": 267,   "vram": 80},
    "RTX_3090":  {"fp16_tflops": 35.6,  "vram": 24},
    "A6000":     {"fp16_tflops": 38.7,  "vram": 48},
}

def cost_per_tflop(gpu_name, dph):
    specs = GPU_SPECS.get(gpu_name, {"fp16_tflops": 1})
    return dph / specs["fp16_tflops"]

# Often RTX 4090 is the best value for inference
# A100 is best for training large models needing >24GB VRAM
# H100 is best only when wall-clock time justifies 10x price premium

Step 2: Spot vs On-Demand Analysis

# Interruptible (spot) instances are 30-60% cheaper
vastai search offers 'num_gpus=1 gpu_name=RTX_4090 rentable=true' \
  --order dph_total --limit 5
# Compare interruptible vs on-demand pricing
# Use interruptible for: batch inference, checkpointed training
# Use on-demand for: final training epochs, production inference

Step 3: Auto-Destroy Safeguards

import time, subprocess, json

def auto_destroy_after(instance_id, max_hours=4):
    """Destroy instance after max_hours to prevent cost overruns."""
    max_seconds = max_hours * 3600
    time.sleep(max_seconds)
    subprocess.run(["vastai", "destroy", "instance", str(instance_id)], check=True)
    print(f"Instance {instance_id} auto-destroyed after {max_hours}h")

# Run in background thread when provisioning
import threading
watchdog = threading.Thread(target=auto_destroy_after, args=(inst_id, 4), daemon=True)
watchdog.start()

Step 4: Idle Instance Detection

#!/​bin/​bash
# Find and destroy idle instances (GPU util < 10% for >10 min)
vastai show instances --raw | python3 -c "
import sys, json
for inst in json.load(sys.stdin):
    if inst.get('actual_status') == 'running':
        gpu_util = inst.get('gpu_util', 0)
        if gpu_util < 10:
            print(f'IDLE: Instance {inst[\"id\"]} GPU util={gpu_util}% '
                  f'(\${inst.get(\"dph_total\", 0):.3f}/​hr)')
"

Step 5: Cost Reporting

def daily_cost_report():
    """Calculate current daily burn rate from running instances."""
    result = subprocess.run(
        ["vastai", "show", "instances", "--raw"],
        capture_output=True, text=True)
    instances = json.loads(result.stdout)

    total_hourly = 0
    for inst in instances:
        if inst.get("actual_status") == "running":
            dph = inst.get("dph_total", 0)
            total_hourly += dph
            print(f"  {inst['id']}: {inst.get('gpu_name')} ${dph:.3f}/​hr")

    print(f"\nTotal: ${total_hourly:.3f}/​hr = ${total_hourly * 24:.2f}/​day")

Cost Optimization Checklist

  • Always search with --order dph_total to find cheapest offers
  • Use interruptible instances for checkpointed workloads
  • Implement auto-destroy timeout on all instances
  • Monitor GPU utilization; destroy idle instances
  • Use RTX 4090 for workloads that fit in 24GB VRAM
  • Only use H100 when wall-clock time savings justify cost premium
  • Pre-install dependencies in Docker images (avoid paying for pip install)

Output

  • GPU cost-efficiency analysis by model
  • Spot vs on-demand comparison
  • Auto-destroy watchdog for cost protection
  • Idle instance detection script
  • Daily cost burn rate report

Error Handling

ErrorCauseSolution
Unexpected $50+ billForgot to destroy instancesImplement auto-destroy watchdog
GPU idle at $2/​hrWaiting for data downloadPre-stage data before provisioning GPU
Spot preemption mid-jobCheapest instance reclaimedCheckpoint frequently; auto-recover

Resources

Next Steps

For reference architecture, see vastai-reference-architecture.

Examples

Budget cap: Set dph_total<=0.25 in search queries and auto_destroy_after(inst_id, 4) to cap any single job at $1.00.

GPU comparison: Run the same workload on RTX 4090 ($0.20/​hr) vs A100 ($1.50/​hr). If the A100 finishes in less than 1/​7th the time, it's cheaper overall.

More skills from jeremylongshore

openrouter-compliance-review logo
jeremylongshore/claude-code-plugins-plus-skills

openrouter-compliance-review

Review OpenRouter integration for regulatory compliance (SOC2, GDPR,

2.8K 0
View
maintainx-data-handling logo
jeremylongshore/claude-code-plugins-plus-skills

maintainx-data-handling

Data synchronization, ETL patterns, and data management for MaintainX.

2.8K 0
View
groq-core-workflow-a logo
jeremylongshore/claude-code-plugins-plus-skills

groq-core-workflow-a

Execute Groq primary workflow: chat completions with tool use and JSON

2.8K 0
View
validating-csrf-protection logo
jeremylongshore/claude-code-plugins-plus-skills

validating-csrf-protection

Validate CSRF protection implementations for security gaps. Use when

2.8K 0
View
api-contract logo
jeremylongshore/claude-code-plugins-plus-skills

api-contract

Configure this skill should be used when the user asks about "API contract",

2.8K 0
View
ideogram-incident-runbook logo
jeremylongshore/claude-code-plugins-plus-skills

ideogram-incident-runbook

Execute Ideogram incident response with triage, mitigation, and postmortem.

2.8K 0
View
gamma-sdk-patterns logo
jeremylongshore/claude-code-plugins-plus-skills

gamma-sdk-patterns

Reusable patterns for the Gamma REST API (no SDK exists).

2.8K 0
View
deepgram-sdk-patterns logo
jeremylongshore/claude-code-plugins-plus-skills

deepgram-sdk-patterns

Apply production-ready Deepgram SDK patterns for TypeScript and Python.

2.8K 0
View
openrouter-upgrade-migration logo
jeremylongshore/claude-code-plugins-plus-skills

openrouter-upgrade-migration

Migrate to OpenRouter from direct provider APIs or upgrade between SDK/model

2.8K 0
View
openrouter-common-errors logo
jeremylongshore/claude-code-plugins-plus-skills

openrouter-common-errors

Diagnose and fix common OpenRouter API errors. Use when encountering

2.8K 0
View
linear-cost-tuning logo
jeremylongshore/claude-code-plugins-plus-skills

linear-cost-tuning

Optimize Linear API usage, reduce unnecessary calls, and maximize

2.8K 0
View

Popular AI tools

Kaiber logo
Video

Kaiber

Generate, edit, and beat-sync AI video with leading models in one workspace.

Paid
View
Vimcal logo
Productivity

Vimcal

The world's fastest calendar for remote work

Free
View

Transform Your Design with AI Designer by ImgCreator.ai

Freemium
View
Akool AI logo
Content & writing

Akool AI

Revolutionizing Video Production with AI-Powered Creativity

Paid
View

Extend an image past the frame and let AI fill the new aspect ratio.

Freemium
View
StarByFace logo
Security

StarByFace

Discover your celebrity doppelgänger with StarByFace!

Free
View
C

ChainClarity explains 700+ crypto whitepapers in plain English, with layered summaries, comparisons, research tools, alerts, and a $4.99 Pro plan.

Freemium
View
Opus Clip logo
Coding & apps

Opus Clip

Opus.ai: Revolutionize Your Web Experience

Free
View