GPU Burst

Overview

GPU Burst provisions multiple GPUs across multiple machines simultaneously. Define your primary GPU type and optional fallback alternatives — if your preferred GPU isn't available, the system automatically uses alternatives within your price caps.

Best for: Distributed training, batch inference, large-scale data processing, and any workload that needs to scale fast.

Key features:

• Up to 100 GPUs in a single order
• Fallback GPU types with price caps
• Automatic provisioning across multiple providers
• Per-order billing and run tracking

How It Works

Create a Burst Order

from gpuniq import GPUniq
client = GPUniq(api_key="gpuniq_your_key")

order = client.burst.create_order(
    docker_image="pytorch/pytorch:latest",
    primary_gpu="RTX_4090",
    gpu_count=8,
    extra_gpus=[
        {"gpu_name": "RTX_3090", "max_price": 0.5},
        {"gpu_name": "A100", "max_price": 1.2},
    ],
    volume_id=9,
    disk_gb=200,
)

print(f"Order ID: {order['order_id']}")

curl -X POST "https://api.gpuniq.com/v1/burst/orders" \
  -H "X-API-Key: gpuniq_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "docker_image": "pytorch/pytorch:latest",
    "primary_gpu": "RTX_4090",
    "gpu_count": 8,
    "extra_gpus": [
      {"gpu_name": "RTX_3090", "max_price": 0.5},
      {"gpu_name": "A100", "max_price": 1.2}
    ],
    "disk_gb": 200
  }'

Order Parameters

Cost Estimation

Estimate cost before creating an order:

estimate = client.burst.estimate(
    docker_image="pytorch/pytorch:latest",
    primary_gpu="RTX_4090",
    gpu_count=8,
)

Check Docker image size before deploying:

size = client.burst.check_image_size("pytorch/pytorch:latest")

Manage Orders

# List all burst orders
orders = client.burst.list_orders(limit=100, offset=0)

# Get order details
details = client.burst.get_order(order_id=1)

# Start / stop / delete
client.burst.start_order(order_id=1)
client.burst.stop_order(order_id=1)
client.burst.delete_order(order_id=1)

Billing & Run History

Track costs and GPU run history per order:

# Billing transactions
txns = client.burst.transactions(order_id=1, limit=50)

# GPU run history
runs = client.burst.runs(order_id=1, limit=50)

Comparison: Burst vs Marketplace vs Dex-Cloud