GPU Compute

GPU Marketplace

Browse 5000+ GPU servers from multiple providers, compare specs and pricing, and rent specific machines with full control.

Overview

The GPU Marketplace gives you full control over individual machines. Browse thousands of GPU servers from providers worldwide, compare hardware specs, pricing, reliability scores, and rent the exact machine you need.

Best for long-running training jobs where you need a particular hardware configuration.

Browse GPUs

Filter the marketplace by GPU model, VRAM, price, location, and more.

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

gpus = client.marketplace.list(
    gpu_model=["RTX 4090", "A100"],
    min_vram_gb=24,
    min_inet_speed_mbps=500,
    verified_only=True,
    sort_by="price-low",
    page=1,
    page_size=20,
)

print(f"Found {gpus['total_count']} GPUs")
for agent in gpus["agents"]:
    print(f"  {agent['gpu_model']} x{agent['gpu_count']} — ${agent['price_per_hour']}/hr")

Filter Parameters

query

gpu_model

Filter by GPU model names (e.g., "RTX 4090", "A100").

query

min_vram_gb

Minimum VRAM in gigabytes.

query

max_price_per_hour

Maximum hourly price in USD.

query

location

Geographic location filter.

query

verified_only

Show only verified providers with proven reliability.

query

min_gpu_count

Minimum number of GPUs on the machine.

query

min_inet_speed_mbps

Minimum internet speed in Mbps.

query

sort_by

Sort order: price-low, price-high, vram, reliability.

Marketplace Statistics

Get an overview of the marketplace: total GPUs online, price ranges, locations.

stats = client.marketplace.statistics(gpu_model=["RTX 4090"])
print(f"Online: {stats['online']}")
print(f"Min price: ${stats['min_price']}/hr")
print(f"Locations: {stats['locations']}")

View Agent Details

Inspect full hardware specs, pricing tiers, and provider reliability for a specific machine.

agent = client.marketplace.get_agent(agent_id=29279811)

Check Availability

Verify a machine is still available before placing an order.

avail = client.marketplace.check_availability(agent_id=29279811)

Create an Order

Synchronous

order = client.marketplace.create_order(
    agent_id=29279811,
    pricing_type="hour",
    docker_image="pytorch/pytorch:latest",
    ssh_key_ids=[1, 2],
    disk_gb=100,
    volume_id=9,
)

Order Parameters

body

agent_id

The ID of the GPU server to rent.

body

pricing_type

Billing interval: hour, day, week, month.

body

docker_image

Docker image to deploy (e.g., pytorch/pytorch:latest).

body

ssh_key_ids

SSH key IDs to attach to the instance.

Disk size in GB.

Persistent volume to attach.

Async (Recommended for Production)

For better reliability, use async order creation which returns a job ID for polling:

job = client.marketplace.create_order_async(
    agent_id=29279811,
    pricing_type="hour",
    docker_image="pytorch/pytorch:latest",
)

# Poll for status
import time
while True:
    status = client.marketplace.get_order_status(job["job_id"])
    if status["status"] in ("completed", "failed"):
        break
    time.sleep(2)

Use async orders for production systems to handle network timeouts gracefully.

Pricing

Type	Description	Best For
Per Hour	Standard hourly billing	Regular workloads
Per Day	Discounted daily rate	Extended training
Per Week	Weekly discount	Production jobs
Per Month	Best monthly rate	Continuous usage

Longer commitments typically offer 20-40% savings compared to hourly pricing.

Available GPU Types

GPU Model	VRAM	Typical Use Cases
NVIDIA RTX 4090	24GB	Training, inference, rendering
NVIDIA RTX 3090	24GB	ML workloads, cost-effective
NVIDIA A100	40/80GB	Large model training, HPC
NVIDIA H100	80GB	Frontier AI, research
NVIDIA A6000	48GB	Professional rendering, ML
NVIDIA L40S	48GB	Inference, video processing
NVIDIA RTX 5090	32GB	Latest generation, fast training

Was this page helpful?

Last updated Feb 22, 2026

Built with Documentation.AI