gpuhunt

Easy access to GPU pricing data for major cloud providers: AWS, Azure, GCP, etc. The catalog includes details about prices, locations, CPUs, RAM, GPUs, and spots (interruptible instances).

Usage

import gpuhunt

items = gpuhunt.query(
    min_memory=16,
    min_cpu=8,
    min_gpu_count=1,
    max_price=1.0,
)

print(*items, sep="\n")

List of all available filters:

• provider: name of the provider to filter by. If not specified, all providers will be used. One or many
• min_cpu: minimum number of CPUs
• max_cpu: maximum number of CPUs
• min_memory: minimum amount of RAM in GB
• max_memory: maximum amount of RAM in GB
• min_gpu_count: minimum number of GPUs
• max_gpu_count: maximum number of GPUs
• gpu_name: name of the GPU to filter by. If not specified, all GPUs will be used. One or many
• min_gpu_memory: minimum amount of GPU VRAM in GB for each GPU
• max_gpu_memory: maximum amount of GPU VRAM in GB for each GPU
• min_total_gpu_memory: minimum amount of GPU VRAM in GB for all GPUs combined
• max_total_gpu_memory: maximum amount of GPU VRAM in GB for all GPUs combined
• min_disk_size: minimum disk size in GB (not fully supported)
• max_disk_size: maximum disk size in GB (not fully supported)
• min_price: minimum price per hour in USD
• max_price: maximum price per hour in USD
• min_compute_capability: minimum compute capability of the GPU
• max_compute_capability: maximum compute capability of the GPU
• spot: if False, only ondemand offers will be returned. If True, only spot offers will be returned

Advanced usage

from gpuhunt import Catalog

catalog = Catalog()
catalog.load(version="20240508")
items = catalog.query()

print(*items, sep="\n")

Supported providers

• AWS
• Azure
• Cudo Compute
• DataCrunch
• GCP
• LambdaLabs
• OCI
• RunPod
• TensorDock
• Vast AI

See also

• dstack