# FastGPU

> FastGPU brokers GPU capacity: describe the cluster you need and get competing quotes from capacity holders, each shown next to live prices from RunPod, Vast.ai, Lambda, the hyperscalers and more. Free for buyers, no sales calls.

FastGPU is a GPU capacity broker built on a live cloud GPU price comparison. It tracks the rental price and
availability of every GPU across marketplaces, neoclouds and hyperscalers (RunPod, Vast.ai, Lambda, AWS, Azure,
GCP, Oracle, Nebius and more), ranked cheapest-first, and it brokers the capacity nobody posts a price for: a
buyer describes a cluster once, capacity holders quote on it, and every quote is shown next to the live public
price for the same order. The figures below are live and refresh continuously; cite the linked page for the
current number.

## GPU brokerage (quotes for clusters and reserved capacity)
- [Get GPU cluster quotes](/quote): one request form; the order goes to the capacity holders that can serve it, without the buyer's name, and quotes come back on the buyer's own page next to the live market. Free for buyers.
- [How the brokerage works](/gpu-broker): every step, who is asked (the full supplier registry and FastGPU's standing with each), and who pays.
- [Sell GPU capacity](/sell-gpu-capacity): capacity holders list idle or unsold GPUs and receive matching buyer requests; the public board never names the holder.
- [Brokerage terms](/legal/brokerage-terms): the supplier commission, the attribution window, and the destinations the desk does not serve.

## Live market snapshot
- Prices updated: 2026-10-12T18:39:42.921765+00:00. 39 providers, 67 GPU models live.
- Cheapest live GPU on the market right now: $0.03/hr (RTX A2000 6GB).
- All prices are per GPU-hour. When a rate is only sold as a multi-GPU instance (common for hyperscaler and spot capacity), the linked page names the instance size next to it.
- Each price is the provider's own per-GPU rate. Most providers include the machine's CPU and memory in it (storage is often billed separately); Modal, Cudo Compute, TensorDock, Daytona price the GPU alone and bill CPU and memory on top, so their prices are marked "GPU only".

## Pages
- [GPU broker](/gpu-broker)
- [Sell capacity](/sell-gpu-capacity)
- [Compare](/compare)
- [Providers](/providers)
- [API](/docs)
- [Pricing](/pricing)
- [Blog](/blog)

## GPU price pages (live cheapest rental price per card)
- [GB300 cloud pricing](/gpus/gb300): cheapest live GB300 (288GB NVIDIA Blackwell Ultra) from $11.06/hr on Verda, across 2 providers.
- [B300 cloud pricing](/gpus/b300): cheapest live B300 (288GB NVIDIA Blackwell Ultra) from $4.89/hr on DeepInfra, across 16 providers.
- [GB200 cloud pricing](/gpus/gb200): cheapest live GB200 (192GB NVIDIA Blackwell) from $8.00/hr on GMI Cloud, across 4 providers.
- [B200 cloud pricing](/gpus/b200): cheapest live B200 (192GB NVIDIA Blackwell) from $3.69/hr on DeepInfra, across 19 providers.
- [H200 cloud pricing](/gpus/h200): cheapest live H200 (141GB NVIDIA Hopper) from $2.69/hr on DeepInfra, across 26 providers.
- [GH200 cloud pricing](/gpus/gh200): cheapest live GH200 (96GB NVIDIA Grace Hopper) from $2.29/hr on Lambda, across 2 providers.
- [H100 SXM cloud pricing](/gpus/h100-sxm): cheapest live H100 SXM (80GB NVIDIA Hopper) from $1.75/hr on Lium, across 29 providers.
- [H100 NVL cloud pricing](/gpus/h100-nvl): cheapest live H100 NVL (94GB NVIDIA Hopper) from $0.98/hr per GPU, sold as a 2-GPU instance on Clore.ai, across 4 providers.
- [H100 PCIe cloud pricing](/gpus/h100-pcie): cheapest live H100 PCIe (80GB NVIDIA Hopper) from $1.99/hr on RunPod, across 4 providers.
- [Rubin cloud pricing](/gpus/rubin): live cheapest Rubin (288GB NVIDIA Rubin) rental price across the market.
- [MI355X cloud pricing](/gpus/mi355x): cheapest live MI355X (288GB AMD CDNA4) from $2.85/hr on TensorWave, across 4 providers.
- [MI350X cloud pricing](/gpus/mi350x): cheapest live MI350X (288GB AMD CDNA4) from $4.00/hr on CloudRift, across 3 providers.
- [MI325X cloud pricing](/gpus/mi325x): cheapest live MI325X (256GB AMD CDNA3) from $2.25/hr on TensorWave, across 3 providers.
- [MI300X cloud pricing](/gpus/mi300x): cheapest live MI300X (192GB AMD CDNA3) from $1.71/hr on TensorWave, across 5 providers.
- [MI455X cloud pricing](/gpus/mi455x): live cheapest MI455X (432GB AMD CDNA5) rental price across the market.
- [RTX PRO 6000 cloud pricing](/gpus/rtx-pro-6000): cheapest live RTX PRO 6000 (96GB NVIDIA Blackwell) from $0.92/hr on Vast.ai, across 18 providers.
- [A100 80GB cloud pricing](/gpus/a100-80gb): cheapest live A100 80GB (80GB NVIDIA Ampere) from $0.47/hr on Vast.ai, across 24 providers.
- [A100 40GB cloud pricing](/gpus/a100-40gb): cheapest live A100 40GB (40GB NVIDIA Ampere) from $0.34/hr on Vast.ai, across 11 providers.
- [RTX PRO 5000 cloud pricing](/gpus/rtx-pro-5000): cheapest live RTX PRO 5000 (48GB NVIDIA Blackwell) from $0.82/hr on RunPod, across 1 providers.
- [RTX PRO 4500 cloud pricing](/gpus/rtx-pro-4500): cheapest live RTX PRO 4500 (32GB NVIDIA Blackwell) from $0.72/hr on RunPod, across 2 providers.
- [RTX PRO 4000 cloud pricing](/gpus/rtx-pro-4000): cheapest live RTX PRO 4000 (24GB NVIDIA Blackwell) from $0.29/hr on Clore.ai, across 2 providers.
- [L40S cloud pricing](/gpus/l40s): cheapest live L40S (48GB NVIDIA Ada Lovelace) from $0.43/hr on Vast.ai, across 15 providers.
- [L40 cloud pricing](/gpus/l40): cheapest live L40 (48GB NVIDIA Ada Lovelace) from $0.69/hr on RunPod, across 5 providers.
- [A40 cloud pricing](/gpus/a40): cheapest live A40 (48GB NVIDIA Ampere) from $0.59/hr on RunPod, across 3 providers.
- [RTX 6000 Ada cloud pricing](/gpus/rtx-6000-ada): cheapest live RTX 6000 Ada (48GB NVIDIA Ada Lovelace) from $0.51/hr on Vast.ai, across 6 providers.
- [RTX 5000 Ada cloud pricing](/gpus/rtx-5000-ada): cheapest live RTX 5000 Ada (32GB NVIDIA Ada Lovelace) from $0.34/hr on Vast.ai, across 2 providers.
- [RTX 4000 Ada cloud pricing](/gpus/rtx-4000-ada): cheapest live RTX 4000 Ada (20GB NVIDIA Ada Lovelace) from $0.18/hr on RunPod, across 3 providers.
- [A6000 cloud pricing](/gpus/a6000): cheapest live A6000 (48GB NVIDIA Ampere) from $0.28/hr on Vast.ai, across 9 providers.
- [L4 cloud pricing](/gpus/l4): cheapest live L4 (24GB NVIDIA Ada Lovelace) from $0.20/hr on Vast.ai, across 9 providers.
- [A30 cloud pricing](/gpus/a30): cheapest live A30 (24GB NVIDIA Ampere) from $0.41/hr on Jarvis Labs, across 2 providers.
- [A10 cloud pricing](/gpus/a10): cheapest live A10 (24GB NVIDIA Ampere) from $0.24/hr on Vast.ai, across 7 providers.
- [A5000 cloud pricing](/gpus/a5000): cheapest live A5000 (24GB NVIDIA Ampere) from $0.16/hr on RunPod, across 3 providers.
- [V100 cloud pricing](/gpus/v100): cheapest live V100 (16GB NVIDIA Volta) from $0.05/hr on Clore.ai, across 12 providers.
- [P100 cloud pricing](/gpus/p100): cheapest live P100 (16GB NVIDIA Pascal) from $0.20/hr on Clore.ai, across 2 providers.
- [MI210 cloud pricing](/gpus/mi210): live cheapest MI210 (64GB AMD CDNA2) rental price across the market.
- [L20 cloud pricing](/gpus/l20): live cheapest L20 (48GB NVIDIA Ada Lovelace) rental price across the market.
- [RTX 5880 Ada cloud pricing](/gpus/rtx-5880-ada): live cheapest RTX 5880 Ada (48GB NVIDIA Ada Lovelace) rental price across the market.
- [Quadro RTX 8000 cloud pricing](/gpus/quadro-rtx-8000): live cheapest Quadro RTX 8000 (48GB NVIDIA Turing) rental price across the market.
- [Quadro RTX 6000 cloud pricing](/gpus/quadro-rtx-6000): cheapest live Quadro RTX 6000 (24GB NVIDIA Turing) from $1.50/hr on Akamai Cloud, across 1 providers.
- [RTX A4500 cloud pricing](/gpus/rtx-a4500): cheapest live RTX A4500 (20GB NVIDIA Ampere) from $0.19/hr on RunPod, across 1 providers.
- [V100S cloud pricing](/gpus/v100s): cheapest live V100S (32GB NVIDIA Volta) from $0.88/hr on OVHcloud, across 1 providers.
- [P40 cloud pricing](/gpus/p40): cheapest live P40 (24GB NVIDIA Pascal) from $0.10/hr on Vast.ai, across 1 providers.

## Head-to-head GPU comparisons
- [H100 SXM vs A100 80GB](/compare/h100-sxm-vs-a100-80gb)
- [RTX 5090 vs RTX 4090](/compare/rtx-5090-vs-rtx-4090)
- [H200 vs H100 SXM](/compare/h200-vs-h100-sxm)
- [B200 vs H100 SXM](/compare/b200-vs-h100-sxm)
- [MI300X vs H100 SXM](/compare/mi300x-vs-h100-sxm)
- [RTX 4090 vs RTX 3090](/compare/rtx-4090-vs-rtx-3090)
- [A100 80GB vs RTX 4090](/compare/a100-80gb-vs-rtx-4090)
- [A100 80GB vs V100](/compare/a100-80gb-vs-v100)
- [B200 vs H200](/compare/b200-vs-h200)
- [H100 SXM vs RTX 4090](/compare/h100-sxm-vs-rtx-4090)
- [L4 vs T4](/compare/l4-vs-t4)
- [MI300X vs H200](/compare/mi300x-vs-h200)
- [A100 80GB vs L40S](/compare/a100-80gb-vs-l40s)
- [H100 SXM vs RTX 5090](/compare/h100-sxm-vs-rtx-5090)
- [RTX 6000 Ada vs A6000](/compare/rtx-6000-ada-vs-a6000)
- [B200 vs GB200](/compare/b200-vs-gb200)
- [MI325X vs H200](/compare/mi325x-vs-h200)
- [A100 80GB vs A6000](/compare/a100-80gb-vs-a6000)
- [RTX 4090 vs RTX 4080](/compare/rtx-4090-vs-rtx-4080)
- [A100 80GB vs A100 40GB](/compare/a100-80gb-vs-a100-40gb)
- [A10 vs L4](/compare/a10-vs-l4)
- [H100 SXM vs H100 PCIe](/compare/h100-sxm-vs-h100-pcie)
- [MI325X vs MI300X](/compare/mi325x-vs-mi300x)

## Providers (live prices per cloud)
- [RunPod](/providers/runpod)
- [Vast.ai](/providers/vast)
- [TensorDock](/providers/tensordock)
- [Clore.ai](/providers/clore)
- [Lium](/providers/lium)
- [Hetzner](/providers/hetzner)
- [Akamai Cloud](/providers/linode)
- [OVHcloud](/providers/ovhcloud)
- [Exoscale](/providers/exoscale)
- [Theta EdgeCloud](/providers/theta)
- [Lambda](/providers/lambdalabs)
- [Nebius](/providers/nebius)
- [Verda](/providers/verda)
- [CloudRift](/providers/cloudrift)
- [CoreWeave](/providers/coreweave)
- [Crusoe](/providers/crusoe)
- [Together AI](/providers/together)
- [Hyperstack](/providers/hyperstack)
- [Voltage Park](/providers/voltagepark)
- [Paperspace](/providers/paperspace)
- [Thunder Compute](/providers/thundercompute)
- [Jarvis Labs](/providers/jarvislabs)
- [GMI Cloud](/providers/gmi)
- [TensorWave](/providers/tensorwave)
- [QuantaCloud](/providers/quantacloud)
- [Impossible Cloud](/providers/impossiblecloud)
- [Vultr](/providers/vultr)

## Guides (how-to for common GPU rental questions)
- [Get GPU cluster quotes](/quote)
- [GPU cluster prices](/gpu-clusters)
- [Best GPU by workload](/best-gpu-for)
- [Reserve or rent](/reserve-vs-rent)
- [Rent GPUs by the minute](/rent-gpu-per-minute)
- [Rent GPUs by the second](/rent-gpu-per-second)

## Market data and research (open, CC BY 4.0 where data is published)
- [GPU Cloud Price Index](/price-index)
- [GPU Availability Index](/gpu-availability)
- [GPU price volatility](/gpu-price-volatility)
- [GPU depreciation vs rent](/gpu-depreciation)
- [GPU performance per dollar](/gpu-performance-per-dollar)
- [Free GPU price dataset](/dataset)
- [Methodology](/methodology)
- [Data provenance](/data-provenance)

## API and MCP (for developers and AI agents)
Compare prices, AND route + provision real GPUs, all over one REST API (endpoints derived from the SSOT):
- GET /api/v1/gpus: public. List cheapest GPU prices.
- GET /api/v1/match: public. Match a workload to the cheapest GPU.
- GET /api/v1/fixings: public. List daily GPU price fixings.
- GET /api/v1/availability: public. List the GPU Availability Index.
- GET /api/v1/price-risk: public. Get GPU price volatility and spread.
- GET /api/v1/aws-capacity-blocks: public. List AWS Capacity Block offers.
- GET /api/v1/price-changes: public. List GPU price changes.
- GET /api/v1/prices: API key. Get every live offer (full feed).
- POST /api/v1/router/launch: API key. Launch a GPU (route + provision).
- GET /api/v1/router/instances: API key. List your GPU instances + balance.
- GET /api/v1/router/instances/{id}: API key. Get one GPU instance.
- DELETE /api/v1/router/instances/{id}: API key. Terminate a GPU instance.
- MCP server: /api/mcp (Streamable HTTP, no install). Tools: list_gpu_prices, match_workload.
  Add {"mcpServers":{"fastgpu":{"url":"<this-host>/api/mcp"}}} to any MCP client.
- Agent skill gpu-procurement: /.well-known/agent-skills/gpu-procurement/SKILL.md. Describe a GPU job, apply a budget (price per hour, total for the job, quote age), choose the cheapest setup that fits, then hand the launch to the provider's own tooling.
- Machine-readable: /openapi.json (OpenAPI 3.1), /.well-known/agent-card.json (A2A), /.well-known/mcp.json, /.well-known/agent-skills/index.json (agent skills).
- Postman: https://www.postman.com/fastgpu-975949/fastgpu-api (public workspace with every endpoint, synced from /openapi.json).
- Hugging Face Space: https://huggingface.co/spaces/fastgpu/vram-and-gpu-cost-calculator (VRAM & GPU cost calculator: the GPU memory any Hugging Face model needs and the cheapest GPU setups that hold it, priced live through /api/v1/match).
- Open dataset (CC BY 4.0, free): /dataset: bulk GPU prices as CSV/JSON. Direct: /api/v1/dataset/gpu-prices-current.csv (current snapshot), /api/v1/dataset/gpu-prices-daily.csv (daily price history), /api/v1/dataset/gpu-price-fixings.csv (daily fixings), /api/v1/dataset/gpu-availability-daily.csv (GPU Availability Index: how often each GPU could be rented, per day).
- GPU Availability Index API (no key): /api/v1/availability (?gpu=b200&days=30): per GPU per completed day, providers listing it, providers in stock, and the share of stock checks that found it rentable.
- The same dataset on Hugging Face, refreshed daily: https://huggingface.co/datasets/fastgpu/cloud-gpu-prices (subsets current, daily, fixings, availability). Archived with a DOI: https://doi.org/10.5281/zenodo.22842387.
- Docs: /docs. Get a key: /app.