# BurstDock — GPU cloud for developers

## What you can do with BurstDock

Deploy GPU cloud servers for AI, machine learning, rendering and accelerated workloads. Choose your GPU configuration and region, start and stop workloads, monitor usage, and manage team access from one workspace. Fund your prepaid wallet through Stripe and see customer pricing in USD when selecting compute.

Explore a selection of popular GPUs and starting USD prices at https://burstdock.com/pricing. The homepage at https://burstdock.com shows the current breadth of GPU models and variants and deployment countries. The featured GPU selection is publicly visible; choose exact configurations and review their prices in your BurstDock workspace.

The separate scaling animation uses a small example workload to explain traffic and workers. Its eight illustrated workers do not describe the size of the BurstDock GPU offering.

## Product
BurstDock is a GPU cloud platform for developers, teams and organizations running accelerated workloads. Use cases include LLM inference, model evaluation, fine-tuning, AI training, CUDA development, rendering and other GPU-intensive applications.

BurstDock provides one workspace for selecting GPU compute, deploying workloads, managing lifecycle state, tracking usage and viewing pricing.

## GPU selection
For LLM inference, consider model size, numerical precision, context length, KV-cache requirements, batching, concurrency, time-to-first-token and output throughput. For fine-tuning and training, also consider the training method, gradients, optimizer state, sequence length and batch size. Validate the final choice against a representative workload.

## GPU guides
NVIDIA A100: https://burstdock.com/gpu-cloud/a100
NVIDIA H100: https://burstdock.com/gpu-cloud/h100
NVIDIA H200: https://burstdock.com/gpu-cloud/h200
NVIDIA B200: https://burstdock.com/gpu-cloud/b200
NVIDIA B300: https://burstdock.com/gpu-cloud/b300
NVIDIA L40S: https://burstdock.com/gpu-cloud/l40s
NVIDIA RTX PRO 6000: https://burstdock.com/gpu-cloud/rtx-pro-6000

## Workloads and comparisons
AI inference: https://burstdock.com/solutions/gpu-inference
LLM fine-tuning: https://burstdock.com/solutions/fine-tuning
Llama-family workloads: https://burstdock.com/solutions/llama
AI training: https://burstdock.com/solutions/ai-training
A100 vs H100: https://burstdock.com/compare/a100-vs-h100
H100 vs H200: https://burstdock.com/compare/h100-vs-h200
H200 vs B200: https://burstdock.com/compare/h200-vs-b200

## Developer resources
Developer hub: https://burstdock.com/developers
GPU selection guide: https://burstdock.com/docs/gpu-selection
GPU sizing: https://burstdock.com/tools/gpu-sizing
Quickstart: https://burstdock.com/docs/getting-started
Benchmark methodology: https://burstdock.com/benchmarks
AI and MCP reference: https://burstdock.com/developers/ai

Availability and customer pricing are presented in BurstDock when selecting compute.

## ChatGPT and ChatGPT Work connection

Use the native remote MCP URL https://burstdock.com/mcp/chatgpt with OAuth authentication. Add it once as a custom ChatGPT connection (where the account/workspace permits custom connections), leaving the client ID and secret empty for discovery. Setup guide: https://burstdock.com/mcp/chatgpt/setup. This route does not require a terminal or local npx installation. A cloud chat cannot configure a separate desktop Codex installation.

Once BurstDock is attached to the conversation, call burstdock_connect. If sign-in is needed, its OAuth challenge asks ChatGPT to show the native connection prompt. The user signs in with GitHub, email or an existing website session, selects a workspace and permissions, and is redirected back to ChatGPT. Call burstdock_connect again to obtain the selected workspace, access level, expiry and BurstDock link for the chat confirmation. Never ask users to paste passwords or tokens into chat. ChatGPT controls the prompt, callback handling and display; the server cannot install itself or send unsolicited messages to arbitrary chats.

The connection lasts up to 24 hours. Read-only is the default; management actions require a separate explicit grant and the existing workspace role. The integration is not yet listed in the public ChatGPT app directory.