Skip to content
runcrate
Console

Data processing

GPU data pipelines for ETL and dataset prep.

Run RAPIDS cuDF and Dask-CUDA on an H100 or A100 instance to clean, deduplicate and tokenize training data, then delete the instance when the job ends.

Launch an instanceSee GPU prices
H100from $3.58/hr

What you can build

Three jobs people run here. Each on a product you can start today.

Instance

DataFrames on the GPU

cuDF reads Parquet, CSV and JSON and runs pandas-style joins and group-bys on the GPU. Dask-CUDA spreads one job across every GPU.

Compare GPUs
Crate

Start from the Devbox crate

The Devbox crate is a ready image your instance boots, with CUDA, Python with uv, Jupyter and Docker. Add RAPIDS with one install.

See what an instance includes
Storage

Data that outlives the job

Keep raw and processed data on a volume between pipeline stages, or read from and write to your own S3 bucket.

Read how volumes work

Examples

Ask your agent. Or read the code it writes.

Connected by the line, Claude Code or Codex calls the Runcrate tools for you: launch, copy files, run, delete.

Connect your agent first

Add Runcrate's MCP server to yourself: run curl -fsSL"https://runcrate.ai/mcp.md"and follow it.

Add to CursorInstall in VS Code
Works inClaude CodeCodexOpenCodeClineRoo CodeKilo CodeCursor

Docs
Launch an A100 with the Devbox crate, run dedupe.py on the dataset, and delete the instance when it finishes. Calls the Runcrate tool create_instance. It returns: status deploying: 1x A100. Calls the Runcrate tool instance_status. It returns: status running, with an address. Calls the Runcrate tool file_upload. It returns: dedupe.py, on the instance. Calls the Runcrate tool ssh_execute. It returns: exit code 0: the dataset, deduplicated. Calls the Runcrate tool delete_instance. It returns: deleted: true.

Models and tools

GPU data tools. All of them run on your instance.

NameRoleRuns on
RAPIDS cuDFGPU DataFrames for ETLInstance
Dask-CUDASpread work across GPUsInstance
cuMLGPU machine learning and featuresInstance
NVIDIA DALIGPU image and audio loadingInstance
JupyterExplore data interactivelyInstance

How it works

From first call to production. Three steps.

  1. 01

    Bring your data

    Copy files over with runcrate cp, read from your own S3 bucket, or attach a volume that already holds the dataset.

  2. 02

    Run the GPU pipeline

    Transform with cuDF, distribute with Dask-CUDA and build features with cuML. Root access means any library you need installs.

  3. 03

    Export and shut down

    Write Parquet, tokenized corpora or features to a volume or your bucket, then delete the instance to stop the bill.

Pricing

What it costs. Live prices, as billed.

See every GPU price

Name

As billed

  • H100

    GPU, 80GB, cheapest offer

    $3.58/hr

  • A100 80GB

    GPU, 80GB, cheapest offer

    $1.43/hr

  • L40S

    GPU, 48GB, cheapest offer

    $0.55/hr