Skip to main content

What is Eugo?

Eugo is a cloud platform for high-performance computing. You write ordinary Python in a browser-based notebook, and Eugo compiles, vectorizes, and distributes it across a cluster of ARM and GPU compute nodes.

No servers to provision, no environments to configure, no autoscaling policy to tune.

The shape of it​

Three things, and the relationship between them:

ThingWhat it is
Interactive sessionA running EugoIDE instance in your browser. Where you write code.
ClusterA head node plus compute nodes, launched on demand. Where code runs at scale.
WorkspaceThe environment that owns your storage, libraries, and resource allocation.

You open a session inside a workspace. From that session you launch a cluster. Your notebook code dispatches work to the cluster and collects results back. When the work is done you shut the cluster down, and the session stays open for the next thing.

What you actually write​

The API is deliberately small. Here is a complete distributed program:

import eugo.hpc

@eugo.hpc.distribute
def process(partition):
# Ordinary Python. Nothing here knows it is running on a cluster.
return expensive_transform(partition)

futures = [process.remote(p) for p in partitions]
results = eugo.hpc.get(futures)

Three things changed from a single-machine version: a decorator on the function, .remote(...) where you call it, and one call to resolve the results. The body of process is unmodified.

That is the design goal. Notebook code stays readable, with no cluster initialization to write and no separate configuration file describing your topology: the Connect button in the EugoHPC Manager panel writes the one cell that connects your notebook to a cluster.

The eugo namespace is empty

import eugo on its own gives you nothing. The distributed-compute API lives under eugo.hpc, so import eugo.hpc is the import you want.

What Eugo is not​

The phrase "cloud HPC platform" covers a lot of ground, so here are the boundaries.

  • Not a managed Kubernetes. You do not write manifests or think about pods.
  • Not a job scheduler you submit scripts to. Work is dispatched interactively from a notebook.
  • Not a rewrite of your analysis. Your pandas, NumPy, and scientific Python code carries over.

Supported libraries​

The scientific Python stack is preinstalled and supported: NumPy, pandas, SciPy, scikit-learn, PyTorch, and the usual geospatial and imaging libraries.

Not supported

Eugo does not support TensorFlow, spaCy, or JAX. If your pipeline depends on one of them, plan for that before you migrate. It is the constraint most likely to surprise you.

What comes next​

The next lesson covers why this exists: the specific problem that makes provisioning your own instances unattractive. If you would rather see it working first, skip ahead to opening your first session.


Video: What is Eugo, in 90 seconds. Same material, with a transcript.