Elastic Python Compute

Scale Python AI workloads on enterprise data

Run the same Python code from notebook to production, with elastic Ray clusters, CPU or GPU compute, and fast access to your data.

문의
Elastic Python Compute
How it works

What is Elastic Python Compute?

Elastic Python Compute is how Teradata AI Studio runs your Python code. Smaller jobs run in a single container. Bigger jobs spread across a Ray cluster, a group of machines that work on the task at the same time. It’s built into AI Studio, so there’s nothing separate to buy or set up.

Benefits

Scale Python without starting over

data_object

Keep your code as you scale

Run the same Python from first experiment to enterprise scale, well beyond the limits of a single notebook or machine.

align_vertical_bottom

Match compute to each job

Run light jobs in a single container. Spread heavy ones across Ray clusters. Choose CPU or GPU for each workload.

insert_chart

Work where your data lives

Keep development connected to governed enterprise data and AI Studio services, so you move faster without opening security or governance gaps.

Elastic Python Compute Architecture
Overview

Python execution built for production AI

Use the Python libraries you already know. AI Studio handles the runtime, scaling, and data access.

Elastic Python Compute is the Python execution layer in Teradata AI Studio. Pick the model that fits each workload: a single container for lighter tasks, or a distributed Ray cluster for larger, parallel work.
 
Read enterprise data, run your code, and save models, results, and job outputs for the applications and analytics that depend on them. 

Workflow

4 steps from notebook to scale

Settings_suggest

Configure

Start in an AI Studio notebook or the Python SDK. Set data access, packages, memory, workers, and CPU or GPU requirements.

takeout_dining

Provision

The platform launches a single container or a Ray cluster, based on your workload design.

terminal

Run

Your code reads and writes data through supported connections and runs tasks across the compute you selected.

save

Save

Model artifacts, results, and job outputs persist for downstream AI Studio and enterprise workflows.

Core capabilities

Compute that adapts to the workload

Single-container execution

Prototype, explore data, and run smaller training jobs without cluster overhead.

Distributed execution with Ray

Parallelize large-scale machine learning, batch inference, tuning, and multimodal pipelines across Ray workers.

Elastic scaling

Set minimum and maximum nodes, and let Ray clusters scale within those limits as demand changes.

CPU and GPU flexibility

Use general-purpose compute for analytics and accelerated compute for machine learning, computer vision, language, and graph workloads.

Open Python ecosystem

Use familiar open-source libraries and frameworks in containerized environments built for custom analytics.

High-performance data access

Read and write in parallel between Ray compute and enterprise data, using Arrow Flight servers and Teradata connectivity.

Workload isolation

Run on dedicated clusters or node pools to keep compute-intensive jobs from slowing other platform activity.

Ray cluster autoscaling

Automatically adds workers as workload demand increases and removes idle workers as demand falls, within configured minimum and maximum limits.

Batch workload submission through Ray jobs

Enables teams to submit, monitor, and retrieve logs for Python workloads through APIs, without keeping an interactive notebook session open.

Use cases

Put scalable Python to work

Start with the execution pattern that fits today’s workload. Move to distributed CPU or GPU compute when data volume, model size, or concurrency grows.

Creating Ray Cluster

Machine learning training and scoring

Train, tune, and score models in a single container or across a Ray cluster, depending on workload size.

Language model and batch inference

Run summarization, translation, sentiment analysis, and other text processing in parallel, using the models and libraries your team chooses.

Multimodal and video intelligence

Distribute OCR, transcription, sound-event detection, object detection, image classification, and annotation across workers. Write structured and semi-structured results to Teradata.

Graph analytics and advanced data science

Run large-scale network, relationship, clustering, and partitioning analysis on distributed CPU or GPU compute.

Elastic Python Compute FAQs

It’s the managed Python execution capability in Teradata AI Studio. It provides containerized Python runtimes, single-container execution, and distributed execution on Ray clusters. It isn’t a standalone product. 

Use a single container for prototyping, ad hoc analysis, and smaller training jobs. Use a Ray cluster when work benefits from parallel processing, such as larger machine learning jobs, batch inference, tuning, multimodal pipelines, and CPU- or GPU-intensive workloads. 

Yes. Choose CPU or GPU node pools to match each workload. GPU is supported only on Teradata Factory.

It connects to Teradata and supported enterprise data sources through platform connections, with parallel reads and writes using Arrow Flight servers. 

Yes. It’s designed for native Python workloads and familiar open-source libraries. 

You follow one repeatable workflow: configure compute, provision the runtime, run your code, and save outputs for downstream AI Studio processes. The code stays the same at every scale. 

Machine and deep learning training and scoring, language model batch inference, graph analytics, advanced data science, and multimodal video-processing pipelines. 

Ray is the distributed execution layer. It spreads Python work across workers and CPU or GPU resources for larger, parallel workloads. 

Take Python from notebook to production

See how Teradata AI Studio runs your Python AI and analytics workloads at scale, on governed enterprise data.

문의