Leverage GPUs Anywhere with Alluxio Enterprise AI 3.2 release

Enables 97%+ GPU utilization, features new native integration with Python ecosystem and more.

wHAT’S NEW

ALLUXIO WEBINAR | What’s new in Alluxio Enterprise AI 3.2: Leverage GPU Anywhere, Pythonic Filesystem API, Write Checkpointing and more

In today’s AI-driven world, organizations face unprecedented demands for powerful AI infrastructure to fuel their model training and serving workloads. Performance bottlenecks, cost inefficiencies, and management complexities pose significant challenges for AI platform teams supporting large-scale model training and serving. On July 9, 2024, we introduced Alluxio Enterprise AI 3.2, a groundbreaking solution designed to address these critical issues in the ever-evolving AI landscape.

In this webinar, Shouwei Chen, product manager at Alluxio, will introduce exciting new features of Alluxio Enterprise AI 3.2: leveraging GPU resources anywhere, enhanced I/O performance with 97%+ GPU utilizationk, achieving the same performance as HPC storage, new Python FileSystem API, and other new features, include advanced cache management, rolling upgrades, and CSI failover.

The Alluxio Data Platform powers many of the most critical data-driven applications in the world.

Why Alluxio

Uniquely positioned between compute and storage, Alluxio provides a single pane of glass for enterprises to manage data and AI workloads across diverse infrastructure environments with ease. Alluxio Data Platform has two product offerings – Alluxio Enterprise Data and Alluxio Enterprise AI. Choose the product offering based on your workload’s needs, and enjoy epic performance, seamless data access, simplified data engineering, and cost savings.

Epic Performance

Accelerate model training by 20X and model serving by 10X.

Maximize Infra ROI

Build on existing data lakes. 90% + GPU utilization.

Seamless Data Access

Eliminate data copies.
Access data seamlessly anywhere.

Data and AI Infrastructure Challenges

Analytics

Data is siloed across different storage systems, clouds, and regions

Data is often remote from the application leading to low performance. Replicating data introduces additional data engineering complexity and costs.

Managing different tech stacks is costly and complicated

Whenever new storage is introduced application developers may need to rewrite applications to accommodate new storage protocols. The myriad storage options require too much labor to maintain and operate. Poor user experience leads to increased user complaints to the data platform team.

Cloud and network costs are difficult to manage and predict

Everytime data is replicated from one silo to another, costs are incurred. API costs and egress costs add up over time and are difficult to predict.

AI/ML

Data infrastructure cannot meet the demands of AI

Data infrastructures are not built to meet the requirements of AI workloads. Data platform teams are tasked to retool the data infrastructure and introduce costly hardware such as high performance computing or specialized storage.

GPUs are scarce and often underutilized

GPUs are in short supply, and training often happens wherever GPUs are available. When training datasets are remote from the GPU clusters, data loading is very slow, wasting GPU cycles. Copying data introduces data engineering complexities.

Data access is slow and models are starved

Models often need to ingest data from multiple sources with different protocols. Thus data is often remote from the models, introducing I/O bottlenecks. Metadata operations on a large number of small files also causes low performance, stalling the training pipeline.

Trusted by the World’s Leading Organizations

“Alluxio has proven to be a valuable solution in addressing the data access challenges of hybrid cloud for Comcast. It has provided us with faster data access, reduced egress costs, and streamlined data management, resulting in more efficient and effective data value creation for the organization.”

“At Uber, we run Alluxio to accelerate all sorts of business-critical analytics queries at a large scale. Alluxio provides consistent performance in our big data processing use cases. As compute-storage separation continues to be the trend along with containerization in big data, we believe a unified layer …”

“With the introduction of Alluxio, we are seeing better performance, increased manageability, and lowered costs. We plan to implement Alluxio as the default cross-region data access in all clusters in the main data lake.”

End-to-End Machine Learning Pipeline Demo

Alluxio’s Senior Solutions Engineer Tarik Bennett walks through a short end-to-end machine learning pipeline demo with Alluxio integrated. See how Alluxio can be provisioned or mounted as a local folder for the PyTorch dataloader, delivering 90%+ GPU utilization and dramatically accelerating data loading times.

1. Data Preparation

2. Setting up the Model

3. Setting up the PyTorch Profiler

4. Model Training