Machine Learning

Pure Storage: Modern Machine Learning with FlashBlade 

Data is the fuel driving rapid innovation powered by artifi cial intelligence. Enterprises need modern data platform purpose- built for machine learning, accelerating insight while simplifying complex data pipelines in analytics.

Legacy, retrofit storage not keeping up

Artifi cial intelligence (AI) revolution today is powered by modern technologies, like deep learning algorithms and GPUs, that are massively parallel. However, legacy storage systems were largely built on decades-old building blocks, designed in the serial era, and the performance gap between compute and storage continues to grow.

FLASHBLADE™ Scale out storage reflects the soul of AI

FlashBlade is the industry’s first scale-out storage purpose-built for modern data analytics like machine learning.

Fast, big and simple, FlashBlade is massively parallel at its core, o ering unprecedented performance and simplicity for data scientists. Purity for FlashBlade operating software is built on a massively distributed key-value pair architecture. Like a GPU-accelerated server, FlashBlade is architected to accelerate parallel workloads, delivering the performance of 10 racks of disk in a 4U size.


Now you can:

Accelerate training by keeping GPU servers busy with data.

Simplify workfl ow with modern, “tuned for everything” storage.

Replace complex, fragile infrastructure of disks with Peta-scale capacity in 4U.

“Tuned For Everything” Architecture

Unstructured data has three qualities: volume is growing rapidly, it is random in its structure, and fi le/object size can vary from small to large. It’s no surprise that AI workloads refl ect these qualities. Bigger training datasets means higher accuracy. Access patterns can be random. And fi les/objects can be small images and text to large streaming videos.

FLASHBLADE is engineered for unstructured data built to deliver the  highest performance no matter what your data needs are 

  • • Delivers linear scaling performance that grows with your data, from TBs to PBs, to thousands of clients.
  • • Offers predictable, ultra-fast performance for any access patterns, random or sequential.
  • • Designed to deliver maximum performance, from small, meta-data heavy to large streaming files.


A leading global, web-scale company has built a large AI supercomputer, powerful enough to be among the world's fastest systems. This impressive system is built with racks of NVIDIA DGX-1 servers and FlashBlade systems, designed to be well balanced for deep learning training. FlashBlade is engineered to deliver maximum performance in any deep learning training environment, reducing data processing time from days to hours.