Together AI launches a data caching accelerator
The story so far
Together AI has launched the Together Storage Accelerator (TSA), providing a distributed caching layer for GPU clusters, accelerating AI data pipelines and reducing data egress costs. TSA runs on each GPU node in the cluster, caching data from object storage and Hugging Face Hub, making subsequent read speeds about 20 times faster, reaching 3.4 GB/s/node, and can be used without modifying application code. TSA is now free and included in every Together GPU cluster, allowing users to use it without changing existing application code or data loaders, supporting existing boto3 code, transformers calls, and CLI workflows, and handles user-configured credential authentication.
AI-written from the reports · Updated 1 hour ago
Coverage timeline
Follow the coverage to see every side of the story.
- Together AI BlogIntroducing Together Storage Accelerator: fast, free data caching for your GPU cluster
Together AI has released Together Storage Accelerator (TSA), providing a distributed caching layer for GPU clusters to accelerate AI data pipelines and reduce data egress costs. TSA runs on each GPU node in the cluster, caching data from object storage and Hugging Face Hub, resulting in a 20x speed increase for subsequent reads, reaching 3.4 GB/s/node, and can be used without modifying application code.
Heat over time
Not enough data yet to draw a trend.