Understand this story
On November 19, Together AI launched Together Storage Accelerator (TSA), a distributed caching layer for Together GPU Clusters that connects to object storage and the Hugging Face Hub without requiring changes to existing applications and is provided free of charge with every cluster. Tests on H100 nodes published by Together AI showed warm-cache read speeds of 3.0–3.4 GB/s for 2 GiB objects, approximately 20 times faster than cross-cloud sources; when 200 nodes retrieved the same checkpoint, the number of source requests could be reduced from 200 to one. The service can now be enabled on the cluster settings page, though actual results may vary depending on the data source, object size, and network path.
AI-generated from public reports
Latest development · Report
Together AI Launches Together Storage Accelerator, Bringing…Developments
1 confirmed developments
- Source:Together AI Blog
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 19
Together AI BlogFirst-party
Together AI Launches Together Storage Accelerator, Bringing…
Together AI has launched Together Storage Accelerator, providing free distributed caching for every Together GPU Cluster and enabling access to object storage and the Hugging Face Hub without requiring changes to existing applications. Tests on H100 nodes showed warm-cache read speeds of 3.0–3.4 GB/s for 2 GiB objects, approximately 20× faster than cross-cloud sources. When 200 nodes retrieve the same checkpoint, the number of origin requests can drop from 200 to one. The service can now be enabled on the cluster settings page. Actual results may vary depending on the data source, object size, and network path.
AI Trends