Together launches free Storage Accelerator: ~20x faster warm reads for GPU clusters
togethercompute · x · 2026-11-20
Together AI launched Together Storage Accelerator (TSA), a distributed caching layer included free with every Together GPU cluster.
- Caches data from any S3-compatible object store and Hugging Face Hub on cluster NVMe; one origin fetch per cluster
- Warm reads up to 3.4 GB/s, 20x faster than cross-cloud pulls, with no application code changes
- Aims to eliminate cold-start waits (e.g., 30-minute model downloads) and costly egress (AWS charges up to $0.09/GB) that leave paid GPUs idle
The pitch: your GPUs are only as fast as your data pipelines, and TSA removes the WAN tax.
More from Infra
- QuixiInfer author: tell Claude 'defeat is not an option' to make it write kernels for your GPU — QuixiAI · 2026-11-20
- Volcano launches Agent Cloud: self-healing deploy platform built for coding agents — rohanpaul_ai · 2026-11-20
- Running Krea-2-Turbo on a GTX 1650: 18 minutes per image cut to 30 seconds — A_NASHEX · 2026-11-20
- subapi open-sourced: turn your Claude subscription into a local API endpoint — Last-Transition249 · 2026-11-20
- How BMW governs hundreds of agents across 12,500 developers and 500M lines of code — AWS ML Blog · 2026-11-20
- Analyst conjectures compute entry ticket may get cheaper; 200K GB300s 'guaranteed-viable' by 2028 — teortaxesTex · 2026-11-20