Infrastructure
Teams running training or inference on Together GPU Clusters can likely cut data-read latency and origin fetch costs by enabling the free cache without modifying their code, so it is worth testing on existing workloads.
Together AI has launched Together Storage Accelerator, a distributed caching layer for its Together GPU Clusters. It claims about 20x faster warm reads and one origin fetch per cluster, requires no application changes, and is included at no cost. The feature targets data loading speed for AI training and inference workloads on GPU infrastructure.
Read the full article at Together AI
CoFabrix summarises and comments on this story. The original reporting belongs to Together AI.
Find out where your organization stands -- and what to do about it.