SambaRack™

SambaRack is designed for AI inference workloads that require high throughput and low latency when running large models in production. It is used to deploy and serve inference on AI models at scale across enterprise, developer, and data-center environments.

SambaRack is a rack-level integrated system built specifically for inference using RDU processors and its Dataflow architecture, whereas GPU clusters typically assemble general-purpose accelerators across multiple servers. This purpose-built design of SambaRack focuses on improving efficiency and performance for inference on AI models. SambaRack can co-exist alongside GPUs in a data center for different workloads.

Yes. SambaRack combines multiple RDU chips in one rack, typically 16 SN50 RDUs, to provide the compute and memory capacity required to run very large AI models efficiently within a single system.

Yes. SambaRack is designed to integrate into existing air-cooled data center infrastructure with minimal modification, enabling organizations to deploy AI inference on-premises as well as in hosted environments.

SambaRack scales by combining multiple RDU chips within each rack and connecting multiple racks together into larger inference clusters. Large deployments can span multiple racks to deliver higher throughput and support large-scale AI inference services.