SN40L Overview

Introduction

SambaStack is a purpose-built hardware and software platform optimized for AI inference, powered by SambaNova’s advanced Reconfigurable Dataflow Unit (RDU) processors. At the core of this infrastructure is a SambaNova SN40L cluster, composed of six SambaRacks—- each housing 16 SN40L RDU systems. This dedicated setup delivers high-throughput, low-latency performance for machine learning workloads. Models running on the cluster are exposed through OpenAI-compatible API endpoints, with each endpoint capable of hosting multiple independently accessible models.

Below are some of the links to SambaNova documentation:

• SambaRack information
• Reconfigurable Dataflow Units (RDUs) — purpose-built for AI
• Accelerated Computing with a Reconfigurable Dataflow Architecture (PDF)

Using Metis

Models running on this cluster are served through the ALCF Inference Service. See Inference Endpoints for the endpoint URL, available models, and usage examples.