ASWELL AI

The SN5000 Series, specifically designed for AI, is the world's only Ethernet switch portfolio optimized for deep learning workloads. It enables cloud-scale GPU computing with speeds of up to 800Gb/s.

The SN4000 product line, built for cloud-scale networking, combines industry-leading features and groundbreaking performance to support the latest scalable distributed data center applications at speeds of up to 400 Gb/s.

The SN3000 series, ideal for leaf and spine data center network solutions, provides maximum flexibility with port speeds of up to 200 Gb/s and port density that allows for full rack connectivity to all servers at any speed.

The NVIDIA Quantum-2 InfiniBand switch offers 400 Gb/s throughput, in-network computing, smart acceleration engines, flexibility, and a robust architecture, achieving superior performance with lower costs and complexity for high-performance computing (HPC), AI, and hyperscale cloud infrastructure.

The NVIDIA Quantum InfiniBand switch delivers 200 Gb/s throughput, high bandwidth performance, low power consumption, and scalability to reduce capital and operational costs while maximizing return on investment. The NVIDIA Quantum switches optimize data center connectivity through advanced routing and congestion management features.

ConnectX-8 InfiniBand SuperNIC provides 800 Gb/s data throughput along with support for NVIDIA's in-network computing acceleration engine, delivering the performance and powerful feature set needed to drive AI factories and scientific computing workloads at scale.

ConnectX-7 Smart Host Channel Adapter (HCA) offers ultra-low latency, 400 Gb/s throughput, and an innovative NVIDIA in-network computing acceleration engine for enhanced acceleration. ConnectX-7 provides the scalability and feature-rich technology needed for supercomputing, AI, and hyperscale cloud data centers.

The ConnectX-6 Smart Host Channel Adapter (HCA), based on the NVIDIA Quantum InfiniBand architecture, delivers 400 Gb/s throughput and high-performance in-network computing acceleration engines to maximize efficiency across HPC, AI, cloud, hyperscale, and storage platforms.

The NVIDIA BlueField-3 DPU is a 400 Gb/s infrastructure computing platform providing line-rate processing for software-defined networking, storage, and cybersecurity workloads. BlueField-3 combines powerful computing, high-speed networking, and extensive programmability to deliver software-defined hardware acceleration solutions for the most demanding workloads, from AI acceleration to hybrid cloud, high-performance computing, and 5G wireless networks.

The NVIDIA BlueField-2 DPU provides innovative acceleration, security, and efficiency across all hosts. The BlueField-2 data center infrastructure combines the performance of NVIDIA ConnectX®-6 Dx with programmable Arm® cores and hardware offloading for software-defined storage, networking, security, and management workloads. BlueField-2 offers 200 Gb/s network connectivity, providing outstanding performance, security, and reduced total cost of ownership for cloud computing platforms, enabling organizations to effectively build and operate large-scale virtualized and containerized bare-metal infrastructure.

The BlueField-3 SuperNIC is a new level of network accelerator built on the BlueField-3 networking platform, specifically designed to enhance hyperscale AI workloads. Designed for network-intensive large-scale parallel computing, the BlueField-3 SuperNIC provides up to 400 Gb/s RoCE (RDMA over Converged Ethernet) network connectivity between GPU servers, optimizing efficiency for maximum AI workloads. Opening a new era of AI cloud computing, the BlueField-3 SuperNIC supports a secure multi-tenant data center environment with deterministic and isolated performance across workloads and tenants.