Mellanox provides the world's first smart switch, enabling in-network computing through the Co-Design SHARPTM technology. SB7890 has the highest fabric performance available in the market with up to 7Tb/s of non-blocking bandwidth with 90ns port-to-port latency. Scaling-Out Data Centers with Extended Data Rate (EDR) InfiniBand Faster servers based on PCIe 3.0, combined with high-performance storage and applications that use increasingly complex computations, are causing data bandwidth requirements to spiral upward. As servers are deployed with next generation proces- sors, High-Performance Computing (HPC) environ- ments and Enterprise Data Centers (EDC) will need every last bit of bandwidth delivered with Mella- nox's next generation of EDR InfiniBand high-speed smart switches. World's First Smart Switch Built with Mellanox's latest Switch-IB TM 2 InfiniBand switch device, EDR uses efficient 64/66 encoding while increasing the per lane signaling rate to 25Gb/s. The SB7890 provides up to thirty-six 100Gb/s full bi-directional bandwidth per port. These stand-alone switches are an ideal choice for top-of-rack leaf connectivity or for building small to extremely large sized clusters. SB7890 is the world's first smart network switch, designed to enable in-network computing through the Co-Design Scalable Hierarchical Aggrega- tion and Reduction Protocol (SHARP) technology. The Co-Design architecture enables the usage of all active data center devices to accelerate the communications frameworks, resulting in order of magnitude applications performance improve- ments. SB7890 enables efficient computing with features such as static routing, adaptive routing, conges- tion control and enhanced VL mapping to enable modern topologies (SlimFly, Dragonfly+, 6D Torus). These ensure the maximum effective fabric band- width by eliminating congestion hot spots. The SB7890 switch has best-in-class design to support low power consumption. ATIS weighted power consump- tion is 122W for a fully-populated switch. Power is further reduced if not all ports are used or if partially utilized. Collective Communication Acceleration Collective communication is a term used to describe communication patterns in which all members of a group of communication end- points participate. Collective communications are commonly used in HPC communication protocols such as MPI and SHMEM (OpenSHMEM). Collective operations have implications on overall application performance and scale. SB7890 introduces the Co-design SHARP technology, which enables the switch to manage collective communications using embedded hardware. Switch-IB 2 improves the performance of selected collective operations by processing the data as it traverses the network, eliminating the need to send data multiple times between end-points. This decreases the amount of data traversing the network and additional benefit of freeing up CPU resources for computation rather than using them to process communication. The network portion of the reduction operation on a fully-populated three-level fat-tree can be completed in less than three microseconds. Building Efficient Clusters The SB7890 is the industry's most cost-effective building block for deploying high performance clus- ters and data centers. Whether looking at price-to- performance or energy-to-performance, the SB7890 offers superior performance, low power and scale reducing capital and operating expenses providing the best return-on-investment.