Bringing the interconnectivity of supercomputing to ultra-efficient processors.
An embedded network fabric switch provides the links necessary to interconnect 10's, 100's or even 1000's of EnergyCore SoCs - without the cost, complexity or cabled-mess of traditional rack-based deployments. Similar to a traditional layer-2 network switch, but enhanced with more advanced power and routing features, the EnergyCore Fabric Switch embedded within each SoC delivers:
- High-bandwidth, low-latency links that deliver up to 10Gb of bandwidth with latencies less than 200ns per hop
- Adaptive link speed minimizes energy consumption by auto-negotiating link speeds between EnergyCore server nodes within the same fabric
- Topology agnostic deployments - up to 4096 nodes - allow system designers to build clusters based on requirements for a variety of workloads
Implemented within the EnergyCore ECX-1000 , and ECX-2000, each SoC contains five fabric links that operate between 1-10Gbps per channel, and supports a variety of fabric topologies such as meshes, grids, butterfly trees, and more. The fabric interconnects are built on industry standard XAUI interfaces, and fully supports native IP (TCP/UDP) as well as lower level protocols. With node-to-node latency under 200 nanoseconds, network round trip times are considerably faster compared to a traditional top of rack switch.
Lower costs, increase performance, and reduce energy consumption.
As an integral component within the EnergyCore architecture, the integrated fabric switch:
- Lowers costs by removing the need for external switching equipment and cabling.
- Delivers higher performance with 10Gb interconnects across an entire cluster, at no additional cost.
- Reduces energy consumption by eliminating or significantly reducing the number of external switch ports required.
- Enables ultra-dense system designs by offering an alternative to traditional network infrastructures and removing the need for physical Ethernet cables and ports.
- Provides network flexibility by supporting various topologies focused on performance throughput, availability, or a balance of both.
Maximizing software compatibility by transparently exposing the fabric switch as Ethernet MACs while leveraging standard Linux drivers. Network communication stacks in your existing applications will "just work". Calxeda's Fleet Fabric is integrated on the same chip as the processor, including 10 Gigabit networking to every server at much lower power and costs.