When Intel showcased its Intel® Scalable System Framework (Intel® SFF) at SC15, it was the fabric that took center stage.
Intel SSF is a flexible blueprint for developing balanced, high performance, efficient HPC systems ranging from the smallest clusters to the largest supercomputers. It’s a cornerstone for what is being heralded as the “HPC everywhere” era.
The framework’s architectural approach will help simplify the use of such essential HPC building blocks as processing, memory, software and fabric technologies to create systems ranging from those designed for traditional HPC to the most demanding Big Data analytic applications. And it’s the fabric – the newly introduced Intel® Omni-Path Architecture (Intel® OPA) – that is a key element of the Intel Scalable System Framework.
“Intel OPA touches every element in the system,” says Joe Yaworski, Intel Director of Fabric Marketing for the Enterprise & High Performance Computing Platform Group. “In HPC systems that can scale from hundreds of nodes to hundreds of thousands, the robustness of the fabric will determine performance.”
He points out that Intel OPA represents an evolutionary approach to fabric with some revolutionary enhancements. “Intel OPA incorporates the best features of existing proprietary and InfiniBand* fabric technologies and introduces significant new features and capabilities to improve overall HPC performance, reliability and QoS. In addition, our strategy is to build upon the Open Fabric Alliance (OFA) and OFA Enterprise Distribution to ensure compatibility and stability with existing applications and systems. These approaches significantly reduce the risk of bringing new technology to the marketplace.”
Among some of the evolutionary/revolutionary features incorporated into Intel OPA are:
- CPU/fabric integration – Intel OPA will be integrated with Intel® Xeon Phi™ processors code name Knights Landing—now in pre-production phase—and will offer two ports with a total bi-directional bandwidth of 50GB. CPU-Fabric integration offers a path toward improved performance, reduced power and costs on the node level, along with increasing computing density and reliability for large-scale HPC deployments.
- Optimized host implementation – Intel OPA provides extremely high MPI message rate performance – and MPI is the mainstay of HPC communications.
- Enhanced fabric architecture – The fabric is tuned for high compute throughput. Each of the 48 switching ports on the switching ASIC can handle 195M messages/sec.
- Cost effectiveness – Intel OPA’s 48-port switch helps enable up to 26 percent more servers than InfiniBand* EDR within the same budget and up to 60 percent lower power consumption for a more efficient switch and system infrastructure.
These are just a few of the enhancements found in Intel OPA with an emphasis on QoS and reliability coupled with peak performance and lower total cost of ownership.
Information released by Intel at SC15 indicates that the growing acceptance of Intel Scalable System Framework underscores the importance of the Intel OPA fabric to the framework. Already Colfax, Cray, Dell, Fujitsu Systems Europe, HPE, Inspur, Lenovo, Penguin Computing, SGI, Sugon, and Supermicro have announced plans to launch systems based on Intel SSF with its Intel OPF fabric early next year.
Intel OPA is currently being used at several large sites, including the Texas Advanced Computing Center and the Pittsburgh Supercomputer Center. In addition, Colfax, Cray, Dell, Fujitsu Systems Europe, Hitachi, Huawei, HPE, Inspur, Lenovo, NEC, SGI, Sugon, Supermicro and other system vendors are announcing Intel OPA-based switches and server platforms, with volume shipments ramping in the first quarter of next year.
As one of the primary building blocks of Intel SSF, Intel OPA is taking its place as a key element in the design of HPC systems ranging from powerful desktop workstations and small HPC clusters to high-end systems that are on the path to exascale.