Inspur launches innovative i48 server optimized for AI, HPC and NFV
On June 26, Germany, Inspur launched the innovative i48 server at the International Supercomputing Conference 2018 (ISC18). Optimized for data-center application scenarios, the server is a high-density computing module that supports the IT infrastructure required by AI, HPC and NFV. It can complete consistent, high-density and rapid deployment of an entire system as well as flexibly configure compute and storage nodes according to application data capacity and processing performance requirements.
i48 features innovative design and advantages of both servers and rack servers. It supports 8 high-density compute nodes, 4 computing-storage balanced nodes and 2 large-capacity storage nodes in a 4U chassis. Compared with cluster deployment of traditional rack servers, i48 increases deployment efficiency by 800% and lowers energy consumption by 18% through a unified power supply and thermal dissipation system. With micro rack design and optimized components, i48 saves 8% and 15% of costs for single node compared with blade servers and rack servers respectively.
i48 is equipped with a consistent, high-density and large-scale computing system, which helps tackle performance challenges of massive scientific data processing and AI data training. i48 single rack supports 80 Intel® Xeon® Scalable processor-based high-performance compute nodes, and each node has the performance of over 3.5Tflops, so the overall performance of single rack exceeds280TFlops; in this way,4 racks are able to complete HPC cluster construction of about 1 PFlop. On high-speed interconnection, i48 supports direct connection of 100G high speed network of all nodes to core switch with low-latency, or two 48-port 100G TOR switches to form a large-scale system via cascade in single rack. With the front/back maintainability of a modern data center, i48 greatly improves the efficiency of large-scale system operation and maintenance.
The agility of i48 allows it to flexibly configure compute and storage nodes according to application data capacity and processing performance requirements. Applying i48 to 3rd generation of gene sequencing for example, an i48 single rack configured with 384TB data storage and more than 200 processor cores shortens time required for genome-wide assembly of humans and similar mammals to 2 days, with data processing capacity 4 times higher than the industry average. The i48’s high-density blade compute nodes allow configuration of Inspur’s F10A FPGA accelerator, boosting GATK’s performance by 340%.
With numerous data channels, i48 is also set to play a leading role in enabling the expansion of emerging businesses. While 5G technology is set to bring more opportunities for telecom operators, allowing engagement in business model innovations including IOT, IOV, edge computing and AI, the staggering amount data communication and low-latency featured by 5G will require operators to virtualize their networks. NFV will be the infrastructure of emerging businesses in the 5G era. With in-depth optimization for NFV application, each node of i48 supports 4 PCIe and 1 OCP/PHY network interface cards, providing up to 16 10-gigabit network communication interfaces and supporting NUMA Blanced optimization. It can improve communication performance by up to 20% without interfering with CPU utilization.
In addition, each high-density compute node of i48 supports 4 NVMe hot-swap HDDs with capacity up to 32TB. Fit for high-speed Burst Buffer data caching, i48 provides a fast storage tiering solution that bursts I/O performance along with the cluster file system, improving overall performance by a factor of 10 and supporting the fast I/O caching file systems of large-scale HPC clusters.
“i48 features the advantages of both a blade server and rack server,” said Peter Peng, vice president of Inspur Group. “It has the quick deployment, flexible extension, ultra-large storage capacity and high cost-effectiveness of a rack server, along with the high density, easy management and high performance per watt of a blade server. It provides a new data center module and is compatible with Intel Xeon Scalable processors and Intel® OnmiPath interconnection technology, allowing global data center users to deploy high-performance IT infrastructure in a more flexible and faster way, to better tackle the challenges of massive data storage and processing in business scenarios including AI, HPC and NFV.”
Please click the link for full report