Inspur Blog

NF5280M5: raising the bar for server versatility and performance

Enterprises are facing the problem of exponential data growth and diversification of business scenarios, resulting in more and more complex IT infrastructure in data centers and expanding on-demand hardware infrastructure. The need for rapid response is becoming more and more urgent.

Based on the newest Intel Xeon Scalable computing platform, Inspur’s M5 series proposes the ultimate design concept to optimize the cloud data center for deployment characteristics like form factor, density, energy consumption and manageability. The strategy to implement this design principle is to effectively combine platform and modularization. The NF5280M5 two-socket flagship server was designed to embody this principle.

SPECjbb2015 results show extreme performance optimization

The NF5280M5 is a 2U two-socket server, which is currently the largest mainstream category of servers. It supports 2 of Intel’s latest Xeon Scalable processors, with up to 28 cores and 56 threads in each CPU. The CPU uses two 10.4 GT/s UPI interconnect links to support M.2 all flash, achieve ultra-high storage throughput, reduce I/O latency to microseconds, and deliver ultimate application performance in database, ERP, big data analysis and other scenarios. According to the measured data, the NF5280M5 delivers up to 1.3 times the performance of the previous generation.

The latest SPECjbb2015 test results released by the world’s two major performance evaluation organizations show that NF5280M5 achieved the max-jOPS of 181,507. This is the index for a system’s maximum processing performance, indicating the system can complete up to 181,507 Java transaction processes without Java failure.

SPECjbb2015 is the definitive benchmark for measuring the performance of server-side Java applications, since Java is still the lingua franca of enterprise applications. Excellent Java performance means excellent application performance in every enterprise application scenario. In addition to SPEC jbb, the NF5280M5 performed equally well in other SPEC benchmarks, breaking 26 world records in SPEC CPU, SPEC Power and other tests.

Versatile, configurable and application-optimized

While the typical 2U two-socket server delivers default scalability and applicability, it is difficult to make further breakthroughs in the form factor. By contrast, the flagship model of Inspur’s two-socket rackmount products can be flexibly configured through component upgrades.

NF5280M5 uses Inspur’s unique spatial layering technology to break boundaries in the traditional server design and divide the system into two parts: the basic component and the variable component. In the storage module, one set of standards and three combinations are built. Implementing an all-flash configuration enables 31 regular hard drive expansions and the maximum support for 20 large-disk expansions is 67% higher than that of the previous generation. In the IO part, a standard OCP module is introduced, which enables the expansion of four GPUs and the integration of storage modules and IO modules through flexible configuration, which greatly improves the diversity of system application scenarios. In addition, through the system resource allocation technology, the standard components are deeply integrated and flexibly deployed, which can meet more than 30 configuration requirements of major applications that include general computing, large-capacity storage, all-flash, IO-intensive, and heterogeneous acceleration scenarios.

With its unique IO expansion module, the NF5280M5 can support up to four full-width GPUs or eight half-width GPUs, making it an ultra-high-density AI training solution with the highest specs in the industry to support more AI applications. Currently it is widely used for deep learning, offline training, and speech recognition at enterprises like Baidu, NetEase, and Iflytek.

High availability, higher reliability

The NF5280M5 uses different levels of high-availability technologies, such as a robust hard disk RAID, hot spare memory, mirroring, error correction, and redundant hot-swap design of power supply fans. It has complete hardware and software error tolerance. Take the intelligent memory management technology as an example. Before the system starts, the memory is diagnosed, and any error or hidden unit is removed from the available area. The memory is monitored, and errors or hidden dangers handled, in real time during system operation. Memory error detection is precise down to the Bit level, and failures caused by memory error is reduced by more than 90%.

This product adopts more security design by integrating more than 80 security technologies at different levels, integrating trusted computing modules, and supporting domestic commercial cryptographic algorithms. So far, third-party organizations such as the Information Security Laboratory of China Electric Power Research Institute and other industry users have conducted more than 100 safety tests on the server, and the test results show full compliance.

Industrial opportunities under smart computing

Inspur NF5280M5 perfectly demonstrates high efficiency, reliability, intelligence and high expansion characteristics in a limited space, adapting to the needs of various application scenarios in the intelligent age, especially for telecommunications, finance, large enterprises and other users who have stringent server requirements. As Inspur gains speed and size in the global server market in recent years – becoming one of the world’s 3 top-selling server vendors — Inspur can deliver application-specific, smart computing products and services to more customers seeking to seize new opportunities in the next-generation IT era.

Leave a Reply

Your email address will not be published. Required fields are marked *