2021-2022 Global Computing Index ReportLEARN MORE
Inspur HF18000G5 high-end all-flash storage is designed for medium/large-sized enterprises and is oriented to structured and unstructured.LEARN MORE
A New Generation High-Density 4U Multi-nodes Server.LEARN MORE
Accelerate your demanding business workloads.LEARN MORE
Learn what's at the heart of new hybrid cloud platforms.Extend the cloud experience across your business in away that‘s open flexible, and hybrid by design.Read the INSPUR paper
Empower an Intelligent Future.LEARN MORE
November 18-21 | Colorado Convention Centre, Denver, USAThinking about attending SC19？Register to get a free pass and join Inspur at the conference for all things HPC.
Computing Intensive DesignStorage Intensive DesignMission Critical SystemsOpen ComputingEdge ComputingGPU Accelerate
CloudArtificial IntelligenceSAP HANAInspur Nutanix HCI SystemsInspur Azure Stack HCI SolutionsData Center Solution
By Arthur Kang, SPEC OSSC Member / OSG ML Chair, Performance Architect, Inspur Information
When it comes to servers, quality is about making the best use of technology. Full command of fundamental technologies allows us to create a superior product, which in turn advances technology further. As of February 2022, Inspur Information has released server test results for a total of 1,209 SPEC (Standard Performance Evaluation Corporation) benchmark tests, in which it broke 292 records. These records are thanks to a world-class R&D team boasting excellent performance optimization capabilities, allowing Inspur to be the top performer in SPEC benchmark tests for new products.
SPEC is a globally renowned third-party application performance testing organization that establishes, maintains, and endorses a set of benchmarks for evaluating server application performance. SPEC metrics have been used as the authoritative reference for both performance and cost of software and hardware platforms for selecting IT systems in key industries such as finance, and telecommunications. Consequently, SPEC performance testing methods, such as SPEC CPU and SPEC Power tests, are widely used when evaluating project bids.
SPEC CPU2017 is a CPU subsystem testing toolkit that includes two sub-projects, Cint for comparing integer performance and Cfp for comparing floating-point performance. It also comes with two test methods—Speed, a single-task mode, and Rate, a multi-threaded mode. SPEC CPU2017 was developed based on real application scenarios for end users, representing common CPU-intensive operations. These scenarios include language compilation, weather forecasting, video encoding, gene sequencing, discrete event simulation, quantum computing, molecular dynamics simulation, ray tracing in physics simulation, fluid mechanics computing, and speech recognition. Inspur NF8480M6 (Intel Xeon Gold 5318H) ranked 1st in the SPEC CPU2017 Integer Rates test with a score of 504, highlighting the powerful high-density integer computing capability of the Inspur M6 server platform.
NF8480M6 also placed 1st in the SPECPower_ssj2008 benchmark test with a score of 11344 ssj/W (number of transactions processed per unit of power as a measure of performance per watt), 26.3% higher than the industry average. In layman’s terms, this means that it will only consumes 73.7% of the power required by comparable products. In a data center with millions of servers, this can result in savings of hundreds of millions of USD in energy costs.
Beyond performance and energy efficiency tests, the Inspur M6 server platform stood out in cloud, data, AI and storage application tests. In SPEC cloud IaaS 2018 performance testing, F5280M6 ranked 1st with a total score of 55. It can ensure the normal operation of up to 39 copies of virtual machines, each deployed within an average of 41 seconds, with relative scalability reaching 85.1% (by the official standards of SPEC, a result above 80% is considered excellent). NF5280M6 also broke the SPEC virt_sc2013 data center virtualization performance world record held for four years with a top score of 4679 from the operation of 263 highly loaded virtual machines on one server.
Credit for the M6 platform's excellent testing results must also be given to its optimized server hardware design and performance:
●High reliability, availability, and serviceability (RAS); hot swap, intelligent discovery/management, and other technologies significantly reduce the risk of unplanned downtime and provide highly efficient, stable, and reliable computing power for analyzing massive amounts of data and processing requests for critical applications.
●Powerful wide heatsinks allow more heat to be dissipated with airflow through inlets and outlets and significantly improving the efficiency of CPU Turbo Boost technology.
●Upgraded chip interlink technology and load balancing topology technology, which distributes loads evenly across CPUs, which can then better support critical enterprise applications.
●T-shaped energy-saving heatsinks enable better heat dissipation, reducing CPU power consumption by nearly 23% and improving server cooling efficiency by 22%.
●Intelligent temperature regulation technology based on the PID algorithm allows fans to adjust their speed to keep the working temperature of components within normal ranges, minimizing fan power consumption and thus reducing power consumption from cooling.
The Inspur team's commitment to exploring more ways to improve product performance is a major contributor to the success of the Inspur M6 platform. With extensive expertise and experience in test optimization, the test team developed feasible test plans based on project cycles and product characteristics, and designed various optimization solutions for benchmark products and mainstream products. They frequently analyzed test results and based on large amounts of data, such as CPU performance, memory throughput, and system energy efficiency, identified areas of potential performance improvement and adjusted optimization strategies in high concurrency, load balancing, software performance and other aspects. They also compiled automated test scripts and modified test scripts to reduce debugging time, improving test optimization efficiency. Leveraging a variety of software and hardware tuning methods, the team has unlocked the ultimate potential of Inspur's M6 platform, making it the reigning champion in SPEC server performance testing.
Inspur AIStation Empowers Efficient GPU Resource Sharing
AIStation is an Inspur-developed AI development platform specifically designed to deal with these issues by offering an easy to set up and refined GPU resource scheduling system.
Inspur joined with Xishuangbanna National Nature Reserve to develop an extensive technology system for the conservation of some 300 Asian elephants in Yunnan, China.
Uncovering the ancient past with Inspur AI and biomolecular archaeology
Inspur teams up with DNA lab to trace the origin of human civilization
Archaeology and AI Unlock the Secrets of Our Ancient History
With the help of today’s intelligent computing, researchers are now more easily able to find out more about our world from critically examining the artifacts of the past.
Inspur Information's Impressive SPEC Benchmark Results Showcase the Benefits of Maximizing Server Performance to Its ful...
By Arthur Kang, SPEC OSSC Member / OSG ML Chair, Performance Architect, Inspur Information
Training Yuan 1.0 – a Massive Chinese Language Model with 245.7 Billion Parameters
The advanced language capabilities of Yuan 1.0 necessitates a high number of parameters, which brings many challenges in model training and deployment. This article focuses on the computing challenges of Yuan 1.0 and the training methods used.
Performance Evaluation of Weather and Climate Prediction Applications on Intel's Ice Lake Processor
The amazing enhancements of Intel's new third-generation Xeon Scalable processors (Ice Lake),
A Deep Analysis on Optimal Single Server Performance of ResNet50 in MLPerf Training Benchmarks
lots of factors can influence the training performance. In this article, we use the ResNet50 model from MLPerf Training v1.0 benchmarks as an example to describe how to improve training speed with hardware and software optimization
OpenRMC Provides Open Rack Management Solution to Help Data Centers Reduce Costs and Boost Efficiency
Integrates hardware and software that helps data centers improve construction efficiency, simplifies operations management, and enhances operational efficiency.
Standardization for Advancing Heterogeneous AI Computing Platforms
New specification provides current system compatibility and a framework for mixed accelerator hardware applications
The Future of Data Centers Is Greener, More Open Computing
As basic IT infrastructure shifts from the on-prem private IT datacenter to the public/hybrid cloud with the growing demand for more computing performance, the world is facing a new challenge
The Evolution of the Rack-scale System in the 5G Era
As the world enters the 5G era, the common impression is that 5G will bring down consumer prices of information services.