Глобальная горячая линия технической поддержки Тел: 1-844-860-0011 1-646-517-4966 E-mail:serversupport@inspur.com Следите Inspur
Home page > Продукты > Облачные вычисления и дата-центры  > HPC > Details on high performance computing

Inspur NF5588M3

Inspur NF5588M3 features the same performance and network as NF5288. It has a higher IO expansion performance and outstanding tower and rack switching ability and therefore is an ideal choice for flexible deployment of high-performance servers. It offers a robust and reliable basic platform for enterprise-level users and help users to acquire hardware and software as well as total-life-cycle aftersales services at an even lower rate of investment.

Functional Features

Functions and features:

Target customers and application fields

Life science: molecular dynamics, gene sorting, protein folding, computing chemistry, etc.

Engineering science: CAD/CAM/CAE, astrophysics, CFD, Mathematics, LabView.

Governmental national defense: weapon, image processing, battlefield simulation.

Medical: MRI, CT, image-added treatment.

Petroleum and petrochemical: earthquake data processing, oil reserve simulation.

Financial: risk analysis, derivative financial commodity simulation, financial molding.

Visual: render farm, game and animation.

Electronic design automation EDA: SPICE, Verilog, 3D EM.

Product advantages

CPU+ GPU isomeric computing architecture

GPU computing unit represents an innovation compared to the conventional singular computing unit. The latest Intel architectural processor core is adopted to collaborate with NVIDIA Tesla/Kepler accelerated computing technology and allows CPU and GPU to work separately. CPU mainly undertakes logic selection, judgment jumping and IO communication duties that it is specifically designed for, while GPU focuses on intensified and highly-parallel computing work, hence allowing the reasonable distribution of computing resources, full release of computing power and increase of computing ability by several to several hundred folds. It is particularly noteworthy that the collaborative computing acceleration architecture is not a denial of the conventional technical route but is complementary to it, as it can provide greater access to the advantages of GPU and allow the user to reduce the time required for problem solving substantially, or increase the scale of problem solving.

Virtual solution with outstanding performances

Compared with the conventional GPU virtual solution, the GRID K1 and K2 cards of Nvidia Kepler architecture are designed to offer richer effects in the virtual environment and they also realize the hardware virtualization of GPU. It also means that multiple users can share a singular GPU, thereby improving user density. Meanwhile, it can provide the real PC performances and compatibility.

Grid card can support the GPU-based virtual solution of Citrix, Microsoft and VMware so that users can make flexible choices from the extensive range of service and solutions.

Large-scale parallel computing processor core

Compared with the multi-core CPU that can only have very few threads working simultaneously, CPU can execute several thousand threads simultaneously and this enables our system to process more data flows, e.g. CPU computing is a special computing task in game and animation rendering, i.e. there is no need to consider the sequence of pixels subject to computing. If there are millions of threads, rendering of all pixels can be realized simultaneously so that all pixels can be computed within the time of one pixel.

High-speed IO switching technology between units

Each device has its special connections. GPU does not need to apply for bandwidth with the bus and it can transmit data at a very high frequency. Compared with conventional PCI bus which can only allow unidirectional transmission within a singular period, PCI-E dual simplex connections can provide higher transmission rate and quality. Their difference is similar to that of semi-duplex and full duplex. PCI-E 3.0 x16 is adopted for Inspur NF5588M3 and the bandwidth reaches 32GB/s.

Latest Intel QPI technology

In order to better utilize the collaborative effects of CPU and GPU, Inspur NF desktop supercomputer features Intel's QPI (Quick Path Interconnect) technology to realize the maximum transmission speed of 8.0GT/s so that communication will be faster and more convenient. QPI bus enables direct interconnection inside multi-core processor. When multiple processors are working together, the processors can transmit data mutually without the need to pass the chip set, hence substantially improving general system performance. With the processor integrated with internal memory controller, PCI-E3.0 graphical interface and even graphical core in Sandybridge architecture, QPI architecture’s advantages will be even more evident.

Excellent programming environment

CUDA™ (Compute Unified Device Architecture) allows GPU to resolve complicated computing problems. It includes CUDA’s instruction set architecture (ISA) and GPU’s internal parallel computing engine. The developer can use C language for programming of CUDA™. C language is the most extensively applied advanced programming language and the programs compiled in this language can run with high performances on a processor that supports CUDA™.

The standard C language for GPU parallel application development.

Standard digital library of Fast Fourier Transform (FFT) and basic linear algebra subprogram (BLAS).

Special CUDA drive, for the fast data transmission calculation between GPU and CPU.

Interoperation between CUDA driver and OpenGL and DirectX graphical driver.

Support Linux 32/64, Windows XP 32/64 and Mac operating system.

Inspur has a robust GPU/MIC application R&D and transplantation team

Leading CPU+GPU/CPU+MIC architecture application optimization team within the domestic market.

Resolve the management and dispatching problem associated with mixed architecture clusters.

Integrated with CPU, GPU and MIC calculation capability.

Extract the methodology to migrate applications to the mixed architecture.

Inspur has accumulated a wealth of  experience in the development of GPU/MIC

1.GPU innovation


Inspur collaborated with an institute of the Chinese Academy of Sciences to jointly develop the GPU-based Blastn algorithm. Compared with mono-core CPU, Blastn algorithm is 35 times better in performance in a single GPU.


Inspur collaborated with several institutes of the Chinese Academy of Sciences to jointly develop the GPU-based ET software ATOM; its performance on Tesla C2050 GPU is 48 times of its performance on a single CPU platform.


Inspur collaborated with a domestic university to jointly develop the GPU-based parallel LBM-LES algorithm. Its performance on a single Tesla C2050 GPU is 55 times the performance on a single CPU platform. Its performance on four Tesla C2070 GPUs is 150 times of the performance on a single CPU platform.

2. MIC innovation


Inspur application service team has developed the MIC-based parallel SIRT algorithm. In the Intel CPU platform, the single-thread running time is 1434s; the 8-core multi-thread OpenMP version running time is 136s; single MIC (KNF) version running time is 55s; the performances are substantially improved. It is one of the five MIC applications displayed in the 2011 SC11 conference.


Inspur application service team has developed the MIC-based parallel BSDE algorithm. In the Intel CPU platform, the single-thread running time is 410s; the 8-core multi-thread OpenMP version running time is 41s; single MIC (KNF) version running time is 21s; the performances are substantially improved.


In the Intel CPU platform, the 8192*8192 network-based single-thread running time is 10324s; the 16-core multi-thread OpenMP version running time is 273s; the single MIC (KNF) version running time is 191s; the performances are substantially improved. It is one of the five MIC applications displayed in the 2012 IDF12 conference. 

Technical Specifications



Intel® Xeon® E5-2600V2 processor seriesup to 150W


Support 512GB DDR3 1600/1333/1066 RECC memory at the maximum; 4-channel reading; support advanced memory error correction, mirroring, hot standby and other advanced features

SATA controller

Integrated with SATA disk controller; optional 8-channel SAS 3Gb and 6Gb disk controller


Support hot-plug disk capacity of 32TB at the maximum

PCI expansion slot

Support 4* Nvidia GPU card or Intel MIC card at the maximum

Integrated I/O port

Integrated with serial port, RJ45 network port, USB2.0 connection port, independent IPMI administration port and VGA interface

Power supply

1+1 backup 1620W 80PLUS Platinum power supply

Operating system

Microsoft Windows Server 2008 SP1 32/64bit

Windows 2003 Enterprise with SP2 32/64bit

Red Hat Enterprise Linux 5U3 32/64bit

SuSE Linux Enterprise Server 10 SP2 32/64bit

SuSE Linux Enterprise Server 11 32/64Bit

The operating system has different supporting abilities in different configurations. Please contact Inspur Group.

Other data

Support Nvidia Tesla K20 (4 units)/K20X (4 units)/K40M (4 units)/K10(4 units); Support Nvidia Grid K1/K2

User Manual

Related Products

Нужна ли помощь.
Ждем Вас