当前位置:首页 >> 计算机软件及应用 >>

high-performance-low-cost-brief


Solution Brief Intel? Omni-Path Architecture High Performance Computing (HPC)

Higher Performance at Lower Cost For HPC Fabrics
A fast, scalable fabric is essential to maximize cluster performance, yet traditional InfiniBand* technologies cannot keep pace with growing needs. As a result, organizations are now spending as much as 20 to 40 percent of their infrastructure budget on fabric elements1—money that could potentially be used to scale compute and storage resources.
Intel? Omni-Path Architecture (Intel? OPA), an element of Intel? Scalable System Framework, rises to meet this challenge by transforming the performance, scalability, and cost models of HPC fabrics. This end-to-end fabric solution delivers 100 Gbps port bandwidth, while providing low latency that stays low even at extreme scale. Optimized packet protocols, dynamic traffic shaping, and advanced quality of service (QoS) deliver efficient support for diverse traffic types, with high throughput, high packet integrity, and low, deterministic latency for critical MPI messaging. Even more importantly, the InfiniBand fabric would soak up as much as 40 percent of the total cluster hardware budget, while an Intel OPA fabric would account for only 21 percent.1 The money saved could be used to purchase up to 26 percent more servers, along with the fabric elements and cables required to connect the new servers to the cluster.6 Researchers and engineers would have substantially more compute power to run increasingly complex studies and obtain faster results.

Reduce Infrastructure Requirements By Up to 50 Percent2
Intel developed a new 48-port switch silicon for Intel OPA, which can reduce the number of switches by as much as 50 percent in a typical fat tree configuration.2 This new silicon powers a family of 24- and 48-port edge switches and 192- and 768-port directorclass switches that can be used to support the full range of HPC requirements. With the Intel OPA 48-port leaf cards, a single 5-hop fabric can support up to 27,648 nodes, which is 2.3 times more than can be supported using current InfiniBand switch designs (an InfiniBand fabric would require 7-hops to go beyond 11,664 nodes).3

Intel? Omni-Path Architecture versus InfiniBand* EDR
Up to 26 percent more compute power within the same cluster hardware budget6
EDR 100 Gb/s INTEL? OPA 100 Gb/s UP TO 26% MORE COMPUTE POWER6

FIXED CLUSTER HARDWARE BUDGET

COMPUTE

SERVERS

750

SERVERS

948

Get More Compute Power Within the Same HPC Budget
An Intel OPA fabric not only scales more easily than current InfiniBand solutions, but also helps to reduce cost, complexity, and power consumption. An Intel OPA fabric for a mainstream HPC cluster with 700 to 750 nodes would require up to 79 percent less rack space than a comparable fabric based on InfiniBand, with two fewer cabinets, which would save 12 square feet of floor space.4 The Intel OPA fabric would also reduce power and cooling costs by as much as 60 percent over a 3-year period.5

FABRIC

40%

24%

Same Hardware Budget, High Cluster Performance.
With Intel? Omni-Path Architecture, organizations can spend more on compute resources and less on fabric elements, resulting in higher overall cluster performance within the same budget.6

It’s more than just a 33% increase in port count over a 36 Radix chip
InfiniBand* EDR (36-port Switch Chip)

Intel? Omni-Path Fabric’s 48 Radix Chip

FIVE-hop Fat Tree

Intel? Omni-Path Architecture (48-port Switch Chip)

THREE-hop Fat Tree

Two (2) 648-port Director Switches

768 nodes

One (1) 768-port Director Switches

(43) 36-port Edge Switches 768 nodes

% Reduction (43) 36-port 1,542 99u (2+ racks) ~680ns (5 hops) Edge Switches Cables Rack Space7 Switch Latency
8

None required 768 20u (<1/2 rack) 300 – 330ns
9

100% 50% 79% 51–55%

Learn More
Intel OPA can help you achieve higher throughput and lower latency for your HPC fabric, while enabling higher overall cluster performance within the same budget.6 For more information, read the Intel white paper: “Transforming the Economics of HPC Fabrics with Intel? Omni-Path Architecture,” and visit www.intel.com/hpcfabrics
1

Internal analysis based on a 256-node to 2048-node clusters configured with Mellanox FDR and EDR InfiniBand products. Mellanox component pricing from www.kernelsoftware.com Prices as of November  3, 2015. Compute node pricing based on Dell PowerEdge R730 server from www.dell.com. Prices as of May 26, 2015. Intel? OPA (x8) utilizes a 2-1 over-subscribed Fabric. Intel? OPA pricing based on estimated reseller pricing using Intel MSRP pricing on ark.intel.com as of November 2015. Reduction in up to ? fewer switches claim based on a 1024-node full bisectional bandwidth (FBB) Fat-Tree configuration, using a 48-port switch for Intel Omni-Path cluster and 36-port switch ASIC for either  Mellanox or Intel? True Scale clusters. Based on 27,648 nodes for a cluster configured with the Intel Omni-Path Architecture using 48-port switch ASICs, as compared with a 36-port switch chip that can support up to 11,664 nodes.  A ssumes 750-node cluster, and number of switch chips required is based on a full bisectional bandwidth (FBB) Fat-Tree configuration. Intel? OPA uses one fully-populated 768-port director switch, and Mellanox  EDR solution uses a combination of director switches and edge switches. Mellanox power data based on Mellanox CS7500 Director Switch, Mellanox SB7700/SB7790 Edge switch, and Mellanox ConnectX-4 VPI adapter card installation documentation posted on www.mellanox.com as of November 1, 2015. Intel OPA power data based on product briefs posted on www.intel.com as of November 16, 2015. All Mellanox power and rackspace information is based on Mellanox installation documentation on www.mellanox.com as of November 3, 2015. Cabinet floor space assumes a cabinet that is 24” wide by 37” deep. A ssumes a 750-node cluster, and number of switch chips required is based on a full bisectional bandwidth (FBB) Fat-Tree configuration. Intel? OPA uses one fully-populated 768-port director switch, and Mellanox  EDR solution uses a combination of 648-port director switches and 36-port edge switches. Mellanox component pricing from www.kernelsoftware.com, with prices as of November 3, 2015. Compute node pricing based on Dell PowerEdge R730 server from www.dell.com, with prices as of May 26, 2015. Intel? OPA pricing based on estimated reseller pricing based on projected Intel MSRP pricing at time of launch. All amounts in US dollars. R ack space based on Mellanox CS7500 Director Switch (28U, 648 ports) and Mellanox SB7700/SB7790 Edge switch (1U, 36 ports) product briefs posted on www.mellanox.com as of November 1, 2015. Intel? OPA  Edge Switch 100 Series and Intel? OPA Director Class Switch 100 Series product briefs posted on www.intel.com, November 2015. L atency data based on Mellanox CS7500 Director Switch (sub-500ns) and Mellanox SB7700/SB7790 Edge switch (90ns) product briefs posted on www.mellanox.com as of November 1, 2015. Intel? OPA Edge  Switch 100 Series and Intel? OPA Director Class Switch 100 Series product briefs posted on www.intel.com, November 2015.

2

3 4 5

All Mellanox power and rackspace information is based on Mellanox installation documentation on www.mellanox.com as of November 3, 2015. Cabinet floor space assumes a cabinet that is 24" wide by 37" deep.

6

7

8

9

 Intel measured data that was calculated from difference between back to back osu_latency test and osu_latency test through one switch hop. All tests performed using Intel ? Xeon? E5-2697v3. Pre-production Intel Corporation Device 24f0 – Series 100 HFI ASIC, Series 100 Edge Switch – 48 port.

Software and workloads used in performance tests may have been optimized for performance only on Intel? microprocessors. Performance tests, such as SYSmark* and MobileMark,* are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more information go to http://www.intel.com/performance This device has not been authorized as required by the rules of the Federal Communications Commission. This device is not, and may not be, offered for sale or lease, or sold or leased, until authorization is obtained. All information provided here is subject to change without notice. Contact your Intel representative to obtain the latest Intel product specifications and roadmaps. Intel does not control or audit third-party benchmark data or the web sites referenced in this document. You should visit the referenced web site and confirm whether referenced data are accurate. Cost reduction scenarios described are intended as examples of how a given Intel- based product, in the specified circumstances and configurations, may affect future costs and provide cost savings. Circumstances will vary. Intel does not guarantee any costs or cost reduction. Intel’s compilers may or may not optimize to the same degree for non-Intel microprocessors for optimizations that are not unique to Intel? microprocessors. These optimizations include SSE2, SSE3, and SSSE3 instruction sets and other optimizations. Intel does not guarantee the availability, functionality, or effectiveness of any optimization on microprocessors not manufactured by Intel. Microprocessor-dependent optimizations in this product are intended for use with Intel microprocessors. Certain optimizations not specific to Intel? microarchitecture are reserved for Intel microprocessors. Please refer to the applicable product User and Reference Guides for more information regarding the specific instruction sets covered by this notice. No license (express or implied, by estoppel or otherwise) to any intellectual property rights is granted by this document. The products described may contain design defects or errors known as errata which may cause the product to deviate from published specifications. Current characterized errata are available on request. ? 2015 Intel Corporation. All rights reserved. Intel, the Intel logo, and Xeon are trademarks of Intel Corporation in the U.S. and other countries. * Other names and brands may be claimed as the property of others. Printed in USA 1115/EG/HB/ PDF 332939-001US


相关文章:
low cost
Low-cost 8页 免费 Low-cost 暂无评价 10页 免费 A Low-cost CAPTCHA 12页 免费 Low Cost Driver 1页 免费 Low cost power supply VI... 9页 2财富值...
单片机课程设计--- 基于51单片机的简约电子钟_图文
关键词: STC11F02E;单片机;电子时钟; I Abstract SCM since its introduction in the 1970s, with its high performance and low cost, the attention and ...
...B、 C和D)中。选出最佳选项. Low-Cost Gifts for Mo...
Low-Cost Gifts for Mother's Day Gift No. I Offer to be your mother's health friend. Promise to be there for any and all doctor's visits...
Clean operating room purifying air conditioner cont...
a high level of expertise, easy to operate, simple structure, low cost. ...Brief Description [0008] Figure 1 is a practical and easy operation of ...
里斯本Low Cost Apartments(FLH - Beato (Low Cost Apa...
里斯本 Low Cost Apartments(FLH - Beato (Low Cost Apartments)) 里斯本 Low Cost Apartments 位于里斯本,里斯本 Low Cost Apartments 是您来里斯本的最佳 落脚点...
06外文翻译
brief explanation of the parameters of cost ...Special-performance buildings, such as laboratories ...For low-rise buildings, one can use stock wide...
walmart:the high cost of low price
walmart:the high cost of low price_法语学习_外语学习_教育专区。影片观后感the high cost of low price Wal-Mart -The High Cost of Low Price I always ...
毕业设计(论文)-基于单片机的酒精浓度测试仪_图文
A/D conversion and processing of signal detection using macro crystal, high performance and low cost microcontroller STC12C5A16AD, finally through the ...
电气外文文献翻译---一种低成本智能式电容位置传感器
I. INTRODUCTION This paper describes a novel high-performance, low-cost, capacitive displacement measuring system featuring: 1 mm measuring range, 1 μm ...
巴塞罗那廉价酒店(Low Cost Barcelona)
巴塞罗那廉价酒店(Low Cost Barcelona) 巴塞罗那廉价酒店位于巴塞罗那,是家 2 星级酒店,巴塞罗那理想的下榻之处,您的旅途也 将因此而与一份温暖、一份感动不期而遇...
更多相关标签: