AMD 와 고성능 컴퓨팅 (HPC)
Take High Performance Computing (HPC) to new heights
Today’s HPC architectures are designed to solve complex scientific and engineering problems across an ever expanding number of HPC and technical workloads.
From desk side and departmental clusters to the largest HPC installations like ORNL’s (Oak Ridge National Laboratory) Jaguar supercomputer running at over 1.7 petaflops with over 2.2 Million processor cores, x86 based architectures are being used to solve a growing number of problems with a growing variety of workloads. With this evolution, it is important for these x86 based architectures to be well balanced and to perform across a board range of HPC codes.
These solutions need to take into consideration key factors that affect performance in HPC solutions including raw GFLOP performance, cache sizes, instruction set support, memory bandwidth and latency issues, network performance, and the growing issue of power efficiency, among others.
With AMD’s continued advancements in multi-core x86 architectural designs, AMD has enabled HPC customers and partners in providing balanced computing solutions, which is helping them to meet these expanding computing requirements.
AMD’s HPC product portfolio consists of multi-core x86 processor platforms and ATI Stream technologies.
On the multi-core x86 side, our AMD Opteron™ 6000 Series platforms with AMD Opteron 6100 Series processors are designed for performance 2-way and 4-way HPC servers; and now our AMD Opteron 4000 Series platforms with AMD Opteron 4100 series processors help expand our HPC portfolio by offering 1 and 2-way server products for HPC Cloud Computing and Web 2.0 deployments, where the need for density, power efficiency and cost optimizations are critical. These products offer outstanding performance-per-watt with the world’s lowest power per core server processor1, setting the foundation for HPC cloud workloads.
Learn more about AMD Opteron 4000 Series platforms and Cloud Computing solutions.
Forming the backbone of AMD’s latest HPC product platform offerings are key architectural features including Direct Connect Architecture 2.0 with HyperTransport™ Technology 3.0 (HT3) for fast communications between CPUs, AMD-P 2.0 helping to drive more power efficient HPC solutions and AMD Virtualization™ (AMDV™) technology, which helps to enhance and accelerate software based virtualization. These features can help you maximize performance and minimize cost.
AMD Opteron™ 6100 Series processors - high performance, low watts
Designed specifically for dense, high-performance environments, AMD Opteron™ processors feature a host of features designed to help increase performance while minimizing power requirements.
Diagram of the AMD Opteron™ 6100 Series processor Features
The table below summarizes some of the major technologies found in the latest generation AMD Opteron™ 6100 Series processor. These technologies are especially important in the increasingly dense environments commonly deployed by HPC users. In these environments, performance/watt is all-important to help reduce facilities costs while providing the performance HPC users require.
Options of 8 & 12 cores per socket
Up to double the cores (over prior generation) for scalable systems with up to 48 cores providing performance and scalability.
Additional cores are important for multi-threaded workloads-often found in HPC application environments.
L2 Cache: 512K/core
L3 Cache: 12MB
Double the L3 cache to accelerate data access and drive optimal performance
Up to four memory channels supporting R/U DDR3 and LV-DDR3 Up to 1333MHz
Allows for system designs with up to 50% more DIMMS and up to 60% faster memory performance(2). Increases memory bandwidth and performance for memory bound HPC workloads.
Planned HE/Std/SE power options to match your HPC workload and power requirements.
HyperTransport™ Technology Assist (HT Assist)
HT Assist increases 4 way server scalability by reducing cache probe traffic between processors by improving memory bandwidth and latency performance. HT Assist can help increase performance for cache-sensitive sections of code, such as those found in compute intensive applications.
Four HyperTransport™ 3.1 Technology (HT3) links (between CPUs) running at up to 25.6 GB/s per link at 6.4GT/s
Nearly twice the I/O bandwidth(3) with four, faster HT3 links and an integrated CPU/chipset platform.
HT3 increases interconnect rates up to a maximum 6.4GT/s, helping improve overall system balance and scalability.
AMD-P and AMD-P 2.0 suite of power management features:
New features include enhanced APML (in APML enabled platforms), AMD CoolSpeed technology, C1E, LV-DDR3 support, and Link Width PowerCap
This collection of power management innovations helps reduce energy use in HPC solutions.
Link Width PowerCap can increase sys-tem level perf/watt by up to 5% during load(4); C1E can reduce system level power consumption by up to 35% during idle(5)
|AMD Virtualization™ (AMDV™) Technology features include Extended Migration and I/O Virtualization
AMDV™ enhances and accelerates software based virtualization to help run more virtual machines with less overhead
The new AMD Opteron™ 6000 Series platforms and AMD Opteron 6100 Series processors were designed with HPC customers in mind. Customers buying servers and workstations based on these products will have access to the latest platform technologies with the added benefit of future same-socket upgradeability to AMD’s next generation AMD Opteron products codenamed: “Bulldozer” extending the life of their infrastructure investments.
More information on HPC system designs taking advantage of AMD’s new high performance products, please visit:
ATI Stream Technologies
In addition to AMD’s latest HPC processor products, AMD offers some of the world’s highest performing GPGPU technologies which complement our CPU offerings. AMD’s ATI Stream Technologies for HPC are a set of advanced hardware and software technologies that enable AMD GPUs to work in concert with our AMD Opteron processors to help accelerate workloads beyond traditional graphics and video processing. AMD’s stream technologies enable balanced platforms to run computationally-intensive tasks more efficiently. To learn more about our ATI Stream Technologies, please visit www.amd.com/stream.
Another key part to AMD’s HPC solutions is the Tools and Compilers supporting those solutions. For details on these supporting tools, please visit our developer pages found at www.amd.com/developer.
Take High Performance Computing (HPC) to new heights
With award-winning high-performance computing (HPC) solutions and continued advancements in multi-core x86 server processor architecture, AMD delivers powerful, well-balanced computing solutions to HPC customers and partners around the world. The latest generation of AMD Opteron processors were built from the ground up to deliver enhanced throughput for scalable computing environments like HPC, cloud clusters, virtualization and database applications.
Powering 11 of the world’s top 25 fastest supercomputers ranked in the bi-annual TOP500® list1 , AMD’s enterprise class 2-way and 4-way AMD Opteron™ processors provide consistent, energy-efficient server platforms that scale to meet any HPC demands.
AMD’s world class AMD Developer Central provides customers with the tools and compilers needed to help build powerful, efficient HPC solutions for solving today’s complex computing requirements.
HPC Computing for Dummies available now
AMD’s Chip Freitag talks about the AMD Core Math Library (ACML) in a Podcast with RCE
Scuderia Ferrari Case Study
AMD Opteron™ processors help drive Scuderia Ferrari further, faster
Cray ORNL - Boeing Case Study
Boeing Uses ‘Jaguar’ to Improve Aerodynamics Codes
Cray ORNL - Combustion Case Study
Simulations Speed Design of Power and Propulsion Devices
Cray ORNL - Hurricane
Researchers Model Hurricanes Using ‘Jaguar’ and Airplanes
Cray and CPTech: Learn how AMD and the Cray XT6 supercomputer are powering Brazil’s largest numerical weather prediction and climate research centers
SGI and University of Sao Paulo Video: Accelerating Astronomy Research with SGI Altix Ice and AMD Opteron™ Series processors
SHARCNET Consortium Builds High Performance Computing Cluster with AMD Opteron™ Processor Technology.
Texas Advanced Computing Center (TACC)
AMD Technology-powered Supercomputer Explores Next- Gen DNA Analysis and Delivers Personalized Medicine
AMD Technology Contributing To An Energy-Efficient World: High Performance Computing
By powering high performance supercomputers, AMD technologies are helping to find solutions to the world’s energy and environmental challenges.
Using ‘Jaguar,’ Researchers Shed Light on Astrophysics’ Dark Spot — Turbulence
ORNL and Purdue Explore Technology at the Nanoscale
‘Jaguar’ Helps Solve the Curious Case of Germanium-72
ORNL’S Cray XT ‘Jaguar’ Helps BMI Win Award, Nation Save Fuel
Helping to predict the impact of large magnitude earthquakes using the Cray Xt4™ system “HECToR” (the UK’s national supercomputing service).
Taiwan's National Center for High-performance Computing (NCHC)
To further their service and support for the growing HPC research needs, the NCHC sought to build a new supercomputing system that can offer scalable HPC resources to many institutions across the island.
NERSC (National Energy Research Scientific Computing Center)
Has a mission to provide high end computing communication and storage services to the Department of Energy’s Office of Science while supporting over 3000 users and 500 different applications.
- Based on TOP 500® List from June 2011 (1-100). See http://www.top500.org/list/2011/06/100.