Intel and Microsoft Boost High-Performance Computing
Intel and Microsoft up the stakes in high-performance computing.
Industry giants Microsoft and Intel Corp. last month independently released new software addressing high-performance computing (HPC).
Microsoft released the second beta of its Microsoft High-Performance Computing Server (HPC Server), the successor to Compute Cluster Server 2003.
Based on Windows Server 2008, HPC Server is slated to ship in the second half of next year and is crafted specifically to run across large compute clusters. Among some of the key new features, Redmond's new HPC software includes a service-oriented architecture (SOA) job scheduler; support for partners' clustered file systems; better failover capabilities; more efficient and scalable management tools; and high-speed networking capability. Microsoft says it's seeing significant performance gains with HPC Server.
"By upgrading to Windows HPC Server 2008 on our 2,048-core production test cluster, we increased the LINPACK performance by 30 percent and were able to deploy and validate the cluster in less than two hours using out-of-the-box software," says Kyril Faenov, general manager of HPC at Microsoft. LINPACK is a software program that solves high-level mathematical equations and is frequently used to benchmark computers.
Intel Introduces Toolkit
Microsoft's release of the latest HPC Server beta comes just a week after Intel launched new software development tools aimed at HPC and cluster-computing environments. The Intel Cluster Toolkit v3.1 provides updated software analysis tools, Message Passing Interface (MPI) libraries, compilers and mathematical libraries to enable clustered-app development, says Intel Chief Software Evangelist James Reinders.
"We see a trend of more and more cores, and we actually see MPI going more and more mainstream," says Reinders, who says the Cluster Toolkit aims to ease challenges related to clustered apps. "Some software needs to be installed on each and every node. This product does that for you."
Margaret Eppstein, assistant professor of computer science and co-director of the Complex Systems Center at the University of Vermont, welcomes anything that can help her focus on the software, rather than the setup.
"Just the mechanics of getting the systems running with MPI, knowing how to link up all the computers right -- those things are more of a pain in the butt than developing the parallel algorithms," Eppstein says.
The Cluster Toolkit supports Fortran, C and C++ programming languages and various Windows and Linux operating systems.
Intel also announced updates to its cluster-aware C++ and Fortran compilers, and continues to promote the Intel Cluster Ready program, an initiative launched in July that aims to create a standard target for cluster-capable software and hardware.
Reinders says the program will reduce the effort involved in configuring and deploying clusters, drive down support costs and improve operating efficiency. Under the program, systems designers and OEMs submit products for testing to receive Cluster Ready certification.
Intel also provides a Cluster Checker tool for certified systems, which checks the cluster configuration and performance. The checker tool analyzes key areas of cluster organization, functionality and performance, and includes advanced testing to assess both per-node and cluster-wide performance.
RapidMind Extends Multi-Core Support
by Doug Barney
Looking to extend the reach of its multi-core software, RapidMind Inc. this month will release version 3 of the RapidMind Multi-Core Development Platform.
The new version enables single-threaded C++ code to execute across both the system processor and Graphics Processor Unit (GPU). Version 3 adds the ability to split workloads among multiple cores on both AMD and Intel processors.
The release should help dev shops take advantage of the rapid adoption of multi-core systems, which today are often underutilized, says Ray DePaul, president and CEO of Waterloo, Ontario, Canada-based RapidMind.
"When you have four cores you really aren't doing four things at once," DePaul says.
RapidMind serves as middleware that sits between program code and the underlying OS and system. Applications can be tuned for the RapidMind environment, which automatically spreads processing loads across all available processor cores.
DePaul says RapidMind offers an efficient way to tap the power of multi-core systems, adding calls to select areas of code. "We look like a simple library. The developer has to identify the area [to optimize] and then hand it to us."
RapidMind supports C++ code and works with any IDE that supports the C++ programming language, including Visual Studio (Visual C++ versions 7 and 8) on Windows and GCC 4 on Linux. Company officials say the company is considering supporting newer languages, such as C#. The company claims typical performance gains of up to 10 times or more.
RapidMind supports GPUs including ATI Radeon, NVIDIA Quadro and GeForce, and the Cell Broadband engine from IBM Corp. OS support includes Windows XP and Windows Vista, as well as Fedora and Ubuntu Linux. It also supports Yellow Dog Linux for the Sony PlayStation.