Cluster Connection » Xeon

One product. Three architectures. You really can have it all…

Ian Miller — Fri, 26 Mar 2010 00:49:53 +0000

Whoever said you can’t have your cake and eat it too hasn’t met the Cray CX1000 supercomputer. Brought to you by Cray’s Productivity Solutions Group (the folks who brought you the Cray CX1 deskside supercomputer), the Cray CX1000 series introduces the idea of high(brid) performance computing. This concept delivers high performance computing (HPC) on your terms – allowing you to select the HPC architecture that best suits the problem you’re trying to solve.

Each of the Cray CX1000's technologies is best-of-class and can be mixed and matched in a single rack – creating a customized hybrid computing platform to meet a variety of scientific workloads.

It's like you can finally get that square peg through the round hole.

Check out the options:
The CX1000-C offers scale-out computing, offering 18 nodes of dual-socket Intel Xeon 5600 processing in a 7U form factor, including a 36 port QDR IB switch and a 24 port Ethernet switch.

The CX1000-G offers scale-through (GPU) computing, offering nine nodes of dual socket Intel Xeon 5600 computing, accelerated by dual NVIDIA Tesla sockets, with the same switches as above.

The CX1000-S offers scale-up computing with SMP nodes built on Intel’s QuickPath Interconnect (QPI) technology offering “fat memory” nodes.

The Genesis of the CX1000
The Cray CX1 system was designed to take Cray’s heritage in supercomputing and make that experience available to users with supercomputing needs – but without big research lab budgets. This $15-100K class of user told us they wanted accessible, easy-to-implement HPC. And we responded with the CX1 deskside supercomputer – our first product brought to you in partnership with Intel.

With the power of HPC now in hand, our new worldwide network of users and resellers started reporting back. They wanted Cray to expand the product family such that the CX1 system’s benefits – ease of use, seamless integration – could be extended to more powerful implementations.

We bit.

We’re very pleased to announce the CX1000 series – a product that builds on the CX1 story, but takes it up to 10 Teraflops and beyond. Starting at under $100K, the CX1000 gives users high(brid) performance computing in both software and hardware. The Cray CX1000 series offers the three most important HPC architectures of the next decade – scale out, scale through and scale up. And it also allows you to select from the widest range of industry-standard HPC software through full support of Windows HPC Server 2008, Linux or dual-boot configurations.

So go ahead, eat that cake.

For more information, go to www.cray.com/Products/CX1000/Chassis.aspx

Blade Systems Make Perfect HPC Cluster Building Block Solution

Maria McLaughlin — Thu, 14 May 2009 20:46:29 +0000

In some sense, blades have always used a more green design. In terms of energy efficiency, the advantage of blades is the ability to use shared power and cooling. The Appro GreenBlade as an example takes full advantage of these features. Each blade contains a dual socket Nehalem motherboard with 8 cores, up to 48 GB of DDR3 1333MHz memory, and an optional 1TB of internal storage. Ten blades fit in 5U chassis for a total of 80 cores, 480 GB of memory, and 10 TB of storage.
By adopting an improved green design, users can expect a 20% overall reduction in power use over previous 1U solutions. Coupled with the power saving capability of the new Intel® micro-architecture, the Appro GreenBlade offers outstanding compute/power efficiency. Power savings are derived from the shared redundant cooling fans and power supplies. The power supplies are high-efficiency (90%+) and configured in a N+1 redundant fashion. Another advantage of bladed systems is the enhanced manageability that 1U solutions cannot offer. Each subrack chassis has a shelf manager that provides power control and staged power-up of all blades. Blades, fans and power supplies can be monitored remotely using standard IPMI/BMC tools.
The system comes in a 5U form factor and holds up to 10 blade nodes in one chassis offering up to 80 processing cores in a computing solution box that can scale as the need for compute capacity expands. By sharing power and cooling in one package, GreenBlade helps HPC users save up-front costs and reduce monthly energy and air-conditioning bills. The Appro GreenBlade system also adds redundancy to each server to prevent unplanned downtime, making it easier to allocate a single spare to protect many servers at once. With its open and modular architecture, it provides the essential elements for building, deploying, maintaining, and upgrading clusters for greater customer flexibility and choice.
In addition, the initial entry price point of the Appro GreenBlade is at price parity with 1U rack-mounted servers, making the crossover point favorable for the GreenBlade system and a better fit for HPC midmarket deployments.
Density and Reliability
A typical 1U solution can hold one or two motherboards. With a single motherboard solution, only 40 cores can be placed in the same 5U space occupied by the Appro GreenBlade System thus requiring additional rack space (and power) to achieve the same core parity. If a dual motherboard solution is used, then the same density can be achieved but at the cost of sharing a single non-redundant power supply for two motherboards. The loss of a power supply or the need to service one of the motherboards requires both motherboards to be out of service. Because the GreenBlades share redundant power supplies, each blade is hot-swappable and can be serviced without affecting any other blade. Eight sub-rack chassis can be housed in a standard 42 rack for a total of 640 cores.

The Intel® Cluster Ready Program enables faster, easier deployment of higher performance HPC clusters based on the Intel® Xeon® processor 5500 series. Intel Cluster Ready makes it simpler to choose, deploy and manage HPC clusters, and ensures application and system interoperability. In doing so, it enables companies to rapidly capitalize on the Intel Xeon processor 5500 series and use it to solve bigger problems, expedite total application performance, and reduce TCO. Bottom line: simpler, more powerful and more cost-effective high performance computing. The Intel Cluster Ready certification gives customers another reason to deploy the Appro GreenBlade solution by offering confidence that the cluster architecture and software are certified to work together. It also gives Independent Software Vendors (ISV) the ability to run their software in this certified cluster platform providing a fast and cost-effective implementation for industry-specific solutions.
Not all blades are created equal. In the HPC sector, blades need to be both green and provide high performance. The Appro GreenBlade System provides an extremely flexible and manageable way to create high-density computing. It is designed to scale, with little effort or re-configuration.
The Appro cluster solution based on the GreenBlade building block offers a variety of interconnect options to include only standard 1U Ethernet or 1U Infiniband (IB) switches. This is accomplished by pre-allocating space in each rack cabinet for two standard 24-port Infiniband or Ethernet switches. This offers flexibility for customers who may choose to deploy either an Ethernet or IB-based HPC cluster. Appro also offers variations in the Infiniband interconnect with ConnectX DDR single IB, DDR dual IB or QDR single IB configurations. The Appro GreenBlade is unique in this category; each blade can provide one or two QDR InfiniBand ports on the motherboard and thus do not require any additional PCIe HCAs to be added to the blade. All these options provide excellent network bandwidth and low latency at an affordable price point to many HPC workloads that require higher performance for their applications.
In addition, the Appro GreenBlade offers standards-based Intelligent Platform Management Interface (IPMI) with a choice of Appro’s remote server management or open source cluster management software solutions.
Long considered to be primarily useful only for HPC-scale deployments, blades are beginning to make inroads into the midsize HPC segment. With the introduction of blade products designed specifically for the needs of this market, such as the Appro GreenBlade system, midsize HPC customers are more likely to turn to blades for their day-to-day server infrastructure needs. The Appro GreenBlade is suited for small to midsize departmental and workgroup high performance computing data centers looking for flexibility of system configuration and network connectivity. This addresses the need for performance, reliability, floor space, power and cooling limitations, storage and remote management options at an affordable price. In addition, the Appro GreenBlade is ideal for HPC applications that require high bandwidth and low latency access to system memory, clustering, and Storage Area Networks (SAN).

Why Intel® Xeon 5500 Processor Series?

Maria McLaughlin — Thu, 14 May 2009 20:39:18 +0000

The new Intel® Xeon Processors 5500 series is a new micro architecture. The most significant change is a true quad-core design with much improved memory bandwidth. In past Intel designs, all the processor sockets on the motherboard shared the same memory controller. As the core count on the nodes increased, this introduced a possible bottleneck for some applications. The Intel® micro-architecture has now placed a memory controller on each processor and added a high speed processor-to-processor interface (12.8 GB/s unidirectional) called QuickPath. The result is a 3.5 times improvement in memory bandwidth as measured by the stream benchmark using a similarly clocked Harpertown processor. Intel reports that applications like LS-Dyna and Fluent see improvements of 2.02 and 2.2 times (respectively) over previous generation hardware.

In addition to improved memory performance, Intel has introduced some other enhancements. The New Intel® Xeon Processors now include Intel® Turbo Boost Technology that improves performance by allowing the operating system to increase core frequencies within the current power envelope. The idea is to increase the clock speed of cores if others are idle while still staying within the thermal specifications. In addition, Intel® Intelligent Power Technology allows for individual cores to be powered down when not in use. The power draw for both memory and the I/O controller on idle nodes can be reduced as well, bringing the idle power consumption for the Nehalem to 10 watts.

There are other improvements worth noting. While not used as much in HPC, Intel® Hyper-Threading Technology is back allowing more threads than cores to operate at the same time. Memory Capacity is now 144GB and the L3 cache is shared between all cores on each Nehalem processor. There have been other improvements in virtualization technology as well.

Pushing The Boundaries ...

Ian Miller — Mon, 11 May 2009 21:08:53 +0000

The HPC community is always pushing on the boundaries of what is possible with computing. While overall scalability and the Flop count is one key metric, equally important is the ability to test codes and evaluate new technologies as they arrive to see if a full "data center" deployment is warranted. One of the ways a product like the CX1 (certified Intel Cluster Ready naturally!) can help the community is by being a powerful yet inexpensive test bed for these technologies. Obviously, the CX1 supports the exciting Nehalem Xeon 5500 processors, and that is it’s core underpinning - allowing codes to leverage the performance and memory speed of that technology.

In addition, several other important capabilities can be evaluated cost effectively on the CX1. For example, the Windows HPC Server 2008 platform is seeing strong market acceptance along with the CX1, and the workstation community see this as a natural evolution from the desktop workstation environment. In addition, users can leverage Cluster Resources Moab dual boot capability to host both Linux and Windows workloads on one CX1 system. Customers who want the simplicity and ease of use of an SMP environment can use ScaleMP's vSMP software on the CX1 platform, and of course GPGPU's are also growing in popularity.

Those are just a few ideas of subjects that are exciting and seeing traction in today's HPC market place. The Cray CX1 team look forward to hearing your comments and ideas for other technologies or applications you would like to see on the platform. Feel free to drop a line to CX1info@cray.com with your ideas or if you have any questions.

Xeon Processor and Intel® Cluster Ready

Steve Conway — Fri, 01 May 2009 17:56:57 +0000

An intriguing aspect of the Intel Cluster Ready (ICR) program is the impressive stable of independent software vendors (ISVs) serving the high performance computing market that have become members. Nearly all of them are mainstream rather than non-established players. The list reads like a who's who of ISVs serving the HPC market and covers a wide range of vertical segments and computational modeling methods.

There's LS-DYNA, used for finite element analysis in the automotive, aerospace, military, manufacturing, and biosciences segments; Acclerys' Discovery Studio and Materials Studio; and a raft of key fluid flow and structures applications, including RADIOSS, Ansys, STAR-CD, FLOW-3D, various flavors of Nastran and Abaqus, and others.

Making key ISV applications available in conjunction with the ICR pre-integrated, pre-tested reference architecture is an important advance, especially for new and less-experienced HPC users. IDC research studies conducted for the Council on Competitiveness and other parties confirmed that one of the biggest barriers to HPC adoption by desktop users was uncertainty about whether the third-party apps they were using would run on clusters. Many of the surveyed desktop users were small engineering services firms, tier 2 and 3 suppliers to large automotive, aerospace, and other manufacturing firms. The vendors of the most popular third-party applications are on the ICR member list. This means that ICR could become a powerful catalyst for new HPC adoption.

But that's not all. Many engineering applications, particularly codes used for structural and fluid-structures analysis, are low scaling and communications-intensive. They require lots of bandwidth at the core and node level. The new Nehalem-based Xeon 5500 series processors are designed to boost the per-core bandwidth that has been greatly needed to advance per-code and per-node performance on applications like these. And for applications that scale beyond a node, Intel's QuickPath Interconnect (QPI) that is part of the Nehalem microarchitecture better supports interconnects that provide high system-wide (bisection) bandwidth.

In tandem, ICR and Nehalem promise to be a winning combination for meeting the "ease-of-everything" requirements of new and less-experienced HPC users (not to mention higher-end users). Here's how the symbiosis works: ICR is designed to make HPC clusters, which are notoriously tough to manage, substantially more tractable; combined with the new Nehalem-based Xeon 5500 processors that provide the bandwidth muscle to boost performance substantially, even on low-scaling engineering codes; and then add in ICR that delivers the Xeon 5500 with its enhanced bandwidth in a pre-integrated, pre-tested form.

From an administration standpoint clusters can be wild beasts. ICR is designed to tame them in a way that might make Siegfried and Roy proud. On the processor side of the symbiosis, early benchmark results for the Xeon 5500 series have been impressive. It will be interesting to see the outcomes when Intel has had additional time to benchmark a broader set of HPC applications.

Intel® Cluster Ready Program Brief - April 2009

ICR Team — Mon, 20 Apr 2009 21:18:46 +0000

[Excerpt] High-performance computing (HPC) is a key differentiator and productivity engine for business. With the Intel® Cluster Ready program, businesses can exploit the power of high-performance computing more rapidly, easily, and cost-effectively.

Developed collaboratively with hardware and software vendors, the Intel Cluster Ready program makes it simpler to buy, deploy, and manage an HPC cluster. Intel Cluster Ready helps ensure application and component interoperability - from the minute you first power up the cluster through the lifetime of the system. And, with an Intel Cluster Ready system powered by the new Intel® Xeon® processor 5500 series, you can unleash even more parallel processing performance. Simulate, analyze, and visualize more complex models faster, and accelerate your data-intensive applications - all in a smaller, denser, and more energy-efficient footprint. The net impact: Faster time to value, lower total cost of ownership (TCO), and greater business impact.

Click here to read more

NEC Adds New Advanced Servers for Virtualization

Dr. Thomas Schoenemeyer — Thu, 16 Apr 2009 21:50:08 +0000

[Excerpt] Next Generation NEC Servers Built on New Intel Xeon Processor; Offer Higher Performance, Energy-Efficiency, High Bandwidth Vital for Virtualization and Consolidation.

Click here to download "NEC Adds New Advanced Servers for Virtualization"

Penguin Announces Releases Based on Intel® Xeon 5500 Processors

Arend Dittmer — Thu, 16 Apr 2009 21:42:42 +0000

[Excerpt] Penguin Computing, an Intel Premier Partner, announced today three new Relion™ 1700 Series products based on the Intel® Xeon® processor 5500 – the Relion 1700 and 2700 dual processor enterprise servers and the Relion 1702 twin-node, dual processor per node HPC cluster server. Penguin’s new servers coupled with advanced fine grained power control deliver significantly greater performance and scalability, offering a foundation to refresh aging IT infrastructure, overcome server sprawl and accelerate return on investment.

Click here to download "Penguin Announces Releases Based on Intel® Xeon 5500 Processors

Appro Unveils High-Density HyperGreen™ Cluster Solution

Maria McLaughlin — Thu, 16 Apr 2009 21:18:00 +0000

[Excerpt] Appro Unveils High-Density HyperGreen™ Cluster Solution to boost Power Efficiency and Performance in the Datacenter. Appro, a leading provider of supercomputing solutions, announced today the launch of the new Appro HyperGreen Cluster Solution and the support of the new Quad-Core Intel® Xeon® Processors 5500 series across all product lines.

Click here to download "Appro Hypergreen Press Release"

Intel® Xeon® 5500 brings new era to HPC

Gilad Shainer — Tue, 07 Apr 2009 18:56:17 +0000

The availability of the new Intel® Xeon® 5500 Processor CPUs brings a new era to High Performance Computing. The available compute resources have now reached new dimensions and can provide competitive advantages for commercial business. In order to ensure high-productivity from the new Xeon®-based systems, one needs to make sure that the systems are balanced, meaning that there are no artificial bottlenecks that will limit the compute potential. One of those potential bottlenecks is the server interconnect. With eight cores or above in a single server, and with the new levels of compute capabilities from each of the cores, you need to be able to transfer data in and out from the servers at speeds that are much greater than 10Gb/s. One option is to have multiple 10Gb/s adapters, which in turn increases the needed real estate, the system’s power consumption and the total cost of ownership. Another option is to use available 40Gb/s networking, such as 40Gb/s InfiniBand to ensure the highest scalability, efficiency and productivity.

I have been doing an analysis on application productivity, comparing Intel® Xeon 5500 with 40Gb/s InfiniBand (using Mellanox's end-to-end networking solutions) to Intel® Bensley with 20Gb/s InfiniBand. One of the applications tested was the Weather Research and Forecasting (WRF) model. With Intel® Xeon based servers and Mellanox 40Gb/s InfiniBand we have seen a 100%(!) productivity increase.

With Intel® Xeon® 5500 support for PCI Express 2.0 and Mellanox 40Gb/s InfiniBand you are now able to achieve more than 6.6GB/s throughput on a single port, almost 1us latency and high message rate of more than 40 million messages per second for mission critical and high-performance applications. The Intel® Cluster Ready program makes it easier for users to take those technologies or solutions, and be able to maximize their performance capabilities to ensure the most efficient and productive use.