Technology Focus
FastCluster 3000 SERIES VXS Systems
The new FastCluster 3000 SERIES VXS systems deliver unrivaled performance in bi-section bandwidth and processing density while preserving absolute code reuse at the application
layer. The 3000 SERIES is targeted to high performance digital signal processing (DSP), SIGINT, radar and sonar applications in airborne, shipboard and UAV platforms where space
and power are at a premium.
The 3000 SERIES redefines processing per-cubic-foot-per-watt and interconnect bi-section bandwidth by
integrating the reconfigurable processing capabilities of Xilinx's FPGA technology with the latest Freescale Multi-Core processors and Myri-10G on a VXS platform. With its
built-in 10-Gigabit Ethernet technology, the 3000 SERIES supports the DOD's vision of "systems of systems" in which embedded systems are not designed, deployed,
and used in isolation but rather in a cooperative way.
The FastCluster 3000 SERIES architecture delivers
uncompromised scalability starting at 20 GBytes/s of bi-section bandwidth for a Single Star Myri-10G fabric. A Dual Star VXS
chassis with 16 payloads doubles the bi-section bandwidth to 40 GBytes/s. A four chassis configuration with 256 PowerPC
AltiVec cores provides 160 GBytes/s of bi-section bandwidth. This approach preserves the advantages associated with a
modular small chassis footprint. In addition support of the 10-Gigabit Ethernet standard facilitates sensor data acquisition
by the FPGA/Multi-Core processing nodes, and interoperability with workstations, raid disks, etc, enabling a low cost and
practical insertion of FastCluster into future DOD platforms.
FastCluster 3000 SERIES air-cooled systems are
scheduled for test bed evaluation by Q4 of 2006 with follow-up conduction cooled versions. In a second phase, 3000 SERIES core
technologies will migrate to enhanced MIL-SPEC rugged platforms with VITA 46/48 packaging and liquid cooling.
Feature Products
FastCluster conduction-cooled blades
are designed to the IEEE 1101.2-1992 standard for mechanical core specifications and ANSI/VITA 47-2005 environmental
specifications for conduction cooled boards to insure reliable system operation when subjected to wide temperature ranges;
moisture, vibration or extreme shock; dirty or corrosive atmospheres; and environments with little or no airflow. The new conduction cooled blades
leverage the established open software environment currently available for the FastCluster convection cooled platform.
The FastCluster standard development platform includes: the VSIPL and ISSPL Optimized Signal Processing Libraries
for PowerPC AltiVec, the optimized MPI Message Passing Interface over Myrinet, the VxWorks™ Real Time Kernel and
Tornado Development Environment and the Linux OS and Open Source Development Package.
Incorporating a low power, high frequency, superscalar processor for embedded applications and the latest in network communication standards for high performance computing the FastCluster 2542C blade boasts an architecture that strikes an ideal balance between performance, power consumption, size and scalability for a variety of applications including radar, sonar and simulation.
Packaged in an industry standard 6U VME form factor the FastCluster 2542C Blade targets high compute density requirements with four independent PowerPC 7447A compute nodes interconnected with a 16 port Myrinet switch (full duplex). The FastCluster 2000 SERIES architecture supports 1 GB/s of interconnect to the backplane Myrinet clustering fabric via the VME PO connector (VITA 26 Myrinet ANSI Standard).
FastCluster 2542C blades offer many high availability features to meet the stringent demands of our defense customers. Standard features include fast booting from a cold start, error-correcting memory, built-in self test and conduction cooled, conformal coated boards for systems intended for use under extreme environmental conditions.
Combining rugged features with high performance computing and I/O streaming capabilities on a single VME 6U board, the StarGate 2523C delivers the computational power of the Atmel PowerPC 7447A with AltiVec® operating at 750 MHz.
The StarGate architecture implements two independent processing nodes. Each processing node supports two fast PCI buses (64 bit/64 MHz) enabling concurrent I/O and data processing to the SDRAM memory. One PCI bus is reserved to handle Myrinet® data transfers, at the full duplex speed of 250 MB/s in each direction, from the LANai network processor to the VME P0 connector on the backplane. The other PCI bus supports a single PMC slot for high speed data streaming operations.
The StarGate 2523C is an ideal choice for use in the harsh and confined environments of shipboard, airborne and land-mobile platforms.
CSPI In the News
CSPI Selects Myricom's Myri-10G for its 2nd Generation of Rugged Clusters
BILLERICA, Massachusetts, October 24, 2005 - CSP Inc. (NASDAQ: CSPI)
MultiComputer Division, a leading supplier of rugged embedded clusters, announced today its selection of Myricom, Inc.'s Myri-10G technologies to power its next
generation of systems. In close partnership with Myricom, CSPI will integrate the Myri-10G chipset, Myrinet Express (MX) and MPI/MX open software stack into its
rugged embedded Linux FastCluster. Combining CSPI's experience in high performance embedded computing with Myricom's new 10-Gigabit Ethernet clustering
technology offers an advanced Open Architecture for DOD programs requiring scalable platforms in harsh environments. "Through our long-term strategic relationship with Myricom we are able to offer
our FastCluster customers a dramatic tenfold improvement in backplane interconnect performance as well as broader interoperability with other 10-Gigabit Ethernet
systems," stated Alexander R. Lupinetti, Chairman, President, and Chief Executive Officer of CSP, Inc.
The selection of the Myri-10G solution extends a ten-year working relationship
between CSPI and Myricom that includes pioneering DARPA's vision for HPEC deployable open platforms based upon MPI (Message Passing Interface).
The new Myri-10G technology will find its way into FastCluster blades that will start reaching the market by the second half of 2006. "CSPI's deployment of our previous generation Myrinet-2000
interconnect in major DOD programs has been very successful both in system performance and in smooth development cycles," says Dr. Chuck Seitz,
Myricom founder and CEO. "With the selection of Myri-10G for its next generation of high-density rugged clusters, we look forward to helping CSPI
deliver new levels of performance in extremely demanding computational and physical environments." The Myri-10G chipset and MX software leverage Myricom's experience
as the leading global provider of high-performance cluster-interconnect technology. Myri-10G features a low power, high density, 16-port crossbar switch and
a high-performance PCI Express (8 lanes) network processor, both with XAUI (IEEE 802.3ae) 10-Gigabit data rate physical layer interfaces to the switched fabric. "Our FastCluster systems will benefit not only from the increased
performance delivered by Myri-10G but from its robust set of features including OS-bypass, overlapping I/O and computation, source-based routing,
route dispersion, cut-through switching, and the MX/Linux open-architecture advantage," said Bernard Pelon, Director of Product Research. More advanced 2nd Generation FastCluster systems will be introduced later
enabling dense and scalable embedded platforms with 100's to 1,000's of Multi-Core processing nodes delivering multi-teraflop capability in harsh environments
where cooling, weight, power and size are critical.
CSP Inc. Delivers FastCluster 2000 SERIES Systems to Lockheed Martin
Partnership brings HPC capability and 10GbE Interoperability to Embedded Cluster Computing
in Support of the Radar Scene Generation Program for the Aegis Combat System
BILLERICA, Massachusetts, March 30, 2006 - CSP Inc. (NASDAQ: CSPI) MultiComputer Division, a leading supplier of embedded high-performance cluster computing systems, today announced that Lockheed Martin (NYSE: LMT) has selected the FastCluster 2000 SERIES for its real-time computing platform in support of the Radar Scene Generation (RSG) Program. The RSG provides a test and evaluation resource at the U.S. Navy's Combat System Engineering Development Site (CSEDS) for the Aegis Combat System (ACS) in Moorestown, N.J.
"Lockheed Martin's selection of our embedded cluster validates and reinforces CSPI's commitment to delivering an open scalable platform capable of supporting the most demanding real-time computing requirements," said William E. Bent, Jr., General Manager and Vice President of CSPI MultiComputer Division.
Utilizing the FastCluster platform reduces development time, efforts and costs for the RSG program and supports the Defense Advanced Research Projects Agency (DARPA)'s code reuse vision of "write once, run everywhere." Specifically, the RSG program will take advantage of the outstanding performance of the FastCluster 2942 6U blades that incorporate all industry-standard software, highly optimized math libraries and the Message Passing Interface (MPI) for interprocessor communication. Fulfilling Lockheed Martin's stringent program requirements, the FastCluster 2000 SERIES is a commercial-off-the-shelf (COTS) hardware and software scalable platform delivering real-time signal processing, optimized performance, code re-use and portability, a flexible design and field scalability.
CSPI shipped several FastCluster systems, for an approximate value of $1 million, in support of the first phase of the program. Follow-on orders are expected to fulfill the requirements of the RSG program.
Corporate News
CSP Inc. Reports Second-Quarter Fiscal 2005 Financial Results
Company Reports 38% Increase in Quarterly Net Income on Strength of MultiComputer Business
BILLERICA, MA, May 15, 2006 - CSP Inc. , a provider of IT solutions, systems integration services and dense cluster computing systems, today reported financial results for the second quarter of fiscal 2006 ended March 31, 2006.
For the second quarter of fiscal 2006, CSP Inc. reported sales of $17.6 million compared with $18.7 million in the second quarter of fiscal 2005. Net income for the second quarter was $1.0 million, or $0.27 per diluted share, compared with $750,000, or $0.20 per diluted share, in the second quarter a year ago. CSP's second-quarter fiscal 2006 net income includes an expense of approximately $384,000 for professional fees related to the previously announced restatement of the Company's interim and annual financial results for fiscal 2004 and fiscal 2003.
The Company recorded equity-based compensation expense of $62,000 for the quarter under SFAS 123R for employee and director stock options and the employee stock purchase plan.
"CSP Inc. performed well in the second fiscal quarter," said Alexander R. Lupinetti, chairman and chief executive officer of CSP Inc. "We reported a 38 percent year-over-year increase in quarterly net income, driven by solid margin performance in our Systems segment from our Multicomputer Division as a result of increased revenues and improved overhead absorption."
"Our Multicomputer Division revenues were from shipments related to a new program win as well as other deployments," continued Lupinetti. "During the quarter, we announced that we had been awarded a contract to supply FastCluster 2000 SERIES computers for Lockheed Martin's real-time computing platform in support of the U.S. Navy's Radar Scene Generation (RSG) Program. The RSG provides a test and evaluation resource at the Navy's Combat System Engineering Development Site for the Aegis Combat System. We shipped a FastCluster system in the second quarter for approximately $1 million to support the first phase of the program. We plan to ship additional FastCluster multicomputers to fulfill the requirements of the RSG program. We also announced the introduction of our second-generation FastCluster 3000 SERIES, which provides our customers with the most advanced interconnect technology available today," added Lupinetti. "We have already begun to see interest in this offering from our customers. Looking ahead, we continue to anticipate significant opportunities for our fully ruggedized open source multicomputer products. As in the past, however, we expect revenues to continue to be lumpy as a result of timing of shipments."
"At MODCOMP, which operates in our Service and Systems Integration segment, our overall market share remains solid and we continue to see demand for our IT systems integration services and best-of-breed integrated IT solutions," said Lupinetti. "However, in order to cut costs, two integration services customers of our German subsidiary have significantly reduced their contract service levels. To align our workforce with current business volume, we are in the process of negotiating a reduction of our German staff. As a result, we expect to incur costs in the range of approximately $500,000 spread over the next two to three quarters. At our U.S.-based IT Systems and Solutions division, we are adding to our sales force to grow our market position as a provider of solutions for integrated IT environments."
At a Glance
Upcoming Events
The Tenth Annual Workshop on High Performance Computing
September 19-21, 2006
MIT Lincoln Laboratory
244 Wood Street
Lexington, MA 02420
Embedded Systems Conference - Boston
September 25-28, 2006
Hynes Convention Center
Boston, MA