Cray extends scalable MPP architecture; introduces industry’s first integrated hybrid supercomputer with Cray XT5 family: Cray today launched the Cray XT5 family of supercomputers, marking a significant step toward Cray’s vision of adaptive supercomputing. Incorporating all the benefits of the successful Cray XT line, the Cray XT5 massively parallel processor (MPP) system includes a new compute blade that quadruples local memory capacity, doubles processor density and improves energy efficiency for a significant reduction in total cost of ownership for customers. Continuing to drive industry leadership in extreme scalability, the Cray XT5 family also includes the industry’s first integrated hybrid supercomputer, the Cray XT5h system. With new programming paradigms and processor technologies, the Cray XT5h system optimizes application efficiency and helps customers tackle problems that cannot currently be solved on a single architecture system.
“The Cray XT5 family delivers superior sustained application performance from a single cabinet to massive scale while offering lower power consumption, industry-leading high density packaging, innovative cooling technologies and a fully upgradeable path from Cray XT3 and Cray XT4 systems,” said president and CEO Peter Ungaro. “This new product family builds on the scalability of the Cray XT product line with the added productivity benefits of hybrid processing to deliver shorter times to solution and competitive advantages to our customers over a wider range of applications.” Cray XT5 Supercomputer
The Cray XT5 system is the world’s most scalable Linux supercomputer. It combines unprecedented sustained application performance with exceptional manageability, lower cost of ownership and, with a Linux operating environment, broader application and tools support to large supercomputer class machines. In addition to supporting the current Cray XT4 compute blades, a new eight-socket Cray XT5 compute blade supports the powerful dual-core and new Quad-Core AMD Opteron processors. Each dual-socket node supports up to 32 gigabytes of locally addressable memory which can be accessed at up to 25.6 gigabytes per second. Using a next-generation Cray SeaStar2+ interconnect chip, the entire system is designed to scale and architected to avoid performance losses associated with communication bottlenecks, memory access delays and operating system jitter.
Cray XT5
Recognizing the growing need to reduce energy usage and control operating costs, the Cray XT5 family employs innovative packaging and technologies that reduce power and cooling requirements. Vertical cooling takes cold air from the floor with a single, high-efficiency turbine fan and efficiently cools the processors on the Cray XT5 blades. The compute blades are designed for optimal airflow with position-dependent custom designed heat sinks. A single cabinet can support very high density CPU configurations of 192 processor sockets or 768 CPU cores. Easily upgradeable and expandable, existing customers can upgrade to the Cray XT5 system from the Cray XT3 or Cray XT4 systems and/or add on to their existing Cray XT systems, thereby leveraging their investment over a longer life. Cray XT5 cabinets can be configured with Cray XT4 compute blades, for optimized compute-to-communication balance, or with the new high-density Cray XT5 compute blades, for memory-intensive and/or compute-biased workloads. Additionally, the Linux operating environment in the Cray XT5 system enables optimal performance across a broader range of applications. Cray XT5h Hybrid Supercomputer
Recognizing that no single architecture is ideally suited for all types of applications, Cray has developed the industry’s first integrated hybrid supercomputer, the Cray XT5h system. With the Cray XT5 system as its foundation, the Cray XT5h supercomputer integrates multiple processor architectures with a complete software development environment into a single system supporting diverse workflows. The Cray XT5h system couples industry-leading scalar processing capability with high memory-bandwidth vector processing and reconfigurable co-processing using field programmable gate array (FPGA) technology, establishing a new paradigm in high performance computing (HPC).
Cray XT5h
“The stalling out of standard microprocessor speeds is starting to spark a rebirth in the HPC industry toward computers that augment microprocessors with other processor types including vector processors, GPUs, accelerators and FPGAs,” said Earl Joseph, IDC program vice president, Technical Computing Systems. “Cray is developing new computer architectures to address this growing requirement with its new hybrid supercomputer. Cray has a long history and experience using multiple processors in its supercomputers including vector processors, custom I/O processors, industry standard microprocessors, multithreaded processors and FPGAs which will add value and capability to their new hybrid designs.” Supporting a variety of processing technologies, the Cray XT5h system integrates its Opteron support with a vector system and FPGAs.
- Vector and Large Memory Processing The vector compute blades for the Cray XT5h supercomputer – called Cray X2 blades, provide breakthrough high bandwidth vector processing capabilities. A Cray X2 compute node is the core building block of the system. It has four vector CPUs implemented as a four-way SMP. With 64 gigabytes of shared memory, each Cray X2 node is capable of more than 100 gigaflops of peak performance and systems can be scaled to 1,024 shared memory processors with 16 terabytes of globally-addressable memory. The Cray X2 vector processor of the Cray XT5h system has unique global addressing capabilities programmable by Co-Array Fortran and Unified Parallel C (UPC), which can solve problems beyond the capabilities of MPI.
- FPGA
A Cray XR1 reconfigurable computing blade consists of two pairs of AMD Opteron processors tightly coupled with two Reconfigurable Processor Units (RPU) A HyperTransport interconnect links general-purpose and reconfigurable processors instead of much slower PCI-based interconnect commonly employed in many reconfigurable solutions. This tight coupling of reconfigurable processors with AMD Opteron processors ensures low latency and high-bandwidth communication between processing elements, allowing users to scale applications to thousands of reconfigurable processors.
