Appro and Penguin Computing are both leveraging the newest chip technologies from Advanced Micro Devices and Intel for high-performance computing systems.
Appro officials on Nov. 2 unveiled plans for the next generation of their Xtreme-X supercomputer, which will be powered by AMD’s upcoming Opteron 6200 “Interlagos” processor and Intel’s Xeon E5, which company executives said in September had gone into revenue production. The system also will support Intel’s upcoming Many Integrated Core (MIC) architecture.
Appro will put the new Xtreme-X supercomputers on display at the SC 11 supercomputer show in Seattle Nov. 14-17.
The new systems are designed to handle high-performance computing (HPC) workloads and offer high-density, improved performance-per-watt capabilities, and multiple connectivity and storage options.
“The Appro Xtreme-X combines the latest cutting-edge technologies with an optimized high-performance computing architecture,” Appro CTO Giri Chukkapalli said in a statement. “This combination delivers outstanding performance, scalability, availability and manageability based on a high-bandwidth and low-latency infrastructure required for today’s HPC workload production requirements.”
The Xtreme-X supercomputer will support two- and four-socket platforms powered by the Xeon E5 chips combined with the MIC architecture, which is designed to create high-performance supercomputing platforms. The Xeon E5, which will hold up to eight cores and run 16 threads per second, is aimed at midrange systems, coming in between the low-end E3 chips and high-end E7 Xeons. During the Intel Developer Forum in San Francisco, Kirk Skaugen, vice president and general manager of Intel’s Data Center Group, said the chip will offer more performance than the current Xeon 5600 “Westmere” processors, including a doubling of the floating-point capabilities and integrating of the PCI-Express bus onto the chip. The Xeon E5 “is the most phenomenal chip we’ve delivered to the server market,” Skaugen said.
He said servers running the chip will hit the market in early 2012.
AMD has begun shipping its Interlagos Opteron processor, which will offer up to 16 cores and a range of features designed to improve performance and energy efficiency. In addition, the Xtreme-X supercomputer also will continue using Nvidia’s Tegra graphics technology in combined CPU/GPU configurations.
The Xtreme-X systems will come in multiple configurations, including one for terascacle and petascale computing that includes integrated Lustre file systems and low-power components. The Xtreme-X for hybrid computing will offer CPU/GPU processing for highly parallel compute workloads.
A configuration for data-intensive computing that calls for high storage and I/O capacity, such as large-scale data analysis, will feature extreme bandwidths between memory, large memory nodes, the use of SMP technologies, and storage using fast disks, solid-state storage and PCI-attached flash. The Xtreme-X for capability computing is aimed at petascale demands and will offer fault tolerance and high availability.
For its part, Penguin is leveraging AMD’s accelerated processing units (APUs), which offer the CPU, graphics technology, memory controllers and a PCI-e interface on the same chip, for a new HPC compute cluster at the Sandia National Labs in Albuquerque, N.M. Penguin’s cluster comprises 104 servers interconnected through an Infiniband networking fabric that will offer a peak performance of up to 59.6 teraflops-or 59.6 trillion floating point operations per second. Penguin’s new Altus 2A00 is the compute platform for the cluster, according to the company.
AMD, which began rolling out APUs for PCs and embedded systems under its Fusion initiative in January, helped Penguin design the Altus 2A00, which Penguin officials said is the first Fusion APU system in a rack-mount chassis and in a 2U (3.5-inch) architecture. AMD’s uniquely designed processor includes 400 parallel processing cores that leverage the OpenCL programming framework.
“With the Altus 2A00, Penguin is the first to bring AMD’s unique APU capabilities to the HPC community,” Penguin CTO Phil Pokorny said in a statement. “We believe that the high level of integration and the resulting benefits for HPC users will further accelerate the adoption of the GPU processing model in HPC. The APU architecture has the potential to become a key component of future exascale systems.”