Hewlett-Packard and Dell are pushing accelerators for their newest high-performance computing systems that not only will give customers greater performance and power efficiency, but also will enable them to better optimize the systems for their workloads.
At the SC14 supercomputing show in New Orleans Nov. 17, HP officials announced new server trays for the company’s Apollo supercomputers that house the latest GPU accelerators from Nvidia and x86 Xeon Phi coprocessors from Intel. Also at the show, Dell officials unveiled the PowerEdge C4130, a highly dense 1U (1.75-inch) server that can offer up to four GPU accelerators or Xeon Phis.
High-performance computing (HPC) organizations are limited in the amount of space they can use for their systems, as well as by the pace of code optimization and cost of code development, and are looking for density and flexibility, according to Brian Payne, executive director of server solutions at Dell.
“Accelerators have been an interesting technology for them to apply,” Payne told eWEEK.
Businesses and institutions with compute- and data-intensive workloads are increasingly turning to accelerated systems to get the performance they need while keeping down the amount of power the servers consume, which helps reduce the overall costs of running the systems. To get the same performance out of CPU-only systems would require larger, more power-hungry servers.
According to Dell officials citing IDC numbers, the use of GPU accelerators and coprocessors grew 1.7 times between 2011 and 2013, while 77 percent of users with HPC environments reported using them. In the latest Top500 list of the world’s fastest supercomputers released at the show Nov. 17, the number of systems on the list using GPU accelerators from Nvidia or Advanced Micro Devices or Xeon Phi coprocessors from Intel hit 75, up from 62 in June. Most use Nvidia’s Tesla GPU accelerators, while the bulk of the others embrace the Xeon Phi platform, according to the list’s organizers.
Nvidia at the SC14 show on Nov. 17 launched the Tesla K80, the latest generation of its GPU accelerator technology that officials said will offer twice the performance and memory of its predecessor, the K40. Also at the show, Intel officials talked about the chip maker’s “Knights Hill” third-generation Xeon Phi, the 10-nanometer coprocessor due out next year.
HP officials in June unveiled its new family of Apollo supercomputers, which they said will offer up to four times the performance of standard rack servers in smaller and more energy-efficient packages. The portfolio, aimed at hyperscale environments, include the air-cooled Apollo 6000 system and the liquid-cooled Apollo 8000 supercomputer. The 6000 began shipping this summer, while the 8000 will ship in December.
HP now is bringing accelerators into the Apollo mix, according to Ed Turkel, group manager of HPC business development for the tech vendor. For the Apollo 6000, the new trays can hold either Xeon Phi 5100 or 7100 coprocessors or Nvidia Tesla K40 or K80 GPUs. For the 8000, the trays will hold two of the Xeon Phi 7120s or two K40 GPU accelerators, Turkel told eWEEK.
Dell’s C4130 is designed to offer high levels of compute capabilities and accelerator density in a 1U form factor. The system can hold up to two Xeon E5-2600 v3 chips and up to four accelerators, including the new Nvidia Telsa K80. It also can run Xeon Phi coprocessors, Payne said. The system can offer up to 7.1 teraflops of performance on a single server, enabling organizations to do a lot of work in a small amount of space.
A key for both the HP Apollo systems and Dell’s C4130 is the ability to be optimized for particular workloads, meeting a growing demand from customers. The Apollo systems can offer a combination of CPUs and accelerators depending on the HPC application needs. According to Dell’s Payne, the C4130 can come in five different configurations that offer different mixes of CPUs and accelerators for optimized performance.
Along with the new system, Dell also announced a storage solution that combines its PowerEdge servers, Storage MD3460 arrays and MD3060e enclosures that support Intel’s EE for Lustre software. The Dell Storage for HPC with Intel EE for Lustre solution offers a highly scalable solution with fast I/O.
“It’s a Lustre solution that can be right-sized for your environment,” Jimmy Pike, vice president, senior fellow and chief architect at Dell, told eWEEK.