SC14: Performance Solutions For The Supercomputing World

The Promise Of Performance: CRN Takes You To SC14

Back in the day, the annual Supercomputer conference was about, well, about supercomputers, the kind that only the largest laboratories or universities could afford.

However, the ever-more-favorable power-to-cost ratio of server, storage and networking equipment has brought the cost of building and using high-performance computing (HPC) and other similar high-end solutions within the reality of the everyday enterprise, where they are now used for a wide range of applications from analytics to simulations to big data.

This month's SC14 was a chance for a wide range of IT vendors to break out their highest-performance gear. But not everyone could make the trip to New Orleans to attend. For those, CRN has brought together a dozen representative technologies sure to bring out the geek in everyone. Turn the page and see what we mean.

Amax: Supporting Nvidia Tesla K80 In New Servers

Amax, Fremont, Calif., said at SC14 that it plans to offer the new Nvidia Tesla K80 family of GPU accelerators in its ClusterMax GPU cluster and ServMax GPU server platforms. The company estimates the Tesla K80 dual-GPU accelerator can offer 10 times higher performance than today’s fastest CPUs, making them useful for complex analytics and computation-intense scientific computing applications.

For example, Amax's ClusterMax SuperG GPU cluster can be configured with up to 539,136 GPU cores, over 604 teraflops of single-precision and over 201 teraflops of double-precision performance, and 2,592 GB of GDDR5 GPU memory per 42U cluster.

ATTO: High-Speed Storage Controller

ATTO Technology, an Amherst, N.Y.-based developer of storage and network connectivity, showed a new family of high-performance components, including the FibreBridge 7500 storage controller and the 16-Gbit Celerity Fibre Channel host-bus adapters (shown). To highlight the high IOPS possible from the technology, ATTO demonstrated the two connected to an industry-standard JBOD configured with HGST’s enterprise-class Ultrastar 12-Gbps SAS SSDs.

The FibreBridge provides up to 1.47 million IOPs with dual controllers for use in building out tiered high-performance SSD and high-capacity HDD solutions with up to 2.4 petabytes of capacity using HGST's upcoming 10-TB drives.

DataDirect Networks: Lustre Storage For HPC And Big Data

DataDirect Networks (DDN), Santa Clara, Calif., used SC14 to launch its latest generation EXAScaler appliance, a Lustre-based storage solution for high-performance computing and big data.

Targeting commercial high-performance computing enterprises in such areas as oil and gas, financial services and manufacturing, the new Lustre-based EXAScaler appliances scale to tens of thousands of clients and petabytes of storage, and support dual-rail InfiniBand. DDN said they offer a 25 percent performance boost over the previous generation, and include InfiniBand optimization, self-encrypting drive support and integrated Lustre support for Apache Hadoop.

Dell: GPU-Dense HPC Rack Server

Dell's new PowerEdge C4130 accelerator-optimized server is a GPU-dense, flexible high-performance computing rack server. Dell claims it to be the only 1U Intel Xeon E5-2600v3-based server to support four GPUs, giving it up to 33 percent better accelerator density and a higher PCIe GPU accelerator-per-processor per rack U ratio than its competitors. That, Dell said, gives them up to 7.2 Teraflops on a single 1U server and 4.17 Gigaflops per watt.

They include up to four Intel Xeon Phi accelerators or NVIDIA Tesla GPUs, up to two Intel Xeon E5-2600v3 processors, 16 DIMMS of DDR4 memory and up to two 1.8-inch SATA SSDs.

Dot Hill: Hybrid Flash Storage Array

Dot Hill Systems unveiled the high-density Dot Hill Ultra56 AssuredSAN hybrid flash storage array, supporting both NEBS Level 3 carrier and MIL-STD government requirements and up to 1.34 petabytes of raw capacity. The Ultra56 AssuredSAN includes dual-active, redundant controllers for up to 6,400 MBps of sustained read performance.

The Ultra56 is targeted at telecom carrier, service provider and government environments, as well as at such applications as big data, cloud, high-performance computing, wireless and cellular infrastructure. They are slated to ship in January with a starting price of about $70,000 for 336 TB capacity in a 4U of rack space.

Huawei: HPC For Cloud Data Centers

Shenzhen, China-based Huawei used SC14 to launch its FusionServer X6800, which features integrated high-density architecture, large storage capacity and reliable management. The X6800 offers a choice of server nodes to meet elastic configuration requirements of computing, storage and I/O resources. It also supports simplified system management, and efficient operation and maintenance for such applications as cloud computing, big data and high-performance computing, including in cloud data centers.

Mellanox: InfiniBand All The Way

Mellanox Technologies showed its new ConnectX-4 single-port and dual-port Virtual Protocol Interconnect (VPI) adapter, which supports both InfiniBand and Ethernet protocols from 10 Gbps to 100 Gbps on nearly all CPU architectures, including x86, GPUs, POWER, ARM, FPGAs and more.

The ConnectX-4 doubles the throughput of the previous generation, and delivers scalable, consistent performance and low 0.7-microsecond latency for HPC, cloud, Web 2.0 and enterprise storage applications. It supports the new RoCE v2 (RDMA) specification and multiple overlay networks technologies, including NVGRE (Network Virtualization using GRE), VXLAN (Virtual Extensible LAN), GENEVE (Generic Network Virtualization Encapsulation) and MPLS (Multi-Protocol Label Switching).

Nvidia: New High-Performance Tesla K80 Accelerator

NVIDIA at SC14 unveiled its latest NVIDIA Tesla Accelerated Computing Platform, the Tesla K80 dual-GPU accelerator, designed for a wide range of machine learning, data analytics, scientific and HPC applications.

The Tesla K80 dual-GPU accelerator delivers nearly twice the performance and double the memory bandwidth of its predecessor, the Tesla K40 GPU accelerator. It delivers up to 8.74 teraflops single-precision and up to 2.91 teraflops double-precision peak floating point performance. Features include 26 GB of GDDR5 memory, 480-GBps memory bandwidth, 4,992 CUDS parallel processing cores and the ability to let GPU threads dynamically spawn new threads.

Open-E: ZFS-Based Enterprise Storage

Marietta, Ga.-based storage software developer Open-E used SC14 to show Open-E JovianDSS, the company’s new ZFS-based enterprise storage software with high-performance and unlimited scalability. Open-E's JovianDSS targets enterprise storage environments with such capabilities as thin provisioning, tiered caching and deduplication, all running on the customer's choice of hardware.

Open-E also showed its storage operating system, DSS V7, in combination with the Avago Syncro controllers. The combination allows Open-E's channel partners to build Fibre Channel or iSCSI active-active dual-controller SANs based on white-box servers with up to half a petabyte of capacity in a high-availability configuration.

SanDisk: Highlighting Its Partners

SanDisk, the Milpitas, Calif.-based SSD and memory developer, used SC14 to highlight its work with several partners in the supercomputing and HPC market.

For instance, the University of Michigan and University of Victoria are working with the CERN Laboratory in Geneva to use SanDisk's Fusion ioMemory solutions to transfer 170-petabyte data sets from the Large Hadron Collider to research locations around the world. Meanwhile, Atlantis Computing demonstrated a hyper-converged system built with SanDisk ULLtraDIMM SSDs and Lenovo System x Servers, Boston Limited showed its new Quattro 1280-6 server featuring SanDisk ULLtraDIMM SSDs, and Supermicro showed several SanDisk flash solutions including the ULLtraDIMM SSD in its HPC servers.

Seagate: Hadoop For HPC Environments

While Seagate is best known for hard drives, the company is expanding its storage reach in several ways, including its new Seagate Hadoop Workflow Accelerator. This is a set of Hadoop tools, services and support for HPC environments to help optimize big data analytics and maximize the performance of Hadoop ecosystem deployments.

Hadoop Workflow Accelerator supports Apache-based Hadoop distributions. It includes the Hadoop on Lustre Connector to let both Hadoop and HPC Lustre clusters use the same data without having to move it between file systems or storage devices. These benefits extend to Seagate ClusterStor, Seagate’s scale-out storage system designed for big data analysis.

SysFera: Custom Simulation Results Over Web Browsers

Lyon, France-based SysFera's latest SysFera-DS version 5.0 HPC and cloud-management software offers immediate visualization of simulation results over custom web pages without downloading and opening results within applications. This, along with an enhanced web interface, make the software more user-friendly, while providing tighter application integration to let engineers and researchers run jobs more quickly and preview results faster, the company said.

SysFera-DS software lets results be previewed via a web browser to help eliminate file search and download time, regardless of the image type their simulation creates for output.