All the latest quantum computer articles

See the latest stories on quantum computing from eeNews Europe
Showing posts with label supercomputer. Show all posts
Showing posts with label supercomputer. Show all posts

Wednesday, March 09, 2016

IDT teams with ProDrive to reduce switch latency

By Nick Flaherty www.flaherty.co.uk

The latency of switching data is a key challenge for today's high speed networks, and IDT has teamed up with Prodrive Technologies to develop a RapidIO switch with 100ns latency for next generation cellular infrastructure.

The embedded devices for these switches will be used in networks running from 750 Gbit/s to 4.8 Tbit/s for next generation 5G cellular networks where the lower latency will open up applications for the Internet of Things (IoT), as well as Mobile Edge and High-Performance Computing applications.

The new products are the result of work the two companies have conducted under the charter of the Open Compute Project (OPC) HPC group, which IDT co-chairs. The switch appliances use IDT's RapidIO technology, which is designed in to many 4G LTE networks.

RapidIO switches are used to connect computing components in complex processing-intensive systems, including SoCs, DSPs, FPGAs, GPUs and custom ASICs. IDT has shipped more than 110 million ports of its 10-20 Gbps RapidIO switches for use in production networks. In addition to using the second-generation switches, the new appliance portfolio will deploy IDT’s recently announced RXS family of RapidIO switches, which enable ports running at up to 50 Gbps with less than 1.2 W per port and 100ns latency.

"The single millisecond round trip latency requirements of 5G infrastructure will place considerable demands on all computing elements in the network, and these new RapidIO-based switching appliances deliver a leading combination of latency and energy efficiency,” said Sailesh Chittipeddi, IDT’s CTO and vice president of global operations. “This collaboration withProdrive Technologies helps meet the needs of 5G edge computing and C-RAN in the telecommunications arena, as well as high-performance computing and data centre analytics.”

Initial systems leveraging switch appliances with 38 ports operating at 20 Gbps have been deployed in wireless C-RAN and data analytics systems. Additional products, expected in the latter half of 2016, use IDT’s 50 Gbps-per-port RXS switch silicon portfolio with 19 inch rack-mountable systems.

“The collaboration with IDT has been instrumental in winning computing and telco customers using our RapidIO- based servers and switch appliances,” said Pieter Janssen, CEO of Prodrive Technologies. “By combining IDT’s interconnect and market channels with the Prodrive design and manufacturing in the EU, we have already deployed 10-20Gbps versions of the switch portfolio in tier 1 telco C-RAN programs as well as top-tier European HPC-oriented data centres. The switch appliances, complete with Prodrive’s RapidIO-connected FPGA/Power/x86/ARM servers, are also attractive for financial trading applications.”

The switch appliances feature scalable modularity with 32 to 96 ports supporting clustering at rack or multi-rack scale for target applications supporting heterogeneous processing with CPUs and accelerators. Multiple switch appliances can be connected in a leaf and spine network leveraging the RapidIO protocol to connect up to 64K processing nodes in a single network. Individual ports can be configured to run at a variety of speeds, including 1, 2.5, 5, 10, 20, 40 and 50 Gbps, and thus tuned to optimize overall system power consumption, depending on application workload.

"Taking an open and collaborative approach to developing solutions for Open Compute and OpenPOWER is accelerating and strengthening the development efforts of our broader open community," said Calista Redmond, president of the OpenPOWER Foundation. "IDT is at the forefront as OpenPOWER grows capabilities into edge computing across system and switch resources."



The switch appliance products are available directly from Prodrive Technologies: http://prodrive-technologies.com/rapidio-gen2-38-ports-top-rack-switch-box/. The 20 Gbps-per-port versions are available now.  Lead customers can contact Prodrive for customized 50 Gbps-per-port solutions featuring IDT’s RXS RapidIO switches.


Tuesday, June 01, 2010

SGI puts petaflop hybrid computing in a single cabinet

By Nick Flaherty www.flaherty.co.uk


Scalable Mixed Technology Uses Open Architecture
 
SGI has developed a breakthrough hybrid computing platform that will deliver a petaflop of performance within a single cabinet using technology from NVIDIA, ATI and Tilera. The platform uses an open software architecture and will scale from tens to hundreds of petaflops, enabling users to tackle the most demanding technical compute requirements.
“We are excited to announce this radically differentiated server technology platform,” said Dr. Eng Lim Goh, senior vice president and chief technology officer at SGI. “Our innovative technology will enable users to attain new levels of scalability and speed with groundbreaking performance capabilities, culminating in a petaflop in a cabinet.”
The new platform was designed to drive performance to specifically address the growing science and engineering technical markets that rely on high-end software to achieve rapid results. It offers GPU processing capabilities from NVIDIA and ATI, as well as accelerator-based technology from Tilera, and other peripheral component interconnect express (PCIe) based solutions. The technology will be seen in SGI products by the end of 2010.

“SGI shows its R&D strength again,” said Steve Conway, IDC research vice president for high performance computing. “Following closely on the heels of the Altix UV series, SGI has introduced a highly dense, scalable technology designed for strong sustained performance on very demanding HPC applications.”

Reblog this post [with Zemanta]

Tuesday, November 17, 2009

Supercomputing moves into harsh environments


GE Fanuc Intelligent Platforms, part of GE Enterprise Solutions, has developed a rugged processing board with the performance of a supercomputer.
The OpenVPX-compatible GRA111 high performance graphics board is one of the first commercially-available products to feature Nvidia’s newly-announced GT 240 GPU (graphics processing unit) and is also the first rugged implementation of a GPU using Nvidia's CUDA (Compute Unified, Device Architecture). The board is the first result of a relationship with Nvidia announced on October 8th.
The GRA111 is the first member of GE Fanuc’s planned family of CUDA-enabled products that are expected to revolutionize military and aerospace embedded computing by bringing a supercomputing capability to applications and environments that require leading edge processing such as radar, signals intelligence and video surveillance and interpretation. For example, radar systems can run up to 15 times faster.
“The enormous potential of the GPGPU – the general purpose graphics processing unit – with its massively parallel computing architecture is just beginning to become apparent to developers in the military and aerospace field,” said Rob McKeel, Vice President, Embedded Systems at GE Fanuc Intelligent Platforms. “What we are announcing today is an exciting first: a new compact, lightweight platform that can deliver exceptional levels of computing performance in demanding applications and challenging environments.”

NVIDIA’s CUDA is a general purpose parallel computing architecture that takes advantage of the parallel compute engine in NVIDIA GPUs to solve many computational problems in a fraction of the time required on a general purpose CPU. It includes the CUDA Instruction Set Architecture (ISA) and the parallel compute engine of the GPU.
CUDA has been shown to deliver a 100x increase in speed in a broad range of applications that lend themselves to parallel computing. Early tests at GE Fanuc have shown that military applications can benefit from similar increases in speed. A major prime contractor in the military/aerospace industry has evaluated the CUDA architecture in a radar system, and found that performance improvement of 15x is achievable with minimal reprogramming effort.
The 3U format GRA111 derives much of its performance from the 96 cores, 128-bit memory interface, 1 GByte of DDR3 video memory, 16-lane PCI Express Gen 2 interface, 540MHz graphics clock and 1.3GHz processor clock of the GPU.
Optionally available as an LRM (Line Replaceable Module) in accordance with the VPX-REDI (VITA 48) standard and with a rich set of I/O, the GRA111 can also be applied to many of the most common graphics and video applications. Dual, independent channels mean that it is capable of driving RGB analog component video, digital DVI 1.0, and HDMI standards. In addition, the GRA111’s video input capability allows integration of sensor data using RS170, NTSC or PAL video formats.

Monday, November 02, 2009

China supercomputer breaks performance barrier


China's National University of Defense Technology (NUDT) has developed a supercomputer that breaks the petaflop barrier, putting China in the same league as the US and Europe for high performance computing.
The supercomputer, called "Tianhe", meaning Milky Way, is based in Changsha, capital of central China's Hunan Province.
The 155-ton system, with 103 refrigerator-like cabinets covering an area of about 1,000 square meters cost $88m and is expected to process seismic data for oil exploration, conduct bio-medical computing and help design aerospace vehicles. It uses 6,144 Intel CPUs and 5,120 AMD GPUs, and reaches a peak performance of 1.206 petaflops and 563.1 teraflops on the Linpack benchmark, making it currently the fourth most powerful machine on the planet.
China's Dawning Information Industry Company is attempting to build its own supercomputer that overcomes the petaflop barrier by 2010, and the NUDT is planning to add "hundreds or thousands of China-made CPUs" to the machine to improve its Linpack performance to over 800 teraflops.