Cray XC series

Cray XC series

Cray XC series is the supercomputers intended for work with the exacting workloads requiring steady performance in several petaFLOPS.

As a part of SK:

• high-performance network interconnect of Cray Aries
• Intel Xeon processors,
• Xeon Phi Intel coprocessors
• GPU NVIDIA Tesla accelerators.

Cray XC40, (2016)

The Cray XC50 supercomputer is provided with GPU accelerators and combines performance on the basis of the graphic processors PCIe Tesla P100 from NVIDIA with high-performance program networked environment of Cray Aries. A system shows peak capacity in 500 petaFLOPS. 1 petaFLOPS - one cabinet.

Cray XC40 offers a combination of interconnect of Cray Aries and the network topology Dragonfly. In composition multi-core processors, the integrated input-output acceleration options, the Cray operating system and a program environment. All this provides performance to 100 petaFLOPS.

The Cray XC40-AC model is equipped with air cooling. Has less dense cabinets, without the need for use of liquid heat carriers or additional air ducts. The reduced network topology promotes cost reduction.

2017: Availability in Azure cloud

The Microsoft corporation and the developer of the Cray supercomputers in October, 2017 agreed about providing computing powers of the Cray XC and Cray CS supercomputers in Microsoft Azure cloud. Under the terms of the agreement, the Cray supercomputers will be connected directly to Azure, integrated with Azure virtual machines and Azure Data Lake storage and also with artificial intelligence technologies and machine learning of Microsoft. Read more here.

2013: Cray XC30-AC

Cray XC30-AC is intended for the customers wishing to receive power of a system of the HPC level taking into account acceptable costs. Feature of this system: it is constructed on the platform of more expensive model.

The supercomputer is expected requirements of computing power from hundred to 1 thousand employees. As a part of a basic configuration one computing rack, maximum - six, with 1024 processors. Believe in the companies that Cray XC30-AC will begin to buy the companies entering the list of Fortune 1000, the small universities, the federal agencies and scientific centers.

Computing power of a system can be from 22 to 176 TFLOPS.

Management

The Cray Linux operating system manages the Cray XC30-AC computer, it is delivered with the tools for application development helping to use the computing power of a product.

Cooling

In Cray XC30-AC air cooling is applied. In one rack for cooling one fan which banishes an air flow on all height is used. In a system there are no fiber optic connections.

Computing module Cray XC30

Heritage

Developers turned on in composition the basic equipment rooms and program components borrowed more expensive XC-30 model. It belongs to processors: XC30-AC supports installation of Intel Xeon processors, Intel Xeon Phi coprocessors and Nvidia Tesla accelerators.

Network topology of the solution

Price

The cost of the solution Cray XC30-AC begins with a mark of $500 thousand and reaches $3 million.

The Cray XC30-AC computer system from four racks

Delivery and installation of Cray XC30 in Swiss Center for Scientific Computing

2012: Cray XC30

The Cray company announced in the fall of 2012 creation of the XC30 supercomputer which performance depending on a configuration can be up to 100 PetaFLOPS.

XC30 constructed on Intel Xeon processors can include to one million (!) computing cores. In the future the company is going to offer the market XC30 configuration in which Intel processors will work together with additional computing elements – Intel Phi coprocessors and AMD accelerators of Tesla.

The statement of the company means that Cray can create a system with a performance of 100 PetaFLOPS if on like that there is a buyer. Orders for XC30 already arrived – in particular, from the National supercomputer center of Switzerland, Energy research center of the USA and the Center of high-performance computing of the university of Stuttgart, but so far nobody ordered a system in the maximum configuration.

Cray XC30 can theoretically be the most powerful supercomputer of the planet, but practically a system in a configuration with a performance capable to apply for leadership in a top 500, is not created yet and is not even ordered yet.

Cray XC30 in battle for virtual leadership is neck and neck with kitayskimtianhe-2 which stated performance is also 100 PetaFLOPS and which also does not exist yet – works on its creation are only begun, are going to complete them in 2015.

The additional intrigue to a race is given by the fact that the Tianhe-1A system already became the most productive on the planet (from November, 2010 to June, 2011), and the Cray Jaguar system had a palm from November, 2008 to November, 2009 and, apparently, will get to the supercomputer of this company in November, 2012.

Let's remind that the rating of Torahs of 500 Supercomputing is officially updated twice a year – in June and in November. Therefore so far already started Cray Titan supercomputer (20 PetaFLOPS) did not get to it even and today the rating still heads IBM Seqoia (16.7 PetaFLOPS).

Let's notice that both systems can so never and not head rating. There are also other companies having competences for creation of systems with a top performance, for example, in rating supercomputers of IBM, Fujitsu, Hitachi and NEC were in the lead, at it there are also systems of Dell, Hewlett-Packard, SGI and other producers.

In the Cray supercomputers data exchange will accelerate four times

At design of the next generation of the supercomputers in Cray company aimed to increase considerably input-output speed on each node. In the Cray supercomputers — XC30 — will be used interconnections of new type under the name Aries and new topology of routing which together promise to increase internal capacity sharply.

"It is important to increase the input-output speed as the quicker there are processors, the more intensively there has to be a data exchange with them to support them in the loaded status — Barry Bolding, the marketing vice president of Cray explains. — The data exchange rate with a node grew, and possibilities of network should grow respectively".

Cray XC30 should leave at the beginning of the next year. It will be the first of a line of supercomputers XC which are going to be released in the next five years. The priority in the new architecture developed more than six years is given to increase in global capacity which Bolding describes as "capability of the machine to process heavy loadings in the conditions of intensive data exchange between all its processors".

Thanks to new interconnections each node can execute 120 million transactions of reading/record now per second whereas at the previous generation of the systems Cray, XK, this indicator was 30-40 million transactions.

Each node in a new system receives the separate block of interconnections Aries on 48 ports. The unique feature of such block is that it performs at the same time functions of the network adapter and router, providing to each node direct access to all other nodes of a system, explains to Bolding.

Cray sold to Intel the practices in the field of technology of interconnections, including the most part of intellectual property on Aries, in May for 140 million dollars. The bus PCIe-3 which is given preference in Intel forms a basis of Aries.

The architecture of the internal network XC connecting all nodes of a system is executed on new direct topology of Dragonfly. It needs less optical links and external switches of the top level are not necessary. Dragonfly is developed according to the results of the research conducted by specialists in networks of Stanford University together with engineers of Cray.

In addition to traditional users of supercomputers — the Pentagon and scientific institutions, Cray offers the new systems to the oil companies, producers of the large-size equipment and other big corporate customers who need powerful computing systems for researches.

The minimum cost of XC30 — 500 thousand dollars. Next year Cray will also offer smaller model for business applications, reported Bolding.

Many of XC30 technologies, including Aries, are partially developed on money of the grant awarded by Cray within the High Productivity Computing System program the Agency of perspective defense researches DARPA. "A task of this program — to create the next generation of highly productive supercomputer systems" — reported Bolding.

Each cabinet of XC30 provides high-speed performance 66 TFLOPS. And the maximum power of a system which can be constructed of such cabinets is more than 100 PFLOPS. The cabinet contains three chassis on 16 computer edges in everyone. Each edge is four nodes. In total in a system there can be about 92,544 nodes having 128 Gbytes of RAM of DDR3.

In the beginning in Cray for the new system will use the Intel Xeon processors E5-2600. And over time in XC it will be possible to install Intel Xeon Phi coprocessors and graphic Nvidia Tesla chips on the basis of architecture of the next generation of Kepler.

In supercomputers XC the new air-to-water scheme of cooling is applied, thanks to it the supercomputer does not need to be placed according to the scheme of alternation of cold and hot ranks.

The Cray Linux Environment software package which Linux kernel, the Lustre file system, a programming language with support of parallelization of Chapel and the scheduler of the SLURM tasks (Simple Linux Utility for Resource Management) enters will work on the new supercomputer.

Cray already sold six copies of XC30. Among buyers — the Swiss national supercomputer center, the Finnish IT scientific center, National computer center of the U.S. Department of Energy and the Center of high-performance computing of the Stuttgart university.