Data Centre

HPC

Alan Turing Institute bags a Cray Urika-GX to crunch numbers for next-gen tech boffinry

But can it run Crysis?

By Chris Mellor

6 SHARE

Cray is supplying an Urika-GX analytics machine to the UK’s Alan Turing Institute for data science research.

The supercomputer appliance – the GX in the model name indicating it has a graph engine – is an x86-based monster with OpenStack, Apache Mesos, Spark, and Hadoop installed. Its CPUs are 18-core Intel Xeon E5-2600 v4s. These Cray beasts can pack up to 1,728 cores across up to 48 nodes (with two processors per node) linked by an Aries interconnect.

It can take up to 22TB of memory, and storage is provided by up to 176TB of Intel P3700 PCIe SSDs. You can build these out of 16, 32 or 48 nodes, and the system is installed in a standard rack.

How much do these Urika’s cost? An ESG study for Cray said the year-one cost for a 48-node system, looking at capital and operational expenditure, was $1,056,260, and the three-year TCO was $1,754,680.

The main lure of the Urika-GX is its graph engine software. This can be used to discover relationships between objects in a graph. The graph here is not a chart but a mathematical structure modeling relationships and processes in an abstract way. An object or node in the graph is related to or connected to other nodes by “edges” or lines and the strength of the relationship can be given a value or weight.

A Urika-GX rack

Such a graph can be analysed to find connected objects, the shortest path between a set of objects, how certain objects are clustered together, and that sort of stuff. The more objects and edges there are, the bigger such analyses get, massively so.

Nvidia suggests one such analysis could be used for parsing speech to determine what is the most likely sequence of words that matches a given set of sounds. There are a huge set of other possible applications, and data science research looks at how information can be organized, what analyses can be done, and how they can be done.

The Alan Turing Institute is the UK's national data science institute, and is hosted at the University of Edinburgh, specifically in the Edinburgh Parallel Computing Centre (EPCC). Researchers there will use the Urika-GX to develop applications in several areas including engineering and technology, defense and security, smart cities, financial services, and life sciences.

Let’s hope they get their million-plus taxpayer quids’ worth of use out of Urika. ®

Sign up to our NewsletterGet IT in your inbox daily

6 Comments

More from The Register

Hard to imagine Google, Facebook building AI without (checks notes) Dell EMC's Data Science Provisioning Portal

If you want to do some ML, and you've got a fat budget, they've got some tech to sell you

Boffins are building an open-source secure enclave on RISC-V

Open source trusted execution component expected this fall

SUSE punts SES v5.5 out door, says storage is going software-defined and open source

Don't mind us, just having a Linux moment

Open-source software supply chain vulns have doubled in 12 months

Hackers 'mainlining' vulns into projects – report

Redis has a license to kill: Open-source database maker takes some code proprietary

It's 'cuz cloud giants use tools without giving anything back

Open source community crams itself into big tent

Can't we just get along? At a sunny California inn with hors d'oeuvres, most definitely

Open Source Security hit with bill for defamation claim

Judge okays $260K in defense costs to Bruce Perens and lawyers under anti-SLAPP

Facebook's new always-listening home appliance kit Portal doesn't do Facebook

Trust us, pleads the Zuck

I got 257 problems, and they're all open source: Report shines light on Wild West of software

It's like a jungle sometimes, it makes me wonder how I keep from going under

Finally: Historic Eudora email code goes open source

'Member that innocent, pre-Zuckerberg time?