facebook rss twitter

Intel talks about the importance of high-performance computing

by Pete Mason on 25 November 2010, 15:34

Tags: Intel (NASDAQ:INTC)

Quick Link: HEXUS.net/qa3bu

Add to My Vault: x

Many, many cores

We may not have been able to attend the Supercomputing Conference 2010 in New Orleans last week, but Intel helped us get caught up by showing us what it was doing in the world of high-performance computing right now, and especially how it was helping research in the UK.

The main message was that HPC is incredibly important, especially in this country, and is continuing to grow as more disciplines realise its potential benefits. In the UK, this means playing to our strengths and using computers in finance, engineering, meteorology - apparently the UK is a world leader in predicting the weather...who knew - and most critically, scientific research. From its roots in physics and computational chemistry, high-performance computing is now becoming critical to gene-sequencing, medical imaging, understanding how diseases work and modelling how systems in the body behave.

The problem with a lot of these sorts of calculation is that they're massively parallel, and so ill-suited to the more traditional multi-core processors that are designed to just do a few things very quickly. This is where Intel's many integrated core (MIC) architecture comes in. The company has been developing this sort of technology for well over a decade in some form or another, and it's finally starting to reach a point where it's ready to be released.

Knights Ferry - which we've seen several times before - is in fact only a software development platform, and not an actual product. Nonetheless, it gives a pretty good idea of what Intel is aiming for. The current board packs 32 vector IA cores into a single chip running at 1.2GHz with either 1 or 2GB of GDDR5 connected by a high-speed ring-bus type interconnect. This might sound a little measly compared to a top of the line GPU, but you can bet that each of these x86 cores will be a lot more capable than a single graphics shader. Not only can each one handle four simultaneous threads - for a total of 128 threads-per-chip - but we're reliably informed that each processing core bears a stunning resemblance to the company's Atom CPUs. They're also specially tuned for massively parallel high-performance workloads.