If you believe that competition is a good thing, then the US Department of Energy is going about the problem of ‘How best to construct tomorrow’s super computers’ in the right way. KitGuru dons flip-flops and heads off to consider the impact of PetaFlops.
While PC labs across the world get very excited about 3DMark scores and the like, serious uber-geeks need to get their fix from Top500. This is a site that’s dedicated to letting everyone who the biggest of big dogs is, when it comes to the ultimate willy-waving competition.
At the desktop/laptop level, people like to have more than one core and – toward the top end of AMD and Intel’s ranges – you can have 8 cores or 6 cores with 2 threads each (respectively). Performance levels are measured here in TeraFlops – or trillions of mini-number-crunching things happening every second. The advent of huge parallelism (plus SLi/CrossFire) in graphics cards kicked this era off around 6-7 years ago.
Come to the Top500 and you need to add ‘at least 3 more zeroes’ to the levels of performance.
Welcome to the land where we’re measuring how many ‘Quadrillion’ operations can be performance a second. When you hit 1,000 TeraFlops a second, you’re delivering a PetaFlop. Until recently, the record was held by the Japanese at just over 10 PetaFlops, but now that crown has swung back to IBM with a system called Sequoia (an IBM BlueGene/Q system installed at the Department of Energy’s Lawrence Livermore National Laboratory).
It uses over 1.5 million cores to deliver over 16 PetaFlops. In this market, Intel leads AMD and IBM for the processors used (76.8%, 12.6% and 11.6% respectively) – while more than 10% of the systems in the Top 500 list now use graphics cards to augment their scores with nVidia (53 systems) leading Cell (2), AMD (2) and Intel (1).
While the number of applications that need the kind of power that a Top500 system can deliver might be limited, the US government is not above a little artificial stimulation.
To that end, it has made multi-million dollar bonuses available to Intel, nVidia and AMD – to get them all keen and interested in working on systems that could, potentially, deliver ExaFlop computing – something that’s 1,000 times more powerful that the scale we use today.
To give you an idea of how much power that is, you would only get 123 PetaFlops if you harnessed all 500 of the world’s most powerful machines today. With that number of ‘cores’ all working together, no wonder energy conservation is such a hot topic.
If you want to look at large numbers and ponder when they will arrive in the ‘desktop computing space’, then here’s a link.
Those involved in the global network projects, like [email protected], will be happy to know that they are now helping sustain close to 6 PetaFlops of performance (and you were the first computing project of any kind to pass the 1, 2, 3, 4 and 5 native PetaFlop marks. Well done to all involved.
Lastly, a big pat on the back to the KitGuru [email protected] team who are now ranked inside the world’s top 200 teams – which you can see here. When you consider that our team has only been going since the start of the site, around 2.5 years ago, you can see what an achievement this is. The ‘Rankings Update’ from Stanford shows teams down to 85,932… So WELL PLAYED KitGuru folders!
KitGuru says: As we discussed in an exclusive interview with Epic’s Mike Gamble last year, the battleground for the next 2 years will be in the mobile space. As more and more compute power is delivered into our hands, virtually for nothing, we’re going to also see a huge leap in what’s possible from large-scale networks. At the same time, distributed systems like [email protected] will also change and expand at a phenomenal rate. Overall, it’s this drive that will help shape tomorrow – so congratulations to the US government for spurring this kind of innovation on.