February 11, 2010
Lately, it seems like I've been spilling a lot of digital ink on two subjects: cloud computing and microprocessors. This week was no exception, as highlighted by our coverage of the International Solid State Circuits Conference (ISSCC) and SGI's Cyclone cloud launch.
Microprocessors and clouds bracket two ends of HPC: the lowest level hardware at the heart of high performance computing and the most abstract layer that can deliver it, respectively. Why the recent focus? Frankly, sometimes it's easier to report on the edges of the HPC ecosystem than having to delve into all the messy software and hardware layers in between. But the other aspect to this is that these two technologies, perhaps more than any others, promise to expand access to high performance computing.
Obviously this is already the case for microprocessors. The well-known Moore's Law has stopped delivering faster processors, but is still providing us with higher price-performance by way of multicore. In the recent processor parade at ISSCC, we noted that IBM's 8-core POWER7 quadrupled its core count compared to the older POWER6. Intel added two cores apiece to its Xeon and Itanium line, with the upcoming Westmere chips at six cores and the new "Tukwila" Itanium at four. Dual-core CPUs now seem quaint by comparison.
There's plenty of circumstantial evidence that cheaper processors, and computing hardware in general, are expanding HPC usage globally. Just this week, we had announcements of new HPC systems at Nanyang Technological University in Singapore, Coventry University in the UK, the University of Melbourne in Australia, the University of Duisburg-Essen in Germany and the Institute of Statistical Mathematics in Japan. Some of these are first-time HPC deployments, and undoubtedly can be attributed to the falling cost of terascale-level computing. Serious supercomputing -- what used to cost a million dollars or more -- can now be had for hundreds or even tens of thousands of dollars.
Note that the aforementioned POWER7 chip is reported to deliver more than 250 gigaflops, and although IBM hasn't gone public with pricing on the four announced systems that will house these multicore wonders, they will likely be no more expensive than their POWER6 progenitors. Heck, I just bought a quad-core PC for around $500, and while that system doesn't pretend to be an HPC workstation, if I had added a late model GPU to it for a couple hundred more, I'd easily have a desktop teraflopper.
But my first personal encounter with HPC is just as likely to occur via the cloud, and I wouldn't be surprised if that was the volume path into HPC for many users. This model seems especially appropriate for organizations that make use of high performance computing in just one segment of their workflow, such as you might find in biotech and manufacturing. I liken it to the way I used to rent a minivan for a week every July for our family's Lake Tahoe vacation so I could travel with my wife, two kids, and three dogs in relative comfort. The rest of year, we just needed our high gas mileage subcompacts to commute to work and do other local travel.
Likewise with high-end computing: why build a tricked-out datacenter and supercomputing infrastructure if you only use it part-time? And now, with the new obsession with energy efficiency and squeezing as much computing as possible from IT budgets, the renting of HPC cycles and software never looked so good.
Of course, the cost of renting something always tends to be higher than owning it over the long-term. But in the IT world, with such a rapid turnover of hardware, and to a lesser extent, software, there's little motivation to actually own something for more than a few years. In fact, from a business perspective, the faster the technology moves, the more incentive there is to rent it.
Don't get me wrong. I'm not predicting cloud computing will replace system ownership, for HPC users or anyone else. Even in the auto industry, we have three relatively healthy purchase models: short-term renting, longer term leasing, and ownership. I suspect the computing business will follow a similar path.
Posted by Michael Feldman - February 11, 2010 @ 6:07 PM, Pacific Standard Time
Michael Feldman is the editor of HPCwire.
No Recent Blog Comments
10/30/2013 | Cray, DDN, Mellanox, NetApp, ScaleMP, Supermicro, Xyratex | Creating data is easy… the challenge is getting it to the right place to make use of it. This paper discusses fresh solutions that can directly increase I/O efficiency, and the applications of these solutions to current, and new technology infrastructures.
10/01/2013 | IBM | A new trend is developing in the HPC space that is also affecting enterprise computing productivity with the arrival of “ultra-dense” hyper-scale servers.
Ken Claffey, SVP and General Manager at Xyratex, presents ClusterStor at the Vendor Showdown at ISC13 in Leipzig, Germany.
Join HPCwire Editor Nicole Hemsoth and Dr. David Bader from Georgia Tech as they take center stage on opening night at Atlanta's first Big Data Kick Off Week, filmed in front of a live audience. Nicole and David look at the evolution of HPC, today's big data challenges, discuss real world solutions, and reveal their predictions. Exactly what does the future holds for HPC?