Here's a collection of highlights, selected totally subjectively, from this week's HPC news stream as reported at insideHPC.com and HPCwire.
>>10 words and a link
Fibre Channel over Ethernet standard proposed to ANSI's T11 Committee,
http://www-03.ibm.com/press/us/en/pressrelease/21331.wss
The Register reports Power6 slips from mid-2007 to “to come,”
http://www.theregister.co.uk/2007/04/05/ibm_power6_tocome/
Sun opens registration for its CommunityOne event, May 7 in San Francisco,
http://www.sun.com/aboutsun/pr/2007-04/sunflash.20070409.1.xml
Iowa State University installs 100 million pixel virtual reality system,
http://www.fakespace.com/press/PR04-10-07ISU_C6.htm
UK-based Very-PC considers cooling computers in tubs of oil,
http://www.newscientisttech.com/article/dn11573-chilled-oil-bath-may-cool-hot-processors-.html
Sun donates parts of its storage solution to open source,
http://tinyurl.com/2eruay
>>Bo Ewald back at SGI
Seven days after it was announced that Bo Ewald was leaving Linux Networx as CEO, SGI announced it was sacking Dennis McKenna and replacing him with … Ewald. Bo, who was previously COO of SGI as well as President and COO of Cray Research, brings a management expertise gained at a series of CEO positions back into the company as well as a little HPC history.
But the HPC landscape for SGI has changed a lot since he was last with the company. For the first time in many years we have several new startups (companies like Liquid Computing, Lightfleet, and SiCortex) trying to bring new technology to the industry. Companies like Sun and IBM are trying to disrupt the delivery model by pushing dollars/cycle and SaaS into HPC. And commodity pressures are snatching all but the thinnest of margins out of the ledger books.
Once its graphics leadership dwindled, SGI traded very effectively on its dominating shared memory technology. But with lower cost alternatives everywhere you look, SGI has to find a way to capitalize on its CC-NUMA technology (and leadership) while remaining relevant to customers not driven by the need for shared memory. Ewald certainly has his work cut out for him.
>>Sony considers creating commercial PlayStation grid
Nicholas Carr's Rough Type has a recent piece reporting on Sony's move to get itself a little attention in the money for cycles crowd. From Nick's article, “Sony is considering tying together gamers' PlayStation 3 consoles into a global supercomputing grid that could be used for commercial applications, reports the Financial Times today. Sony has already teamed up with Stanford in the nonprofit Folding@Home initiative, in which PS3 users donate the spare cycles of their machines to analyze protein cells. Some 12,000 people have signed up for Folding@Home.”
The PS3 console is based on the Cell processor, of course, which has a lot of potential but is still pretty unusable from a programmer's point of view. It appears that Sony has some sort of value stream in mind for itself, selling cycles from the “PlayGrid” to pharmaceutical companies, for example, that need HPC but don't have their own systems.
It seems that this will put Sony in the position of having to provide value to its product owners in exchange for making enough systems available often enough for there to be a “grid” to use.
Smells like PR to me, but I suppose it might work. Until the Cell processor presents a more robust development environment, though, adoption is going to be a tough move for users to make. Organizations looking to do supercomputing on the cheap using something like PlayGrid aren't likely going to want to fork over big dollars for specialized HPC expertise to port and tune their codes for Cell processors.
Full article at Rough Type, http://www.roughtype.com/archives/2007/04/sony_may_launch.php.
>>IBM pushes chip stacking technology into production
IBM announced this week that chips using its new 3D stacking technology will be generally available by 2008.
IBM says that it expects to use the third dimension to reduce the distance that signals have to travel by 1,000 times. Early targets for the technology include improving the power efficiency of wireless products by up to 40 percent and uniform power delivery to all parts of manycore chips.
IBM will also apply the technology in stacking high-performance chips (processor-on-processor or memory-on-processor) in its Blue Gene supercomputer, “to fundamentally change the way memory communicates with a microprocessor, by significantly enhancing the data flow between microprocessor and memory.” Full release at http://www-03.ibm.com/press/us/en/pressrelease/21350.wss.
—–
John West summarizes the headlines in HPC every day at insideHPC.com, and writes on leadership and career issues for technology professionals at InfoWorld and on his own blog at onlytraitofaleader.com. You can contact him at [email protected].