Here’s a collection of highlights, selected totally subjectively, from this week’s HPC news stream.
>>10 words and a link
What about all those cores? HPC enters the mainstream;
http://www.itpro.co.uk/features/123256/get-ready-for-hpc-in-the-mainstream.html
BNL researchers use Sun’s Network.com for quark-gluon plasma;
http://blogs.sun.com/HPC/entry/bnl_uses_sun_grid_for
Independent study says AMD more power efficient than Intel;
http://www.hpcwire.com/hpc/1756595.html
AMD continues blame death spiral rather than creating customer value;
http://www.dailytech.com/article.aspx?newsid=8608
Hawking’s COSMOS research consortium buys new SGI Altix;
http://biz.yahoo.com/prnews/070905/aqw112.html?.v=16
IBD re-running retrospective piece on Seymour Cray’s career;
http://biz.yahoo.com/ibd/070904/lands02.html?.v=1
>>Intel launches new quadcore Xeon 7300 series
Intel Corporation has unveiled the industry’s first quad-core processors specifically designed for multi-processor (MP) servers running applications requiring uncompromised performance, reliability and scalability.
The press machine over at Intel is, well, enthusiastic. But this is news.
There are six new processors in the 7300 line, and today’s announcement marks the completion of the company’s transition to the new Core microarchitecture. 7300 Xeons top out at 2.93 GHz (130 watts) at the high end, and run all the way down to an eco-friendly 1.86 GHz model (50 watts). In addition to having twice the cores the 7300 also supports four times the memory of Intel’s previous MP products.
More at Intel, http://www.intel.com/pressroom/archive/releases/20070905comp.htm?cid=rss-90004-c1-182325.
>>HP, Verari add support for Intel quadcore
Verari announced this week that they’ve added support for the Intel Xeon multi-processor (MP) server platform featuring Quad-Core Intel Xeon Processors (the 7300 series).
From the company’s release (http://www.verari.com/news/archive/PR090507.asp):
“Verari Systems’ BladeRack 2 is a powerful and dense blade server platform built for virtualization and consolidation,” said David B. Wright, CEO and Chairman of the Board of Verari Systems. “The addition of the Intel Xeon MP Platform to our family of highly scalable systems provides our customers with the type of enterprise reliability not available elsewhere and will further increase the IT efficiency and system utilization for VMware customers.”
And HP expanded its quadcore offerings this week as well with the addition of the same Xeon 7300s to the ProLiant line (http://biz.yahoo.com/bw/070906/20070905006548.html?.v=1):
The rack-based HP ProLiant DL580 G5 server and the HP ProLiant BL680c G5, HP’s first four processor (4P) quad-core server blade, offer increased performance with double the number of processor cores. In addition, the DL580 G5 has double the memory capacity of its predecessor.
Verari’s offering will be available in Q4 according to the company; you can pick yourself up a shiny new ProLiant from HP today.
>>The Facebook of HPC
ClearSpeed Technology announced this week that they’ve licensed their accelerator technology (the same magic sauce that pumped up Japan’s TSUBAME super and made ClearSpeed the darling of SC06) to BAE Systems for use in BAE’s satellites.
What does this have to do with HPC or Facebook? Hang with me.
I got to talk to ClearSpeed CEO Tom Beese last week (thanks Christin!) and we had a very lively and wide ranging discussion. It turns out that the technology ClearSpeed is licensing to BAE is the same stuff they sell for plugging in to supercomputers. While BAE is understandably close-lipped about what happens inside the case of these billion dollar devices, word is that they’ll be used for in-flight data processing tasks, not satellite control functions.
ClearSpeed chose to license the technology rather than simply producing boards for BAE because the harsh operating conditions in space dictate production methods for electronic components that are not in line with ClearSpeed’s primary business. You can read more at http://www.hpcwire.com/hpc/1760618.html.
What’s really interesting about this story, though, is the peek it gives us into the future. Various pundits and people paid to have opinions are predicting anywhere from 80- to 128-core chips by the end of this decade. The price of a computation engine is falling dramatically, and we can predict that eventually cores will be “free” (that’s free as in cell phone, not free as in beer).
At that point, which is likely not too far off, there will be no undoing the commoditization of HPC, and even IBM might finally give up the ghost and bury its Power line. This doesn’t mean we’ll have to give up specialized processing though, and the ClearSpeed deal points the way.
Although vendors have been talking about constellations of heterogeneous processors for some time, it’s not been clear how the economics of the industry would allow a robust market to develop for these products at reasonable prices. Ganging components together in ways specific to HPC will always be too expensive relative to the dynamics of commodity product pricing. And niche companies developing coprocessors (or accelerators or whatever) for HPC would likely never be large enough to become reliably innovative over the long run.
But as cores become free, opportunities open up for processing in entirely new markets and products, and existing markets have the opportunity to enhance the function of processing in their products. With both major chipmakers adopting non-bus based architectures and opening their interconnects to third party hardware makers (as in AMD’s Torrenza initiative) there is a clear opportunity for accelerator companies to turn themselves into platforms for specialized computing.
It works like this. ClearSpeed (and companies like them) develop an acceleration vehicle with specific customization points. The accelerator developed for HPC transforms into a real-time performance analytics engine in Ford’s next generation vehicle by simply swapping out or reprogramming an ASIC. These devices can stay relatively simple because of the wealth of cores lying around that can be harnessed to support their functions, and they will remain relatively inexpensive because there is a demand in so many markets. And the companies making the devices will become large and diverse enough to survive mistakes in any one market, and they’ll have a shot at innovating reliably over the long term.
It may be a while before software can take advantage of all the power we’re creating, but a hardware solution may be just around the corner.
—–
John West summarizes the headlines in HPC news each week for HPCwire. You can reach him at john.e.west (at) gmail.com.