As published in HPCWire
Returning to ISC after a hiatus of several years and viewing the event from the vantage point of an industry analyst, the show appears to have made
a quantum leap in terms of the size and sophistication of the exhibition, and the degree and intensity of business activity. The exhibit hall,
while not on the scale of the Supercomputing event, included all the major suppliers to the industry, a large array of smaller middle-tier suppliers,
plus strong entries from the user community. Clearly, suppliers attended to do business and not simply “show the flag,” and generally reported
good results.
The event also struck me as being more European than in past years. There are several possible factors here: increased interest in HPC within western
Europe, expansion of eastern European economies and computing efforts, reduced travel budgets in the US, and so on.
One trend that was apparent at the ISC exhibition was the growing role of technology providers in the industry. Companies such as AMD, Intel, Mellanox,
LSI, Supermicro and STEC were prominent at the show, and from this observer’s perspective, they were doing good trade.
Intel’s New Processors
Intel announced its plans for a multicore coprocessor family: a current software development version codenamed Knights Ferry and a future production
version codenamed Knights Corner based on 22 nm process technology. This line of multicore processor boards is a direct challenge to NVIDIA’s CUDA-based
dominance of the HPC accelerator market. The primary keys to success/failure for this effort are: first, Intel’s ability to deliver its software
development environment for the Knights family, which the company indicates will be compatible with its current tools suite; and second, NVIDIA’s
ability to capture as much market and mind share for its GPU-based Tesla and Fermi product line before Intel can begin to deliver production-level
products. This should provide NVIDIA with about a two-year head start to meet its most significant competitive challenge to date. In the meantime
Intel will need to propagate Knights Ferry programs as widely as possible, and work to convince the market the wait for Knights Corner will be
worth it.
On the TOP500 List
I have a love/hate relationship with the TOP500 list.
On the love side, the computer science behind the list is excellent. The folks who maintain the list do a great job of explaining what the data means
and how it is relevant (although few people seem to listen). The list provides good data on high-end system technologies and historic trends, and,
like a top 25 college sports team list, it is fun to track, especially if your college is on the list.
On the hate side is that outside of the HPC technical community, the computer science part of the TOP500 is largely ignored, reducing one of the most
complex technologies and markets in the world to a few dozen statistics. This is like sending your child to medical school based on which institution
has the highest-rated basketball team, or like assuming that one can understand chemistry by examining the bottom rows of the periodic table. (But
I digress…) That said, Cray, with its long history in the TOP500 race, seemed philosophical about being upstaged by the number 2 system from
Dawning. There is little surprise that China with its program to develop its technical infrastructure appeared well up on the list. Perhaps more
important than gaining the number 2 slot was China’s placing a second system in the top 10 — an NUDT system in the 7th slot, which is indicative
of a general effort to provide computing capabilities to a broad group of scientists and engineers within the country. The odds-makers are favoring
China to gain the top slot on the list by the next publishing cycle or two.
It sometimes seems that entry into the TOP500 list is a right of passage for companies, institutions and countries.
The major winners from this year’s TOP500 competition may be the exascale system advocates. In addition to capturing headlines, China’s success
will certainly lead to political concerns, which may mean additional funding for next generation supercomputing efforts will open up. I am a great
believer in government support for IT technology and infrastructure in general, and HPC in particular. If the TOP500 list can be used to leverage
more resources into HPC programs, it may ultimately be a good thing. However, there is a real risk that such programs will be overly focused on
a single requirement while shorting requirements in other areas (e.g., applications software development, education, support for industrial users
particularly at the midrange and low end, and so on). I sometimes get the feeling that organizations are striving to produce the most powerful
locomotives in the world without training engineers, building the passenger and freight cars, train stations and assembly yards, or even adequate
rail lines.