As some of you have already noted, the most recent figures from IDC’s sweep of the HPC server market are in—and from the surface, they don’t suggest a stellar season ahead for supercomputing. However, when put into some broader context, particularly on the international scale with a few massive, surprise systems added to the mix, the big picture is anything but grim.
To recap quickly, the research group found that worldwide factory revenue for the HPC technical server market dipped 7.2%–knocking it to the $10.3 billion mark for the whole of 2013. Interestingly, this decline is based on the previous numbers from 2012, which showcased a record $11.1 billion. This was driven, in large part, by the power of a few versus the seismic forces of an entire market.
“The high end of the market has always been lumpy,” explained Steve Conway, Research Vice President in IDC’s technical computing division. “Ever since supercomputing has existed, there’s been a lot of variability quarter to quarter and year to year. It’s heavily driven by sales of the largest systems, which then throws the bell curve—just as it did here. He notes that even without those very few but massive sales, the high end is marching steadily forward. “If you don’t look at this market as a one-year thing, but rather as a trend line, it’s continuing to grow.”
Conway reminded us that 2012 was very strong but was powered by some anomalies—a few strong sales in particular. The biggest was the K system in Japan, which was $550 million alone. “This, coupled with the Tianhe-2 in China, provided a sudden injection of spending that hadn’t been predicted—and very likely won’t be repeated in the near future. In other words, the decline in these numbers was well-anticipated.
When taking a look at the server market across all segments, HPC, despite a decline in overall spending in the last year, is doing far better than its general business computing brethren. The average selling price of HPC systems is already high and continues to mount in comparison. The difference in IDC’s most recent server market figures is not a matter of selling price, it’s the number of units sold. While that might be lower, the general server market has remained flat over the last several years with around 1% CAGR versus the relatively steady 7% for HPC.
Conway says that there are many considerations that contribute to this disparity in server market numbers, including what customers require from their systems and of course, HPC’s outsider status on some trends that have cheapened commodity enterprise computing, namely virtualization and server consolidation.
Speaking of that difference between the HPC and general business computing markets, there has seemed to be a convergence underway. We’ve been tracking how an increasing number of enterprise users are looking to technical computing servers and tools, which could signal further growth in these IDC HPC numbers. But let’s reverse that for a moment—are HPC shops looking to the commodity, stripped down (think Open Compute type) servers that are storming the enterprise?
According to Conway, these two spaces—HPC and high-end enterprise—are merging, but uni-directionally. He says more commercial firms are now HPC servers but on the flipside, the HPC folks are not really “down shifting” to look to enterprise servers. “The market is getting a little complicated now because there are workloads where the stripped down servers make sense, but on the other end, there’s a higher expectation from other users that even more should be included in clusters.” He says that they’re watching the trend but predicting it is no easy task once we move out another two or three years.
Another fuzzy area is around how new movements, including Hadoop, could alter future IDC HPC server market numbers. At the core of this is the “big data” trend, a great deal of which can be classified as technical computing under IDC’s larger definition of HPC and what is thrown into the mix. With analytics at the center of this defined as technical computing, Hadoop cannot be ignored. But interestingly, despite the noise around Hadoop in commercial circles, the risk-averse business world is less likely to adopt Hadoop for production environments. Conway said that Hadoop use in HPC trumps it in enterprise with 29% of shops reporting they were already using it. While there are a number of ways it must be modified (swapping out the native file system and other efforts to boost performance) it’s striking that the most visible talk around Hadoop’s role in the world (solving big commercial problems one batch at a time) is from the one segment where the adoption is the slowest.
This all calls to mind the work being done at smaller and mid-sized houses—commercial and academic alike. The server market numbers were brightest for these segments, however, again, this is not just indicative of some massive market push, but rather other, more subtle forces. The most profound culprit for this growth is the recession (remember that?). These segments are bouncing back while large-scale HPC investments weren’t as affected since the wheels of those deals for big systems had been set in motion and were “too big to fail” in a sense. On the other hand, orders around $50,000 (for instance) were more easily put on ice or even cancelled during that uncertain time. They’re back now, reflecting that bounce, but also decent sales.
Among the vendors who stood out in the last year were HP (32.3% share), IBM (27.7%) and strong showings from others, including Cray, which had some sales of both supercomputers and internal products that boosted their revenue 23.4%. Another surprise was Dawning, which closed out the year with a whopping 73.8% revenue growth total over the course of 2012.
Conway stressed the importance of the overall trend line over multiple years for a healthy but hilly market like HPC. He plans on continued growth for the accelerator and coprocessor market, continued investment worldwide in exascale initiatives (although he argues that one common theme among nations is the need to provide a solid ROI argument for sustained funding), and what is already shaping up to be an exciting 2014—from the very peak of computing down to the smaller technical computing cluster levels.