This post is an offshoot of the previous one (which still under construction), on the global industrial organization of HPC market. It is a part that went too long and had enough substance to live in an independent post.
Since its beginning the HPC sector has experienced a series of ups and downs with new player entries and fails. The long-lived SGI (recently acquired by Rackable Systems), Quadrics (1996-2009, one of the few EU players in HPC, member of the robust aerospace and defence hi-tech conglomerate Finmeccanica, closed operations on June 2010; support contracts and possibly IP were later been bought by Vega, a company of the same conglomerate), Woven (interconnect vendor, acquired in August 2009 by Fortinet) and Sicortex are later examples of fails. And there are more: see for instance the Inside HPC article about HPC dead pool.
This high rate of fails does not come as a surprise. Unless you have a disruptive innovation that you are able to bring to the market very quickly, with very short new generation product innovation cycles, and make it a hit from the first release, it is unlikely that new players in the sector, specialized in a mono-segment, HPC, will survive. Your competitors, the big players, are offering products in all computing system markets (from ASIC to supercomputers),
including mass markets, so that they have enough financial muscle to support the higher risk investment in HPC. Cray is a mono-segment specialist, but he took advantage of the lead and has managed to keep it until today. Here more about Cray strategy. Moreover the continuity of the Cray name masks a rugged history for the company (Cray, Cray-SGI and Cray-Tera). In our days they are diversifying through strategic alliances with Microsoft and Intel in order to attack the low-end mass market. And of course, not all small niche players fail: see for instance Appro. This later case might be the subject of a post in the near future.
Sicortex was founded in 2003 by three HPC specialists (in technology and marketing). They raised up to 68 M USD from VCs in rounds A and B (VC´s were: Hercules Technology Growth Capital, Flagship Ventures, Polaris Ventures, Prism VentureWorks, JK&B Capital, and Chevron Technology Ventures) but after outsourcing the third financing round to a Wall street firm, failed to raise the more than 30 M USD they needed. Despite the fact that they were backed by a good IP portfolio, an apparently good differentiation strategy (low-cost custom green HPC systems targeted for fine-grained parallel applications where communication can be a bottleneck), some technological innovations (its systems were made of six-core MIPS chip node connected by a Kautz graph topology): “the company used relatively slow, inexpensive processors but stitched them together with a very fast, distributed communications network in which every processor contains some network ability, in a so-called Kautz graph topology”, from Networkworld) and growing sales (80 clients portfolio and growing), they had to shutdown operations in the middle of 2009. More data about Sicortex shutdown in this interview and here from Inside HPC, or here or here from other online mdia. I found very interesting the first link, an interview with one of the Sicortex founders.
Once closed, what remains to do is to sell the assets, in this case through a specialized firm, Gerbsman Partners (see their blog here and post). According to the later link, “SiCortex’s assets include: patents (granted and pending); proven, energy-efficient computer systems (the company’s core products); sophisticated systems and management software; core engineering team members; advanced simulations; chip design verification; existing and partially complete next generation products such as sub-assemblies, backplanes, racks, fans, cabinets, design tools, testing lab tools and products; software and hardware roadmap; trademarks; customers (including government intelligence agencies); The Pathscale® Compiler Suite, OEM, ISV and reseller relationships; external manufacturing relationships; experienced systems management and sales team; market presence”. They also say that “Patents support the unique technology. 1 issued, 12 pending. These patents center around the unique communications architecture built into the machine (no wires or cables) that provides for <1 microsecond access time to over 5800 processors”. According to the USPTO searching engine Sicortex has the following granted patents and the following pending applications. To clarify, at present already 7 patents of the 13 applications has been granted. Six were granted during 2010. As far as I know (see this source: source, at the end of 2009 no deal on Sicortex assets had been done yet, except regarding the Path Scale Compiler Suit. Some more news about Sicortex IP assets destiny, all from year 2009, here, here and here from online media, and also comments from Cray web page. Apparently the sale has focused on the Path Scale Compiler suite. So, did Cray bought also the rest of IP portfolio ? I ignore it. What happened to the Sicortex IP portfolio ? Was it sold ? If it was not, why wasn´t it ? These questions about Sicortex IP can be labeled as Sicortex first mystery.
At first glance there is no clear explanation for the Sicortex operations shutdown. Even in the hardest of the crisis companies with a comfortable horizon are able to raise capital. If the horizon is cloud free 30-40 M USD, the amount needed by Sicrotex to keep on running operations, is not that much. So what happened with Sicortex ? What went wrong ? That´s the second mystery. I´ve read the opinions from “insiders” such as Sicortex founders, from outsider commenters such as several high-tech online media. But I miss here Sicortex VCs´s opinions. My take on this, and this is just a working hypothesis, is that the cause of their failure might be just that they targeted the wrong market segment. As we have said there is a heavy binary vertical differentiation in this sector: or your system is the one with best performance and then you lead the high-end, or your system, up to a given performance threshold, is the cheaper and then you can compete in the low-end. Sicortex marketing strategy was based on the idea that there was a niche in the middle and there was not, there is not. Best performance will probably mean in the future that a vendor becomes a specialist in a given application, until the performance for the hardest cases in this application field reach the performance satisficer / cost minimizer stage: for instance in the molecular dynamics simulation field, until the protein folding of the hardest 2000 residue size chain can be predicted with molecular dynamics simulation methods at 100% accuracy in milliseconds. We all know what to be the cheapest mean. Of course Sicortex sales were not zero, they had sales growth expectations and promising internationalization plans (maybe implemented too late), but after several years in the USA market with their gen 1 systems, VCs probably could see that the company was running against market forces. In fact green supercomputing, the label that defined Sicortex market positioning, means less power consumption during computation, which can be reduced to cost. In short: they were competing with commodity low-end vendors. To get more evidence for the validity of this working hypothesis I would need to collect information about prices of Sicortex and competitors products and that´s beyond my free time availability.