SEC Filings

CRAY INC filed this Form 10-Q on 10/30/2018
Entire Document

a decrease in our accounts payable balance of $28.0 million due to the timing of payments to vendors, largely in connection with inventory purchases in the fourth quarter of 2017, and a decrease in customer contract liabilities of $29.3 million.
Market Overview and Challenges
Significant trends in the HPC industry include:
convergence of traditional supercomputing modeling simulation with big data analytics and AI;
supercomputing with many-core commodity processors driving increasing scalability requirements;
increased micro-architectural diversity, including increased usage of many-core processors and accelerators (such as graphics processors or GPU’s), as the rate of increases in per-core performance slows;
data I/O and storage capacity needs growing as fast as computational needs;
the rise of AI along with machine learning and deep learning algorithms that utilize HPC technologies for performance and scale;
technology innovations in memory and storage allowing for faster data access such as high bandwidth memory, non-volatile memory and storage, solid state and flash devices;
the increasing commoditization of HPC hardware, particularly processors and system interconnects;
the growing concentration of very large suppliers of key computing, memory and storage components in the industry;
the growing commoditization of software, including more capable open source software;
electrical power and system cooling requirements becoming a design constraint and driver in total cost of ownership determinations;
increasing use of AI and analytics technologies in both the HPC and big data markets;
increased adoption of cloud computing as a solution for loosely-coupled HPC applications;
much higher memory costs during the past year; and
significant variability in market demand for high-end supercomputers from quarter-to-quarter and year-to-year.
Several of these trends have recently impacted the growth rate and related improvements in price-performance of products in the industry and has contributed to the expansion and acceptance of loosely-coupled cluster systems using processors manufactured by Intel, AMD and others combined with commercially available, commodity networking and other components, particularly in the middle and lower portions of the supercomputing market. These systems may offer higher theoretical peak performance for equivalent cost, and “price/peak performance” is sometimes the dominant factor in HPC procurements. Vendors of such systems often put pricing pressure on us, resulting in lower margins in competitive procurements.
In the market for the largest, and most scalable systems, those often costing in excess of $10 million, the use of generally available network components can result in increasing data transfer bottlenecks as these components do not balance processor power with network communication and system software capability. With increasing processor core counts due to new many-core processors, these unbalanced systems will typically have lower productivity, especially in larger systems running more complex applications. We and others augment standard microprocessors with other processor types, such as graphics processing units, in order to increase computational power, further complicating programming models. In addition, with increasing scale, bandwidth and processor core counts, large computer systems use progressively higher amounts of power to operate and require special cooling capabilities.
To position ourselves to meet the market’s demanding needs, we concentrate our research and development efforts on technologies that enable our supercomputers to perform at scale - that is, to continue to increase actual performance as systems and applications grow ever larger in size - and in areas where we can leverage our core expertise in other markets whose applications demand these tightly coupled architectures. We also invest relatively significantly in next-generation technology to successfully and uniquely address the challenges of “Exascale computing” (systems with exaflops-levels of performance, one to two orders of magnitude faster than current supercomputers). In addition, we have industry leadership in developing an integrated supercomputing software stack with demonstrated expertise in system and performance software for several processor architectures. We expect to be in a comparatively advantageous position as larger many-core processors become available and as multiple processing technologies become integrated into single systems in heterogeneous environments. In addition, we have continued to expand our