The News & Analysis article “Utility sacrificed for speed, supercomputer critics say” (D. Normile, 5 October, p. [26][1]) discussed our project, the Japanese 10-petaflops supercomputer (or “K computer”). Some criticisms reported in the article were based on misunderstandings, which I would like to clarify. Contrary to the title “utility sacrificed for speed,” I am confident that the K computer is the most user-friendly among the world's top-ranking supercomputers. In designing it, we focused not only on peak performance but also on sustained performance on real applications in a wide range of science and engineering fields. For example, the memory and network bandwidth per performance ([ 1 ][2]), which are key factors for application efficiency, are much better than those of IBM Sequoia (BlueGene/Q) at Lawrence Livermore National Laboratory ([ 2 ][3]) or Cray Titan at Oak Ridge National Laboratory ([ 3 ][4], [ 4 ][5]). We have also adopted the widely used standard programming model (MPI/openMP) as in BlueGene/Q. The K computer demonstrated an extraordinary level of stability for one of the world's largest-scale systems; the overall system comprised of 88,128 CPUs ran without a single failure for 29.5 hours. All these features (high bandwidth, standard software environment, and high reliability) require extra hardware and software costs, which decrease the peak performance possible for a given cost. Thus, we are proud to say that in reality, we sacrificed speed for utility in the K computer design. ![Figure][6] Supercomputer. Japan's K computer contains 864 watercooled cabinets. CREDIT: D. NORMILE/ SCIENCE In the article, Jun Makino implied that we did not explore the design sufficiently. The truth is that we accepted proposals on architecture from academia and industry in the design process. In the selection of benchmark applications, we paid special attention to the parallelizability (i.e., the marginal number of processors that minimize an execution time of an application) of the algorithms we used, which is strongly related to optimal supercomputer architecture. Performance was never the only criterion: Some proposals were retracted, and others judged as infeasible due to their weak developmental systems. I am confident that we explored various architectural possibilities, taking into consideration their relations to algorithms. Takafumi Matsui said that the project was designed just to serve the computer industry. This is not the case. Of the total project cost of $1.4 billion, $500 million was devoted to software development for life sciences and nanoscience, in addition to buildings and facilities. We recognize that the project's main goal was the development of the supercomputer, with the goal of broad application in science and engineering. Is this a bad thing? It would be if no one wanted to use our computer. The good news, as the article noted, is that many users have applied to use the K computer. Its true value will soon be proved. 1. [↵][7] 1. Riken , K Computer ([www.aics.riken.jp/en/][8]). 2. [↵][9] IBM, Blue Gene ( ). 3. [↵][10] 1. Cray , Cray XK7 ([www.cray.com/Products/XK/XK7.aspx][11]). 4. [↵][12] 1. Nvidia , Tesla GPU Accelerators for Servers ([www.nvidia.com/object/tesla-servers.html][13]). [1]: /lookup/doi/10.1126/science.338.6103.26 [2]: #ref-1 [3]: #ref-2 [4]: #ref-3 [5]: #ref-4 [6]: pending:yes [7]: #xref-ref-1-1 View reference 1 in text [8]: http://www.aics.riken.jp/en/ [9]: #xref-ref-2-1 View reference 2 in text [10]: #xref-ref-3-1 View reference 3 in text [11]: http://www.cray.com/Products/XK/XK7.aspx [12]: #xref-ref-4-1 View reference 4 in text [13]: http://www.nvidia.com/object/tesla-servers.html