A New Kind of Computer Company
It is surprising when you look at it this way, but today's high performance computers have come either from teams of 300 or teams of three. Unfortunately, neither approach has produced sustainable success, except at the very high and very low end of the price scale. SiCortex has a different approach that is better matched to both the technical demands and the economic realities of the HPC marketplace.
Until recently, HPC systems were designed by teams of 300 or more. These teams produced extraordinary computers, but staggered under the burden of doing everything from scratch - including full custom chips, operating systems, and development tools.
Today, HPC users and vendors typically apply teams of three or less to combine processors from vendor A, interconnect from vendor B, and software from the open source community to build cluster computers. This approach has also produced remarkable results, but it forces HPC users to settle for system designs that are optimized for markets other than HPC. This limits scalability and delivered performance.
There is a better way. With a skilled team, SiCortex is creating a family of cluster computer systems that are optimized from the silicon up for HPC applications. We avoid the cost burdens of full custom design by leveraging the ecosystem of silicon manufacturers and suppliers of silicon intellectual property that has grown up over the past decade to serve vendors of specialized products such as cell phones and graphics processors.
With this approach, SiCortex can deliver a single-chip cluster node that is tuned for HPC applications. Citing just one example, the SiCortex node leaves out most of the power-hungry branch prediction logic found on commodity CPUs, which has little or no impact on HPC applications. This makes room on the chip for (among other things) a complete set of switching and switch interface logic, which has first-order impact on performance, cost, and power consumption.
Focusing on the areas where we can add the most value keeps our design cycles short and our cost structure low. The result is a company that is small enough to support a healthy business serving the HPC marketplace, but large enough to regularly produce new generations of machines with outstanding delivered performance.
