Although there are many HPC startups that are focused on software, services, and consulting, to keep the list focused this year, we singled out a few key small companies that are taking their chances to bring new processors and server gear to the supercomputing table.
It is no simple task to “break into” high performance computing where the contracts are large and the need for (as near to) guarantees is critical at scale. As we noted in our analysis of the Top 500 supercomputer results that were published this week at the International Supercomputing Conference (ISC ’15) in Frankfurt, Germany, the stable of vendors that most HPC centers pull from is stocked with the expected names–IBM, HP, Cray, and Dell.
To address the much larger playing field, we handpicked some key companies that might be poised for growth as we usher in the pre-exascale era and its need for performance in a more dense, lower latency, and most important, energy efficient manner. The selections are in not in rankedorder and run the gamut–from companies that have been steadily plugging away for five or more years to fresh startups.
Adapteva is an innovator in parallel processing delivering an energy efficient and scalable multicore processor chip designed for high performance computing.
The Epiphany multicore architecture represents a new class of massively parallel computer processors that is aimed at a wide range of markets, from compact low-power devices to next-generation supercomputers The company says its Epiphany chip is 27x more energy efficient than an Intel Xeon, supporting parallel processing for applications that need large scale computational strength in a small package.
As mentioned in our discussion about some trends in accelerated and low-power systems here, there are several companies in Japan and elsewhere in Asia that are taking a fresh approach to system and processor design and integration. Among these is Exascaler, Inc, who in conjunction with its sister company, specialist chipmaker, PEZY, have claimed two new machines in Japan.
Exascaler has developed its own submersion liquid cooling technology, which when coupled with the PEZY-SC 1,024-core MIMD processor, provides highly energy efficient performance, which was demonstrated by its #2 placement on the Green 500 list of supercomputers in November 2014.
Kalray is a fabless semiconductor & software company proposing disruptive manycore processors for high performance applications. The company’s MPPA Manycore processor is a massively parallel architecture designed to provide the best power efficiency and scalable computing platform, which the company says can provide a 5 to 50 x improvement on the GFlops/Watt ratio over conventional processors.
Kalray was founded in July 2008 and has raised more than $20 million in funding. While they are not necessarily small (mostly because of their foothold in embedded computing) they do have an opportunity with their processor technology for key data-intensive and deep learning workloads that are being tied into existing HPC application areas.
E4 Computer Engineering caught our attention this week during ISC because it has been working with some clever designs featuring GPU-accelerated Cavium ThunderX server platforms design for HPC applications.
The small Italian company, which was founded in 2002 and has worked with CERN, among other European research institutions, designs and manufactures complete HPC and enterprise solutions for both industrial and scientific research and designs each system individually to deliver highly personalized, cost effective approaches to high performance computing.
Gidel is not a new company, and it is not a particularly small one either, but it is positioning its FPGA offerings squarely in line with the trend curve around FPGA technology for HPC and data-intensive applications. The company was founded in 1993 as a high-end system development and integration company. They have since created several powerful and advanced architectures for high-performance computation development.
This week at ISC they were demonstrating a new architecture based on FPGAs enables to dramatically reduce power consumption while significantly increasing computing performance for OpenCL approaches to FPGAs.
German HPC company, Extoll, is focused on supercomputing network technology with its novel network chip, Tour-Malet, which targets low-latency, high hardware message rates, high bandwidth, a switchless design, and several other features focused on fault tolerance and scalability.
For performance computing users whose applications are currently limited by latency, scalability or power, EXTOLL says it has found some clever solutions to avoid costly and power-hungry external switches. The EXTOLL solution integrates host-interface, network interface controller and network functions within a single chip.
This South Korean company is focusing on future exascale systems with a new approach to system integration application development, interconnects and processor design. At the core of their company is the Klimax-210 system, which CoCoLink says is capable of handling 30 teraflops in a single node with the capability of installing 20 high performance processors in each node, SSDs, and expanding the PCIe buses. These can scale out to petaflops range and can also be optimized for GPUs.
CoCoLink says that with their approach, the cost of configuration is five to ten percent of existing CPU systems. More on their novel interconnect and architectural features can be found here.
It is easy to get wrapped up in the server and performance hardware startups, but there are a number of relatively new companies that are targeting the energy efficiency of large-scale HPC datacenters. Among these is Calyos, a Belgian provider of advanced two-phase cooling solutions for high performance computing servers. Adaptable to liquid or air cooled racks, the company’s offerings aim to provide a silent high efficiency platform that will enable the use of high temperature cooling fluid. The goal is to significantly reduce operating costs of supercomputing datacenters, as well as pave the way for the next generation of very high compute density.
This week at ISC 15, Calyos was showing off its cooling solutions for HPC servers in hot water cooled racks and showing how, by combining a two-phase latent heat effect with a passive capillary pump, they can provide potential improvements over existing full liquid-cooled solutions.
On the energy-efficient systems front, another smaller company—and one that is newer to HPC—is making some waves recently. Asetek is just pushing the limits of what can be called a small company within HPC, particularly since it had a lead with cooling solutions for non-HPC systems at its inception. Now, with over 1.7 million units deployed, it has found inroads for energy efficient liquid cooling systems for supercomputing datacenters, and HPC workstations.
The company says is RackCDU provides data center cooling cost reductions exceeding 50% by capturing 60% to 80% of server heat with 2.5x-5x increases in data center compute density. They also note that their RackCDU ISAC sealed servers capture nearly 100% of server heat. The company is targeting its RackCDU at HPC and high-utilization data centers to improve TCO by decreasing energy consumption, increasing data center compute density and enabling server energy reuse.
While this is not an exhaustive list (and there were plenty of startups and small companies to talk to this week at ISC) it does show that there is fresh investment rolling into HPC hardware–from chips, to networks, to targeted solutions for more efficient HPC datacenters.