By William Fellows
Star Bridge Systems Inc, the Sandy, Utah, company founded by former car magnate Al DiMora to build go-faster supercomputers called hypercomputers, believes that within in two years it should be able to offer a $1,700 PC that performs at least six times better than today’s fastest Intel desktop or 1.2 billion calculations per second. The fastest supercomputers, built by IBM, Intel, and Cray, can only handle up to 0.4 billion calculations per second. The PC supercomputer it envisages will run Windows on top of Star Bridge’s Viva operating system services, which also features a programming language said to be akin to the graphical JavaStudio tools. Star Bridge hopes to encourage ISVs to rewrite the applications to run on Viva using its Implementation Independent Algorithm Description Language. Star Bridge claims the supercomputer PC will change the industry forever. Upgrades will be downloaded from the web.
Meantime, the company has delivered its first ‘reconfigurable’ supercomputer to internet search engine company called Icaveo.com Inc, which is jointly-owned by DiMora and Star Bridge co-founder and CTO Kent Gilson. The company can also shout more loudly about its Hyper Algorithm Logic HAL supercomputer this week having completed writing Viva. Until now the system has been difficult to program, Gilson admits. Now it’s got the Raytheon, TRW and Ames research labs interested enough that it’s starting up a Viva programming course and has around 15 people signed up. Its Viva GUI is an ActiveX form and it is currently developing a 3D GUI. It will host a system on the web for tire-kicking purposes.
Star Bridge has so far built 11 systems and is about to deliver its first. The HAL-300GrW1 it is currently touting uses 280 Field Programmable Gate Array chips supplied by Xilinx Inc, essentially chips that can be reprogrammed or reconfigured on the fly without having to be rebuilt. This is the system it is supplying to Icaveo. This particular configuration – the company’s HAL-4 generation – can perform a sustained 12.8Tflops executing a four- bit adder or 3.8Tflops using a 16-bit adder. It costs $26m to lease over four years and is about the size of a PC, sits on the desktop, weighs less than 150lbs and plugs into a 100 volt power outlet. By comparison IBM Corp’s Pacific Blue SP parallel system costs $94m, performs 3.9Tflops peak performance, requires 5,856 PowerPC 604s, uses 3.9 megawatts of power, takes up 8,000 square feet and needs air conditioning. Although it has not built a 100Tflops system, Gilson says it’s feasible with current HAL 4 technology; it has HAL 8 generation, 500 Tflops technology in hand. Unlike conventional computing environments, performance on reprogrammable systems is guaranteed and improvements in chip technology mean system-wide resource improvements – availability is squared – rather than a PC performance improvements which relies on improved clock speeds and shrinking of processor size.
Star Bridge has 60 staff but wants to be at between 9,000 to 12,000 within two years. It is entirely privately funded with no venture capital money. As well as selling and leasing systems Star Bridge is also pursuing a novel sales model which is to co- own joint venture with a customer which leases a HAL system. For example it is offering Constellation configurations of HAL for back-ending switching systems it claims offer near zero latency performance. It has created subsidiary, Voice Packet Exchange Inc, to use the Constellation products as part of an Internet telephone network. A Star Bridge Scientific Inc subsidiary will develop high-speed, low-latency, digital, packet-switched telecommunications payloads for low-earth satellites. And it’s planning a production studio that will have access to 8 HAL- 300GrW1 hypercomputers for graphical modeling. It is looking for partners for these ventures. One aim is to get on the Top 500 supercomputer list in short order.
