Convex Computer

Convex Computer Corporation was an American high-performance-computing company founded in Richardson, Texas, in 1982 by Bob Paluck and Steve Wallach. The company targeted scientific, engineering, and other technically demanding customers that needed more performance than conventional business computers could provide but could not justify the cost or complexity of the largest Cray-class systems. Its commercial strategy was to deliver vector-processing capability at a lower price by using comparatively mainstream semiconductor technology, virtual memory, and software intended to make it easier to port existing programs. The company’s first major product was the C1, introduced in 1985. Architecturally, it resembled the broad design of contemporary Cray vector computers, but used less expensive CMOS implementations for the processor and main memory. Convex sought to compensate through stronger vector units, larger vector registers, and optimizing compilers that automatically converted suitable scalar code into vector operations. The system delivered substantially less peak performance than a Cray-1, but its price-performance proposition and programming environment made it attractive to smaller and medium-sized scientific-computing installations. Convex systems ran a Unix derivative initially known as Convex Unix and later as ConvexOS. The software environment included COVUE, which provided elements of DEC VMS compatibility, as well as compiler and language features familiar to Cray users. The C2, released in 1988, expanded the design into a crossbar-connected multiprocessor system supporting as many as four CPUs. It became Convex’s most successful product. The C3 followed in 1991, adding faster processing and support for systems with up to eight CPUs. Convex also pursued the C4 program and became an early vendor of a gallium-arsenide-based high-performance system. These efforts reflected the company’s attempt to extend the vector-supercomputer model as semiconductor technology advanced. Convex’s business was challenged by the rapid improvement of RISC microprocessors and the emergence of increasingly capable, lower-cost workstations and server clusters. The development and manufacturing economics of a specialized, relatively low-volume supercomputer company became more difficult, while delays in delivering the C3 contributed to lost sales. Convex’s later GaAs efforts increased technical capability but did not reverse the company’s financial and competitive problems. In 1994, Convex introduced the Exemplar family, a fundamental shift from its earlier vector architecture toward parallel computing. The SPP1000 and subsequent SPP1200 systems used Hewlett-Packard PA-RISC processors connected through Scalable Coherent Interface technology. The products were intended to attract customers to a parallel-programming model, but Convex encountered difficulties with operating-system performance and with the mismatch between the technology’s programming assumptions and the preferences of many of its established customers, who tended to favor Fortran-based, brute-force vector computing. Convex developed a working relationship with Hewlett-Packard as it tried to adapt Exemplar systems to HP technology. In 1995, Hewlett-Packard acquired Convex. HP subsequently marketed related systems under S-Class and X-Class designations and incorporated elements of Exemplar technology into its V-Class systems. The Convex corporate brand eventually disappeared, although its engineering work continued to influence HP’s high-end server and technical-computing products.

Convex Computer was established in 1982 in Richardson, Texas, initially under the name Parsec. Its founders, Bob Paluck and Steve Wallach, set out to build a less expensive alternative to the large vector supercomputers supplied by Cray Research. Rather than pursue the most technologically aggressive design, Convex relied on more conventional chip technology and attempted to recover performance through vector-processing enhancements, software optimization, and a favorable price-performance ratio. The company’s first commercial system, the C1, appeared in 1985. Its overall organization was comparable to the Cray-1, but its processor and main memory used slower and cheaper CMOS technology. Convex equipped the system with comparatively capable vector units, including vector registers holding 128 64-bit elements, and used virtual memory rather than the static memory approach associated with Cray systems. The C1 was rated at approximately 20 MFLOPS peak in double precision and 40 MFLOPS in single precision. Although these figures were below those of leading Cray systems, Convex emphasized affordability and ease of program migration. Its automatic vectorizing compilers were designed to extract vector operations from existing scientific programs, while the operating system provided Unix compatibility together with selected VMS- and Cray-oriented features. The C2, released in 1988, was a crossbar-connected multiprocessor development of the C1. It could be configured with up to four CPUs and used newer CMOS and ECL gate-array technology to increase clock speed. The C2 became Convex’s most successful commercial product. In 1991, the company released the C3, which offered faster processors and configurations of up to eight CPUs. Convex also pursued a C4 design using gallium-arsenide technology. The company was among the early vendors to ship a GaAs-based product, but the engineering effort came at a time when specialized supercomputer economics were deteriorating. The market around Convex changed rapidly during the early 1990s. RISC microprocessors became more powerful and less expensive, making it increasingly difficult for a small, specialized manufacturer to develop a competitive proprietary high-performance architecture. The C3 was delivered later than planned, affecting sales, and the C4 program could not compensate for the broader shift toward commodity processors and distributed or parallel systems. Convex was losing money while competing against both established supercomputer companies and emerging workstation and server technologies. In 1994, Convex introduced Exemplar, a new family based on parallel rather than traditional vector processing. The SPP1000 used Hewlett-Packard PA-7100 processors, and later SPP1200 systems used PA-7200 processors. The processors were connected using Scalable Coherent Interface technology. Convex expected customers to adopt a parallel-programming model, but this conflicted with the preferences of much of its installed base, which favored Fortran and relatively direct, brute-force scientific computing. Exemplar also suffered from operating-system performance problems. Convex worked with HP’s hardware and software organizations and ultimately moved toward porting HP-UX to the platform rather than maintaining a fully separate binary-compatible environment. Hewlett-Packard acquired Convex in 1995. HP sold related systems as S-Class multiprocessors and X-Class cache-coherent NUMA systems. Elements of Exemplar technology were later incorporated into HP’s V-Class machines, which ran HP-UX 11.0. The Convex corporate identity was eventually retired, but the acquisition extended aspects of Convex’s parallel-computing work into HP’s technical-server portfolio.

• 1995

  Hewlett-Packard acquires Convex

  [1]Convex Computeren.wikipedia.org

  HP acquired the company and continued related technology in S-Class, X-Class, and later V-Class systems.

• 1994

  Exemplar platform is introduced

  [1]Convex Computeren.wikipedia.org

  Convex moved into parallel computing with PA-RISC-based SPP and Exemplar systems connected through SCI technology.

• 1991

  C3 is introduced

  [1]Convex Computeren.wikipedia.org

  The C3 increased processing speed and supported configurations with up to eight CPUs.

• 1988

  C2 is released

  [1]Convex Computeren.wikipedia.org

  The C2 added crossbar-connected multiprocessing and became Convex’s most successful product.

• 1985

  C1 enters the market

  [1]Convex Computeren.wikipedia.org

  Convex launched its first vector minisupercomputer, emphasizing lower cost, virtual memory, and automatic vectorizing compilers.

• 1982

  Convex is founded in Richardson, Texas

  [1]Convex Computeren.wikipedia.org

  Bob Paluck and Steve Wallach founded the company, initially using the name Parsec during early development.

Lower-cost, Unix-based vector and parallel high-performance computing for scientific and engineering organizations.

• Vector minisupercomputer · 1985

  Convex C1

  Released in 1985, the C1 was Convex’s first commercial system. It used a Cray-influenced vector architecture but implemented its CPU and main memory with less expensive CMOS technology. Virtual memory, enlarged vector registers, and optimizing compilers were intended to make the system practical for scientific and engineering users migrating existing code.

• Multiprocessor vector supercomputer · 1988

  Convex C2

  The C2, introduced in 1988, developed the C1 design into a crossbar-connected multiprocessor system. It supported as many as four CPUs and used newer CMOS and ECL gate-array technology to raise clock speed and performance. The C2 was reported as Convex’s most commercially successful product.

• Vector supercomputer · 1991

  Convex C3

  Introduced in 1991, the C3 retained the general C-series architecture while increasing clock speed and supporting up to eight processors. Configurations delivered different performance levels, but the product faced delays and entered a market increasingly shaped by powerful RISC processors and lower-cost parallel systems.

• Gallium-arsenide supercomputer program

  Convex C4

  The C4 represented Convex’s attempt to extend its vector-computing business with gallium-arsenide-based hardware. Convex was an early vendor to ship a GaAs-based product, but the program arrived as the economics of proprietary, low-volume supercomputers were deteriorating and did not restore the company’s competitive position.

• Parallel computer · 1994

  Convex Exemplar SPP systems

  The Exemplar family, initially associated with the SPP1000 and later SPP1200 and SPP-1600 designations, shifted Convex toward parallel computing. It used Hewlett-Packard PA-RISC processors connected by Scalable Coherent Interface technology. HP later marketed related implementations under S-Class and X-Class names and reused aspects of the technology in V-Class systems.

• Operating system

  ConvexOS

  ConvexOS was Convex’s Unix-based operating environment, initially known as Convex Unix. It included compatibility features associated with DEC VMS through COVUE and language and compiler features intended to assist customers familiar with Cray systems.

• 1995

  Sell the company to Hewlett-PackardM&A

  Background

  Convex faced increasing competitive and financial pressure as specialized supercomputer development became harder to sustain independently.

  Action taken

  Hewlett-Packard acquired Convex and integrated its technology with HP hardware and software activities.

  Aftermath

  HP offered related Exemplar-derived systems as S-Class and X-Class products and later incorporated elements into the V-Class family.

  Source

• 1994

  Shift from vector systems to parallel computingStrategy

  Background

  The rise of RISC processors and commodity-based high-performance computing challenged Convex’s proprietary vector-computer business.

  Action taken

  Convex introduced Exemplar systems based on HP PA-RISC processors and SCI interconnects, aiming to develop a scalable parallel-computing platform.

  Aftermath

  Operating-system performance problems and customer preference for established Fortran-oriented vector workflows limited the strategy, and the technology was subsequently absorbed into HP product lines.

  Source

• 1982

  Pursue lower-cost vector supercomputingStrategy

  Background

  Convex entered a market dominated by expensive, technologically aggressive vector systems and sought to serve smaller organizations with a more affordable alternative.

  Action taken

  The company used more mainstream chip technology while investing in vector units, virtual memory, and automatic vectorizing compilers to improve practical performance and software portability.

  Aftermath

  The approach established Convex in scientific and engineering computing and supported the commercial success of the C2, although later RISC-based competition weakened the model.

  Source

• 1995

  Hewlett-Packard acquires Convex Computer[1]Convex Computeren.wikipedia.org

  M&A

  Hewlett-Packard purchased Convex and continued offering related Exemplar technology in S-Class and X-Class systems before incorporating elements into the V-Class line.

• 1994

  Convex announces the Exemplar parallel-computing platform[1]Convex Computeren.wikipedia.org

  Product launch

  Convex shifted from its C-series vector architecture to Exemplar systems using HP PA-RISC processors and Scalable Coherent Interface interconnects.

• 1991

  Convex introduces the C3[1]Convex Computeren.wikipedia.org

  Product launch · Product generation

  The C3 increased clock speed and supported systems with as many as eight processors, but arrived as RISC-based computing was intensifying competitive pressure on specialized supercomputer vendors.

• 1988

  Convex launches the multiprocessor C2[1]Convex Computeren.wikipedia.org

  Product launch · Product generation

  The C2 extended the company’s vector architecture with crossbar-connected multiprocessing and support for configurations of up to four processors.

• 1985

  Convex introduces the C1 vector supercomputer[1]Convex Computeren.wikipedia.org

  Product launch · Product generation

  Convex released its first commercial machine, combining a Cray-like vector-computing approach with lower-cost CMOS hardware, virtual memory, and vectorizing compiler technology.

1. 1Convex Computeren.wikipedia.org