NCUBE

nCUBE
Company type	Private (until acquisition)
Industry	Computer hardware, Video on demand technology
Founded	1983; 43 years ago in Beaverton, Oregon, United States
Founder	Steve Colley, Bill Richardson, John Palmer, Doran Wilde, Dave Jurasek
Defunct	2005
Fate	Acquired by C-COR in 2005; later absorbed into ARRIS in 2007
Headquarters	Foster City, California (later headquarters), United States
Area served	Worldwide
Key people	Larry Ellison (majority owner and interim CEO), Ronald Dilbeck (former CEO), Michael J. Pohl (CEO from 2002)
Products	Parallel computing machines (nCUBE 10, nCUBE 2, nCUBE-3, nCUBE-4), Video-on-demand systems (MediaCUBE 4)
Number of employees	Not specified
Parent	C-COR (from 2005), ARRIS (from 2007)

nCUBE referred to a line of parallel computing machines created by the eponymous company. The initial versions incorporated proprietary microprocessor architectures, but later iterations shifted to utilizing commercially available server-grade processors designed for extensive parallel processing, with a primary focus on the on-demand video sector.

Establishment and initial expansion

The company was established in 1983 in Beaverton, Oregon by former Intel engineers Steve Colley, Bill Richardson, John Palmer, Doran Wilde, and Dave Jurasek. nCUBE originated as a response to Intel’s reluctance to fully engage in parallel computing despite launching the iPSC/1 simultaneously. Its inaugural hypercube system debuted in 1985, followed by subsequent models: N2 in 1989, N3 in 1995, and N4 in 1999.

In 1988, Oracle Corporation's co-founder and CEO, Larry Ellison, acquired a controlling interest and relocated nCUBE’s headquarters to Foster City, California to foster closer ties with Oracle. Under Ronald Dilbeck’s leadership beginning in 1994, the company geared up for an initial public offering (IPO).

Shift towards video technology

After Dilbeck's exit in 1996, Ellison reduced the company's scale and took on the role of interim CEO, repositioning nCUBE as part of Oracle’s Network Computer initiative. Subsequently, the firm concentrated its efforts on developing video server technology and implemented its first video-on-demand system at Dubai’s Burj al-Arab hotel.

By 1999, nCUBE had acquired SkyConnect, a Colorado-based software company specializing in digital advertising and cable TV VOD solutions. At this stage, nCUBE had transitioned from its roots in parallel computing to become a prominent supplier of VOD systems, having delivered over 100 installations capable of streaming 17,000 simultaneous video feeds and collaborating with Microsoft TV. Plans for another IPO were postponed due to the collapse of the dot-com market.

Legal disputes and post-dot-com challenges

In 2000, SeaChange International initiated a patent infringement lawsuit against nCUBE concerning its MediaCube-4 product. Although a jury initially sided with SeaChange and granted damages, this decision was reversed by the U.S. Court of Appeals in 2005. Meanwhile, after acquiring patents from Oracle's interactive TV division in 2001, nCUBE filed suit against SeaChange alleging infringement on its VOD delivery technology patents. This lawsuit resulted in a favorable judgment awarding nCUBE more than $2 million, which was upheld upon appeal in 2004.

The financial health of nCUBE suffered due to both the dot-com downturn and ongoing litigation. In April 2001, the company reduced its workforce by 17%, subsequently closing offices in Foster City (2002) and Louisville (2003), consolidating operations at its original manufacturing site in Beaverton. In 2002, founder Larry Ellison resigned as CEO and named Michael J. Pohl—former CEO of SkyConnect—as his replacement.

Acquisition

In January 2005, C-COR purchased nCUBE for roughly $89.5 million following an SEC filing made in October 2004. Later, in December 2007, C-COR was itself acquired by ARRIS, thereby enhancing ARRIS’s capabilities within the video delivery technology market.

nCUBE 10

Introduced in late 1985, the nCUBE 10 was among the earliest models in the nCUBE series. It incorporated custom-designed chips, with each compute node equipped with a 32-bit ALU, a 64-bit IEEE 754 FPU, specialized communication instructions, and 128 KB of RAM. Each node delivered performance metrics of 2 MIPS, 500 kiloFLOPS in single precision, or 300 kiloFLOPS in double precision. Boards contained 64 such nodes. The host board, powered by an Intel 80286 processor, operated Axis, a proprietary Unix-like operating system, while individual compute nodes ran a minimal 4 KB kernel named Vertex.

The designation nCUBE 10 derived from its ability to construct a hypercube topology of order ten, enabling configurations of up to 1,024 CPUs within a single system. The architecture included specialized modules such as the nChannel storage controller for dedicated I/O tasks, frame buffers, and InterSystem cards designed to link multiple nCUBE units. At least one host board was necessary to manage user terminals and could subdivide the system into "sub-cubes," facilitating multi-user resource allocation.

nCUBE 2

The succeeding generation brought forth the single-chip nCUBE 2 processor, initially clocked at 25 MHz delivering approximately 7 MIPS and 3.5 megaFLOPS; later models like the 2S variant increased the clock speed to 30 MHz. Memory capacity per module expanded significantly to between 4 and 16 MB, with various physical formats including both single and double-wide modules. The I/O subsystem supported multiple protocols such as SCSI and HIPPI.

Each nCUBE 2 CPU was outfitted with thirteen I/O channels operating at 20 Mbit/s—one dedicated to I/O operations and twelve for CPU interconnections utilizing wormhole routing techniques. These channels were arranged in an order-twelve hypercube topology that could scale up to systems containing as many as 4,096 CPUs.

Modules executed a compact microkernel named nCX, approximately 200 KB in size. Instead of traditional host controllers, Sun Microsystems workstations served as front-end interfaces. The nCX kernel incorporated a parallel filesystem capable of up to ninety-six-way striping to optimize throughput. Supported programming languages encompassed C, C++, NQS, Linda, and Parasoft's Express, all supported by an internal compiler development team.

The most extensive nCUBE 2 deployment was at Sandia National Laboratories featuring a system with 1,024 CPUs that achieved peak performance of 1.91 gigaFLOPS during benchmarking. In addition to running nCX, this system also operated the SUNMOS lightweight kernel for experimental research purposes. Significantly, Sandia researchers Robert Benner, John Gustafson, and Gary Montry earned the inaugural Gordon Bell Prize in 1987 utilizing the nCUBE platform.

nCUBE-3

The nCUBE-3 processor introduced a 64-bit arithmetic logic unit (ALU) fabricated using a refined 0.5µ process technology, which enhanced clock speeds to 50 MHz with future plans targeting frequencies of 66 and even 100 MHz. It featured superscalar capabilities alongside separate 16 KB caches for instructions and data and incorporated a memory management unit enabling virtual memory support.

Its I/O architecture was expanded by adding two dedicated I/O links plus sixteen interconnect channels, allowing hypercube configurations scaling up to an impressive maximum of 65,536 CPUs. These channels operated at an upgraded rate of 100 Mbit/s using two-bit parallel lines rather than previous serial connections. The nCUBE-3 also introduced adaptive fault-tolerant routing mechanisms complementing traditional fixed routing schemes.

At full scale, an nCUBE-3 system could deploy up to 65,536 processors delivering around three million MIPS and achieving peak floating-point performance of approximately 6.5 teraFLOPS. Memory capacity could reach as high as 65 TB with network I/O bandwidth totaling up to 24 TB/s—highlighting a strong focus on input/output throughput. The associated nChannel board provided sixteen I/O channels each capable of sustaining data rates of 20 MB/s.

Although a custom microkernel was initially developed for this generation, it was eventually discontinued in favor of adopting the Transit operating system from Plan 9.

nCUBE-4

The nCUBE-4 transitioned to using standard Intel IA32 server processors for each node, focusing primarily on video streaming applications rather than scientific computation. Each hub combined a single hypercube node, one CPU, two PCI buses, and supported up to 12 SCSI drives. Later iterations included the n4x, n4x r2, and n4x r3 models, with the final two adopting Serverworks chipsets in place of Intel components.

The nCUBE-5 maintained similarities with the n4 series but incorporated two hypercube nodes per hub and enhanced video streaming capabilities through Gigabit Ethernet connectivity.

In 1999, the company introduced the MediaCUBE 4, capable of delivering 80 concurrent streams at 3 Mbit/s and handling as many as 44,000 simultaneous video-on-demand sessions across MPEG-2, MPEG-1, and medium bit-rate formats—highlighting its strong position in the VOD market.

External links

Overview of nCUBE Corporation's systems