Ceased trading in 1992, although some installed platforms are still being operated.
Alliant Computer Systems, formerly known as Dataflow, was founded in May 1982 by Gruner, Mundie and McAndrew to fill the science and engineering market niche for small to medium sized parallel systems.
Alliant's first parallel machines were announced in 1985, and a number of subsequent models were based on custom designed processors including the FX/40 and FX/80 shared memory multiprocessors, which were superceded at the end of the 1980's by the FX/2800 so-called "standards based supercomputer", which was based on Intel i860 microprocessors.
More than 650 Alliant systems were known to be installed in over 400 customer sites worldwide in 1992.
Typical applications included image/signal/seismic processing, computational fluid dynamics, computational chemistry, visualisation and structural analysis. The FX/2800 quickly took over and expanded the large application base of its predecessors.
The CAMPUS/800 product was announced at the end of 1991, and consisted of a number of FX/2800 shared-memory nodes, and unified-memory: physically distributed memory with a shared-memory view, controlled by sophisticated compilers and normal imperative language constructs, rather than explicit message-passing. Much of the development work, particularly the parallelising compiler, was done in collaboration with CSRD (Illinois) and is based on the Cedar project. Alliant had several installed CAMPUS/800 machines (which were probably still prototypes), with the largest known installation having 75 processors.
(No longer applicable)
Alliant Computer Systems Corp. 1 Monarch Drive Littleton, MA 01460 508-486-4950 Alliant Computer Systems UK Ltd 10 Heatherley Road Camberley Surrey GU15 3LW UK 0276-682765 FAX 0276-65235
Alliant CAMPUS/800 information folder including:
The FX series of machines were vector register, parallel, shared-memory architectures. The series included the FX1,4,8, 40 and 80. See also the
.
Compute Hardware: Computational elements (CEs) execute applications code using vector instructions. The CEs transparently execute the code of an application in parallel.
Weitek 1064/1065 plus ten different gate array types with 2600 to 8000 gates. First-generation computational elements (FX1, FX4, FX8) may be added in the field, increasing performance without recompilation or relinking. Advanced Computational Elements (ACEs) for second generation (FX40, FX80, VFX) are based on the BIT floating point chips. Each CE has 8 vector registers, each with 32 64-bit elements, and 8 64-bit scalar floating point, 8 32-bit integer, and 8 32-bit address registers.
Interactive Processors (IP) were Motorola 68020. 4 Mbyte local memory in each IP. ACE 64-bit processor 20,000 gate CMOS VLSI gate array, with BIT floating-point processors. 64 Kbyte instruction cache.
The cycle time is 170 nsec. Only six different PC boards are used.
Floating-point unit: IEEE 32- and 64-bit formats including hardware divide and square root and microcoded elementary functions.
Interconnect / Communications System: CEs are cross-bar connected on the backplane to a 512 Kbyte write-back computational processor (CP) cache (FX/80). Bandwidth is 376 Mbyte/sec.
Connectivity: crossbar (CE to cache), bus (cache to memory, cache to cache)
Memory System: Each 32-Kbyte IP cache is connected to 1-3 IPs (FX/80) or 1-2 IPs and a CE (FX/1). The FX/80 has 1-4 IP caches; the FX/4 and FX/40 have 2 IP caches; the FX/1 has one IP cache.
The CP and IP caches are attached by two 72-bit buses to the main memory. Memory bus bandwidth is 188 Mbyte/sec, and memory cycle time is 85 nsec.
Range of memory sizes available: 32-64 Mbytes (FX/1), 32-160 Mbytes (FX/4 and FX/40), and 32-256 Mbytes (FX/80), using 1 Mbit chips with ECC.
Virtual memory: 2 Gbytes per process.
Advanced CE's (ACE's).\\ Scalar 32-bit : 7.2 mips / CE (14700 Kwhetstones).\\ Scalar 64-bit : 6.2 mips / CE (13700 Kwhetstones).\\ Vector 32-bit and 64-bit : 23.5 Mflops / CE.\\ FX/80 on 1,000 x 1,000 LINPACK benchmark: 69.3 Mflops. Peak performance 188.8 Mflops.
System Software was Concentrix, Alliant's enhancement of Berkeley 4.2 UNIX with multiprocessor support.
Languages supported included: Fortran, C, Pascal, Ada, Lisp, STSC APL, 68020 Assembler.
The Fortran implementation was known to have the following characteristics: F77 - Conformed to 1978 ANSI standard; Extensions - Most of VAX/VMS extensions and Fortran 8x array extensions; Debugging facilities existed.
Vectorizing/parallelizing capabilities - Automatic detection of vectors and feedback to user via diagnostic messages.
Can employ COVI (concurrent outer, vector inner) on nested loops. User control of transformations via directives in the form of Fortran comments Interprocedural dependency analysis for automatic determination of parallel subroutine calls in loops.
Interactive Processors (IPs) execute operating system, interactive code, and I/O operations. An FX/1 has 1-2 IPs. An FX/4 and FX/40 have 1-6 IPs. An FX/80 has 1-12 IPs.
Stand-alone (hostless) configuration. TCP/IP network support.
Size: FX/1 system - 28" x 13" x 25" (the FX/1 I/O expansion cabinet is the same size); FX/4, FX/40, and FX/80 systems - 43.5" x 29.5" x 33.8" (the I/O expansion cabinet is 24.5" and same height and depth, while the tape cabinet is 61" in height).
Cooling: All systems are air-cooled with power consumption of FX/1 -- 1155 Watts (max. configuration), 725 Watts (I/O Expansion); FX/4 -- 4500 Watts, 2100 Watts (I/O Expansion); FX/40 -- 4200 Watts, 2100 Watts (I/O Expansion); FX/80 -- 5100 Watts, 2100 Watts (I/O Expansion).
Peripherals available included:
800/1600/6250 BPI start-stop tape drive; 550 Mbyte (formatted) Winchester disk drives; 45 Mbyte cartridge tape drive; Floppy disk drive; 8/16 line multichannel communications controllers; 600 lpm printer; Ethernet controller.
Applications included: Engineering and scientific end-user and OEM applications, stand-alone or as a computational server to a network of engineering workstations.
First beta delivery May 1985; first production shipment September 1985. Alliant's customers include Asahi Chemical Corp, AT\&T, Boeing Airplane Co., Ford Motor Co., Hughes Aircraft Corp., Motorola Inc., Siemens, The Whittle Laboratory at the University of Cambridge, CERFACS at Toulouse, and the Jodrell Bank Observatory at the University of Manchester.
A widely used reliable platform.
The FX/2800 architecture consists of up to 7 modules of 4 processors arranged as 2 SCEs (super computational elements) and 2 SIPs (super interactive processors).
Compute Hardware: Intel i860.
Interconnect / Communications System: Shared memory on bus system.
Memory System:
Alliant claimed 40 Mflops/node and 1.1Gflops peak performance. Achieve single-node performance of 6.76 Mflops and 14-processor performance of 89.3 Mflops on SLALOM benchmark (which corresponds to a respectible 94\% parallel efficiency).
Data transfer rates of 376MB/s maximum bandwidth to (up to) one 512kB cache per module were reported.
Operating System --- Concentrix full-function Unix and POSIX compliant OS, runs autonomously within ClusterNodes, and coordinates services across 'Nodes for a unified-system view; finely-threaded internally (for efficiency and balance), and providing extensions for parallel processing, large memory and filesystems, and high-performance distributed I/O. OS dynamically partitions based on user requirements and changing workloads. Also FX/RT real-time response and scheduling executive.
Languages --- Compilers available from the vendor for Fortran (including Fortran-90 extensions), C and Ada. Sophisticated parallelising Fortran compiler for scalar, vector, scalar concurrent, vector concurrent and concurrent-outer-vector-inner operations.
Programming Environment --- Parallel libraries and kernels provided developed for shared-memory, including optimised mathematics libraries for parallel EISpack, LINpack, NAG and IMSL. Sophisticated programming, profiling, debugging and visualisation tools. Also graphics support for PHIGS+, X, OSF/Motif, in addition to NFS/NQS and TCP/IP networking.
Scalability --- up to 28 Intel i860 RISC chips in up to 7 modules of 4 processors, sharing memory through a proprietary 16x8 crossbar switch to a memory bus serving processor caches.
CAMPUS stands for Clustered Architecture, Massively Parallel, Unified-memory System.
The CAMPUS/800 architecture consists of clusters of processors in ClusterNodes sharing up to 4GB of memory, with intra-ClusterNode (1st level switch) latency of 1us and 1.12GB/s bandwidth. ClusterNodes closely resemble (and can be upgraded from) FX/2800 machines. High-speed Memory Interconnect (HMI) connecting ClusterNodes, which have distributed, and may even be remotely sited, with inter-ClusterNode (2nd level switch) latency of typically 30us and 2.56GB/s bandwidth.
Node --- ClusterNodes consist of up to 25 i860s sharing up to 4GB of memory (and a 2MB cache)
Performance --- Claim 40 MFLOPS/node peak performance. 4.7 GFLOPS achieved by 192 processor CAMPUS/800 on massively-parallel LINPACK benchmark (25 MFLOPS/node achieved).
Data Transfer --- Two-level switch: intra-ClusterNode, 1 us latency, 1.12 GB/s bandwidth; inter-ClusterNode, 30 us latency, 2.56 GB/s bandwidth. HMI based on HiPPI channel standard, allows ClusterNodes to be physically separated (up to 30km) at different locations without loss of bandwidth.
Operating System --- Concentrix full-function Unix and POSIX compliant OS, runs autonomously within ClusterNodes, and coordinates services across 'Nodes for a unified-system view; finely-threaded internally (for efficiency and balance), and providing extensions for parallel processing, large memory and filesystems, and high-performance distributed I/O. OS dynamically partitions based on user requirements and changing workloads. Also FX/RT real-time response and scheduling executive. Communication Paradigms --- Express (on CAMPUS/800). Languages --- Compilers available from the vendor for Fortran (including Fortran-90 extensions), C and Ada. Sophisticated parallelising Fortran compiler for scalar, vector, scalar concurrent, vector concurrent and concurrent-outer-vector-inner operations. Programming Environment --- Parallel libraries and kernels provided developed for shared-memory, including optimised mathematics libraries for parallel EISpack, LINpack, NAG and IMSL. Sophisticated programming, profiling, debugging and visualisation tools. Also graphics support for PHIGS+, X, OSF/Motif, in addition to NFS/NQS and TCP/IP networking.
Scalability --- up to 800 PAX-conformant Intel i860 RISC chips in a maximum of 32 clusters of up to 25 processors.
Fault Tolerance --- considerable apparent flexibility in reconfiguring around faulty nodes, and since the technology is reasonably mature and standard, robustness should not be a problem. Remotely-controlled diagnostic software programs for pinpointing failed and/or weak components.
First sale and installation of CAMPUS/800 to CSRD (February 1992)
This was a promising machine, serving an important market sector. I believe Alliant were hit by unexpected financial problems rather than technical ones.