NHSE ReviewTM 1996 Volume First Issue

Overview of Recent Supercomputers

| <- HREF="node42.html" Prev | Index | Next -> |
NHSE ReviewTM: Comments · Archive · Search


Chapter 4 -- Systems Disappeared from the List

As already stated in the introduction the list of systems is not complete. On one hand this is caused by the sheer number of systems that are presented to the market and are often very similar to systems described above (for instance, the Volvox system not listed is very similar but not equivalent to the listed Alex AVX system and there are numerous other examples). On the other hand there are many systems that are still in operation around the world, often in considerable quantities that for other reasons are excluded. The most important reasons are: Below we present a table of systems that fall into one of the categories mentioned above. We think this may have some sense to those who come across machines that are still around but are not the latest in their fields. It may be interesting at least to have an indication how such systems compare to the newest ones and to place them in context.

It is good to realise that although systems have disappeared from the section above they still may exist and are actually sold. However, their removal stems in such cases mainly from the fact that they are not serious candidates for high-performance computing anymore.

The table is, again, not complete and admittedly somewhat arbitrary. The data are in a highly condensed form: the system name, system type, theoretical maximum performance of a fully configured system, and the reason for their disappearance is given. The arbitrariness lies partly in the decision of which systems are still sufficiently of interest to include and which are not. For instance, the Convex C-1 is not included, while the Alliant FX/80 is. The reason is that the C-1 is conceptually not different from the later generations of single-processor Convex vector processors, while the Alliant FX/80 was fairly different from its successor the Alliant FX/2800.

Machine: Alliant FX/2800.
Type: Shared memory vector-parallel, max. 28 processors.
Theoretical Peak performance: 1120 Mflop/s
Reason for disappearance:Manufacturer out of business.

Machine: BBN TC2000.
Type: Virtual shared memory parallel, max. 512 processors.
Theoretical Peak performance: 1 Gflop/s
Reason for disappearance: Manufacturer has discontinued marketing parallel computer systems.

Machine: Cambridge Parallel Processing DAP Gamma.
Type: Distributed memory processor array system.
Theoretical Peak performance: 1.6 Gflop/s (32-bit)
Reason for disappearance: replaced by newer Gamma II series (see 3.2.2).

Machine: Convex C3200, C3400, C3800.
Type: Shared memory vector-parallel, max. 8 processors (C3880).
Theoretical Peak performance: 960 Mflop/s
Reason for disappearance: replaced by newer C4 series (see 3.3.3).

Machine: Convex Meta Series.
Type: Distributed memory network of workstations.
Theoretical Peak performance: 200 Mflop/s per processor
Reason for disappearance: replaced by newer SPP-1200 series (see 3.4.8).

Machine: Convex SPP-1000.
Type: Distributed memory RISC based system, max. 128 processors.
Theoretical Peak performance: 25.6 Gflop/s
Reason for disappearance: replaced by newer SPP-1200 series (see 3.4.8).

Machine: Cray Computer Corporation Cray-2.
Type: Shared memory vector-parallel, max. 4 processors.
Theoretical Peak performance: 1.95 Gflop/s
Reason for disappearance: Manufacturer out of business.

Machine: Cray Computer Corporation Cray-3.
Type: Shared memory vector-parallel, max. 16 processors.
Theoretical Peak performance: 16 Gflop/s
Reason for disappearance: Manufacturer out of business.

Machine: Cray Research Inc. APP.
Type: Shared memory RISC based system, max. 84 processors.
Theoretical Peak performance: 6.7 Gflop/s
Reason for disappearance: Product line discontinued, gap expected to be filled by Cray J90 (see 3.3.1).

Machine: Cray T3D.
Type: Distributed memory RISC based system, max. 2048 processors.
Theoretical Peak performance: 307 Gflop/s
Reason for disappearance: replaced by newer T3E (see 3.4.4).

Machine: Cray Research Inc. Cray Y-MP, Cray Y-MP M90.
Type: Shared memory vector-parallel, max. 8 processors.
Theoretical Peak performance: 2.6 Gflop/s
Reason for disappearance: replaced by newer T90 (see 3.3.1).

Machine: Cray Y-MP C90.
Type: Shared memory vector-parallel, max. 16 processors.
Theoretical Peak performance: 16 Gflop/s
Reason for disappearance: replaced by newer T90 (see 3.3.1).

Machine: Digital Equipment Corp. Alpha farm.
Type: Distributed memory RISC based system, max. 4 processors.
Theoretical Peak performance: 0.8 Gflop/s
Reason for disappearance: replaced by newer AlphaServer clusters (see 3.3.4).

Machine: Fujitsu VPP500 series.
Type: Distributed memory multi-processor vectorprocessors, max. 222 processors.
Theoretical Peak performance: 355 Gflop/s
Reason for disappearance: replaced by the VPP300 series (see 3.4.6).

Machine: Fujitsu VPX200 series.
Type: Single-processor vectorprocessors.
Theoretical Peak performance: 5 Gflop/s
Reason for disappearance: replaced by the VPP300 series (see 3.4.6).

Machine: Hitachi SR2001 series.
Type: Distributed memory RISC based system, max. 128 processors.
Theoretical Peak performance: 23 Gflop/s
Reason for disappearance: Replaced by the newer SR2201 (see 3.4.7).

Machine: IBM ES/9000 series.
Type: Shared memory vector-parallel system, max. 6 processors.
Theoretical Peak performance: 2.67 Gflop/s
Reason for disappearance: IBM does not pursue high-performance computing by this product line anymore.

Machine: IBM Power/4.
Type: Shared memory RISC based system, max. 4 processors.
Theoretical Peak performance: 336 Mflop/s
Reason for disappearance: Product line discontinued, gap expected to be filled by SP2 (see 3.4.9).

Machine: IBM SP1 series.
Type: Distributed memory RISC based system, max. 64 processors.
Theoretical Peak performance: 8 Gflop/s
Reason for disappearance: Replaced by the newer SP2 (see 3.4.9).

Machine: Intel iPSC/860.
Type: Distributed memory parallel hypercube, max. 128 processors.
Theoretical Peak performance: 7.7 Gflop/s
Reason for disappearance: replaced by newer Intel Paragon XP (MP) series (see 3.4.10).

Machine: Kendall Square Research KSR2.
Type: Virtually shared memory parallel, max. 1088 processors.
Theoretical Peak performance: 400 Gflop/s
Reason for disappearance: Kendall Square has terminated its business.

Machine: Meiko CS-1 series.
Type: Distributed memory RISC based system.
Theoretical Peak performance: 80 Mflop/s per processor
Reason for disappearance: Replaced by the newer CS-2 (see 3.4.12).

Machine: NEC SX-2.
Type: Single-processor vector processors.
Theoretical Peak performance: 1.3 Gflop/s
Reason for disappearance: replaced by newer SX-4 series (see 3.3.5).

Machine: NEC SX-3R.
Type: Shared memory multi-processor vector processors, max. 4 processors.
Theoretical Peak performance: 1.3 Gflop/s
Reason for disappearance: replaced by newer SX-4 series (see 3.3.5).

Machine: Parsys SN9000 series.
Type: Distributed memory RISC based system, max. 2048.
Theoretical Peak performance: 51.2 Gflop/s
Reason for disappearance: Replaced by the newer TA9000 (see 3.4.16).

Machine: Siemens-Nixdorf VP2600 series.
Type: Single-processor vectorprocessors.
Theoretical Peak performance: 5 Gflop/s
Reason for disappearance: replaced by the VPP300 series (see 3.4.6).

Machine: Stern Computing Systems SSP.
Type: Shared memory multi-processor, max. 6 processors.
Theoretical Peak performance: 2 Gflop/s
Reason for disappearance: Vendor terminated its business just before delivering first systems.

Machine: Thinking Machine Corporation CM-2(00).
Type: SIMD parallel machine with hypercube structure, max. 64K processors.
Theoretical Peak performance: 31 Gflop/s
Reason for disappearance: was replaced by the newer CM-5 (but see below).

Machine: Thinking Machine Corporation CM-5.
Type: Distributed memory RISC based system, max. 16K processors.
Theoretical Peak performance: 2 Tflop/s
Reason for disappearance: Thinking Machine Corporation has stopped manufacturing hardware and hopes to keep alive as a software vendor.

Machine: Transtech Paramid series.
Type: Distributed memory RISC based system, max. 64 processors.
Theoretical Peak performance: 6.4Gflop/s
Reason for disappearance: Transtech now mostly manufactures PC extension boards with IBM MC603 processors as performance boosters.

Copyright © 1996 Aad J. van der Steen and Jack J. Dongarra


| <- HREF="node42.html" Prev | Index | Next -> |
NHSE ReviewTM: Comments · Archive · Search
NHSE: Software Catalog · Roadmap


Copyright © 1996 NHSE ReviewTM All Rights Reserved.
Lowell W Lutz (lwlutz@rice.edu) NHSE ReviewTM WWWeb Editor