CDC CYBER 2000's!!!!!!!!
g_ah...@vaxa.uwa.oz
has twice the power of the current high-end CYBER 990. The company also
announced a new disk subsystem for its entire CYBER mainframe line.
The new hi-end machine, the CYBER 2000 supermainframe, is aimed at scientists,
engineers and other users of high-performance, general purpose computers. The
CYBER 2000 is aircooled and uses the densest ECL gate-array technology in the
industry. It has the fastest processors in the mainframe class.
"The CYBER 2000 is a machine for the most demanding workloads," said James E
Ousley, president of Control Data's Computer Products Group. "Today,
high-performance computer users want more than just computational processing.
Besides this, they require high-performance computers and software that also
analyse, control and manage large amounts of complex data in large enterprises.
"The CYBER 2000 gives high-performance information management functionality,
with the power to handle computational problems and transaction processing."
Control Data identifies the machine as a "supermainframe" because it has
characteristics of both a supercomputer and a mainframe.
When running the standard 100x100 LINPACK benchmark, a common benchmark for
supercomputers, the CYBER 2000 performs at 30 MFLOPS. The Cray 2 runs at 38
MFLOPS. Peak performance on the CYBER 2000 is 220 MFLOPS per processor. Channel
speed on the new supermainframe is the industry's fastest at 25 megabytes per
second.
The CYBER 2000 was designed using a new 14,000 gate logic chip. Most
high-performance mainframes use 2,000 to 6,000 gate ECL arrays or slower CMOS
arrays. ECL chip density at this level translates into higher performance,
increased reliability and lower maintenance.
Two cache memories with 5 nanosecond chips have been separated on the new
machine: one for data and one for instructions. With data ready to access in
large, fast cache memories, the system doesn't have to access main memory as
often, which significantly increases performance.
There are two CYBER 2000 models, which will be available in late 1990 in
single- and dual-processor versions. The CYBER 2000V provides both vector and
scalar processing capabilities for users who have computationally intensive
applications. The CYBER 2000S is streamlined for database management and
information processing. The CYBER 2000S is field upgradable to a CYBER 2000V.
Both run Control Data's NOS/VE operating system.
The new Disk Array Subsystem (DAS) 5830 for the CYBER mainframe line provides
nearly eight times the data transfer rate and four times the capacity of
Control Data's previous disk system. It can transfer data almost six times
faster per channel than an ibm 3380 disk system.
The DAS introduces several new technologies that provide a breakthrough to
levels of speed, capacity and reliability not found on other systems. The
combination of solid-state disk and magnetic disk technologies enables the
CYBER mainframe to optimize data transfer and access. A new array controller
allows a site manager to configure disks so the mainframe can read or write
data across multiple storage devices simultaneously. The new product's
sustainable transfer rate jumps from 2.5 megabytes per second on the 9853 disk
subsystem to 16.7 megabytes per second on the DAS.
The new disk subsystem adds significant capacity whereby a single subsystem
goes from 4.4 gigabytes of storage on the 9853 to 32 gigabytes on the new
system.
The array controller also provides a parity drive that checks for errors and
automatically replaces affected data without interruption.
The Disk Array Subsystem with capability for serial operation will be available
in June 1990, and for parallel operations in the fourth quarter of 1990.
89-086
2097H
END OF NOTICE 2 (shortened)
Seems this is where the ETA technology has really hit the spot. The CYBER
2000's have the appearance of a CYBER 960 without the hightop angled cabinet.
According to the benchmark figure from the newest CDC FOCUS magazine, this
beauty blitzes practically all of it's "competitors", and at a price without
competition. Great Work Guys!!!!!!!
TP1 benchmark figures :
1,500 transactions per second = CDC CYBER 2000 !!!!!
1,000 transactions per second = Honeywell DPS 9000
350 transactions per second = IBM Sierra 3090-180S
250 transactions per second = Amdahl 5890-600E
50 transactions per second = digital VAX 8600
Fortran Kernel figures :
57 MIPS = CDC CYBER 2000V
40 MIPS = Cray X/MP
35 MIPS = CDC CYBER 2000S
25 MIPS = CDC CYBER 990
26 MIPS = IBM Sierra 3090-180A
LINPACKS 100x100 benchmark figures :
84 MFLOPS = Cray Y/MP
53 MFLOPS = Cray X/MP
38 MFLOPS = Cray 2
30 MFLOPS = CDC CYBER 2000V
19.2 MFLOPS = IBM Sierra 3090-180A
19 MFLOPS = Amdahl 1400
16 MFLOPS = Convex C-210
12 MFLOPS = CDC CYBER 992 31
5 MFLOPS = Honeywell DPS 90
3.1 MFLOPS = CDC CYBER 962 31
Also introduced was a new cartridge drive the 5680
Tim_CDC...@cup.portal.com
Can anyone tell me if pricing was discussed at the announcement(s), either
in Paris or at VIM, or were you told to talk to your local sales
representative? It's not necessary to post the pricing, I just want to
know if it was publicly discussed.
Tim_CDC...@cup.portal.com | I Survived The
...!sun!portal!cup.portal.com!tim_cdc_roberts | Great Quake of '89.
Rob Peglar x615
(verbiage deleted)
> Control Data identifies the machine as a "supermainframe" because it has
> characteristics of both a supercomputer and a mainframe.
Does that mean it's not as powerful as a supercomputer, yet costs more
than a mainframe? 1/2 :-)
>
> When running the standard 100x100 LINPACK benchmark, a common benchmark for
> supercomputers, the CYBER 2000 performs at 30 MFLOPS. The Cray 2 runs at 38
> MFLOPS. Peak performance on the CYBER 2000 is 220 MFLOPS per processor. Channel
> speed on the new supermainframe is the industry's fastest at 25 megabytes per
> second.
Oops. See the thread about Linpack running in comp.arch.
>
> There are two CYBER 2000 models, which will be available in late 1990 in
> single- and dual-processor versions. The CYBER 2000V provides both vector and
> scalar processing capabilities for users who have computationally intensive
> applications. The CYBER 2000S is streamlined for database management and
> information processing. The CYBER 2000S is field upgradable to a CYBER 2000V.
> Both run Control Data's NOS/VE operating system.
^^^^^^
Huh? If so, what's all this re-org at AHSE about? Oops.
Rob
--
Rob Peglar Control Systems, Inc. 2675 Patton Rd., St. Paul MN 55113
...uunet!csinc!rpeglar 612-631-7800
The posting above does not necessarily represent the policies of my employer.
Philip Leverton
>>
>> There are two CYBER 2000 models, which will be available in late 1990 in
>> single- and dual-processor versions. The CYBER 2000V provides both vector and
>> scalar processing capabilities for users who have computationally intensive
>> applications. The CYBER 2000S is streamlined for database management and
>> information processing. The CYBER 2000S is field upgradable to a CYBER 2000V.
>> Both run Control Data's NOS/VE operating system.
> ^^^^^^
>Huh? If so, what's all this re-org at AHSE about? Oops.
What exactly do you mean by this? It's not very surprising that VE will be
the o/s of the Cyber 2000's since it appears that native UNIX on the 900 series
will not ever be available as a released product. The 2000 machines will provide
and upgrade path for people running VE on the 800 and 900 series machines.
(ok, this is a proprietary path, but that goes without saying if we're talking
about VE). Does anyone know what the "re-org at AHSE" is about?
It could be to do with the deal announced with MIPS, to use MIPS CPU
technology (not CDC) and CDC's expertise with periperals, I/O, controllers
and other non-CPU bits to build UNIX machines. Let's hear some unbridled
speculation about this! (facts? who needs facts?)
VE at last is getting to be a usable o/s, more like what it should have
been in the first releases. The high development rate of VE (about two releases
a year) means that trying to keep software developed on VE in line with VE
is like trying to hit a moving target. Shouldn't really complain though I
suppose, since VE is getting better.
Philip Leverton
Systems Group
University of Melbourne
Rob Peglar x615
>
> What exactly do you mean by this? It's not very surprising that VE will be
> the o/s of the Cyber 2000's since it appears that native UNIX on the 900 series
> will not ever be available as a released product. The 2000 machines will provide
> and upgrade path for people running VE on the 800 and 900 series machines.
> (ok, this is a proprietary path, but that goes without saying if we're talking
> about VE). Does anyone know what the "re-org at AHSE" is about?
> It could be to do with the deal announced with MIPS, to use MIPS CPU
> technology (not CDC) and CDC's expertise with periperals, I/O, controllers
> and other non-CPU bits to build UNIX machines. Let's hear some unbridled
> speculation about this! (facts? who needs facts?)
OK.
1. I'm not at all surprised by anything CDC does anymore. Just when you
think you've got them figured out (e.g. native unix on 9xx) they reverse
all engines.
2. VE on Cyber whatevers (8xx,9xx,2xxx,...) is just more of the same
for present CDC customers. And, that's all well and good. As Eugene
said (paraphrasing), "NOS is good. Let 'em run NOS". For the rest of
us, however, it's truly disappointing. Just wasted opportunity for a larger
audience to use some fine h/w.
3. My statement about "the re-org at AHSE" is with tongue firmly in cheek.
If you'd like to discuss the fine points of the re-org, from one who has
a fairly close contact :-) at AHSE (well, used to...:-() e-mail me privately.
No sense spilling the beans over the net.
>
> VE at last is getting to be a usable o/s, more like what it should have
> been in the first releases. The high development rate of VE (about two releases
> a year) means that trying to keep software developed on VE in line with VE
> is like trying to hit a moving target. Shouldn't really complain though I
> suppose, since VE is getting better.
Well, don't count on too many more new features in VE. Anyway, I'm glad you
are (relatively) satisfied with VE. However, just as it happened in NOS, you
are among a dwindling set of people. Kinda like being on an island - life
is good, but the isolation is terrible and gets worse with time.
Rob
>
> Philip Leverton
> Systems Group
> University of Melbourne
Tim_CDC...@cup.portal.com
When considering the future of NOS/VE, it is important to remember that
Corps of Engineers contract. We have committed to supporting them until
at least 1995, and they will be running /VE. We can't (and won't) let it
shrivel up with such a major contract on the line. Plus, bear in mind
that the ARHOPS re-org was designed and implemented PRIOR TO the COE
award.
Also, remember that most of NOS/VE is written in Cybil. Cybil compilers
already exist for the 170 instruction set, the 180 instruction set, and
the 68000 instruction set. Speculate on this: what would happen if some
enthusiastic systems programmer were to take it upon himself to generate
a code-generator backend for Cybil that supported the MIPS 6280 instruction
set? What if he went one step (actually several steps) further and ported
the NOS/VE System Core? Ah HA - now the future of NOS/VE begins to look
brighter. Brings a whole new light to our MIPS Agreement, in which we
promise to "bring our data center operations experience to the RISC
marketplace", doesn't it?
Tim_CDC...@cup.portal.com | I survived the
...!sun!portal!cup.portal.com!tim_cdc_roberts | Great Quake of '89
Jerry Kreuscher
From: Tim_CDC...@cup.portal.com
[ ... ]
Also, remember that most of NOS/VE is written in Cybil. Cybil compilers
already exist for the 170 instruction set, the 180 instruction set, and
the 68000 instruction set. Speculate on this: what would happen if some
enthusiastic systems programmer were to take it upon himself to generate
a code-generator backend for Cybil that supported the MIPS 6280 instruction
set? What if he went one step (actually several steps) further and ported
the NOS/VE System Core? Ah HA - now the future of NOS/VE begins to look
brighter. Brings a whole new light to our MIPS Agreement, in which we
promise to "bring our data center operations experience to the RISC
marketplace", doesn't it?
This is a chain of reason worthy of Mae Brussel.
The existence of code generators for cybil on three machines by no
means implies that there will be or that there should be a cybil code
generator for a fourth machine. And the likelihood that a single
zealot, or even a handful of zealots, could port cybil and the rest of
nos/ve to a new architecture as a pirate project is not worth
consideration.
However, the management of that company has shown a remarkable
inability to distinguish cost from value. "Leveraging" off all that
cybil and nosve "technology" could be irresistible to them. How well
they resist that urge will be the real measure of whether they can
turn the company around.
This is my opinion, in case anyone wonders.
Jerry Kreuscher (g...@sun.com)