Main Page | Report Page

 

  Computers Forum Index » Computer Architecture » Fujitsu starts shipping 800 rack 80,000 chip 'K'...

Author Message
Robert Myers...
Posted: Tue Oct 05, 2010 5:51 pm
 
Whilst checking to see if Fujitsu is still a manufacturer of
mainframes (it is), I found this

http://www.datacenterdynamics.com/ME2/dirmod.asp?sid=&nm=&type=news&mod=News&mid=9A02E3B96F2A415ABC72CB5F516B4C10&tier=3&nid=23D84DF2E79B46CB895BECA4998E58BE

80000 SPARC64 VIIIfx co-developed by Fujitsu and (clears throat)
Oracle

6-dimensional mesh torus interconnect developed by Fujitsu.

Robert.
 
Del Cecchi`...
Posted: Thu Oct 07, 2010 5:15 am
 
Robert Myers wrote:
Quote:
Whilst checking to see if Fujitsu is still a manufacturer of
mainframes (it is), I found this

http://www.datacenterdynamics.com/ME2/dirmod.asp?sid=&nm=&type=news&mod=News&mid=9A02E3B96F2A415ABC72CB5F516B4C10&tier=3&nid=23D84DF2E79B46CB895BECA4998E58BE

80000 SPARC64 VIIIfx co-developed by Fujitsu and (clears throat)
Oracle

6-dimensional mesh torus interconnect developed by Fujitsu.

Robert.


So, about the bandwidth? (sounds like 6 links per chip but no info
about the links per se)
 
Robert Myers...
Posted: Thu Oct 07, 2010 5:15 am
 
On 10/6/2010 10:50 PM, Del Cecchi` wrote:

Quote:
So, about the bandwidth? (sounds like 6 links per chip but no info about
the links per se)

I've gotten a little behind on my homework. I'll be all over it. Blue
Waters has, on its to-do list, exactly the kind of calculation that has
had me exercised since Blue Gene was announced, and I'll be surprised if
the Japanese don't do the same. There will be plenty to talk about.

Robert.
 
Brett Davis...
Posted: Thu Oct 07, 2010 5:15 am
 
In article
<6c4c7dd3-c193-4800-9e2b-c308fb84fb78 at (no spam) g18g2000yqk.googlegroups.com>,
Robert Myers <rbmyersusa at (no spam) gmail.com> wrote:

Quote:
Whilst checking to see if Fujitsu is still a manufacturer of
mainframes (it is), I found this

http://www.datacenterdynamics.com/ME2/dirmod.asp?sid=&nm=&type=news&mod=News&mid=9A02E3B96F2A415ABC72CB5F516B4C10&tier=3&nid=23D84DF2E79B46CB895BECA4998E58BE

80000 SPARC64 VIIIfx co-developed by Fujitsu and (clears throat)
Oracle

6-dimensional mesh torus interconnect developed by Fujitsu.

So now for mainframes we have POWER, a Fujitsu chip and two ASIC based fringe lines at Unisys.
The irony is that a ASIC design while slow, will keep getting "free" upgrades as the ASIC it is based on is updated.
It might never die. ;)

I checked Bull-HoneyWell-GE, Groupe Bull switched their customers to Itanic.

Any other fringe companies?

Brett
 
Michael S...
Posted: Sat Oct 09, 2010 8:34 pm
 
On Oct 7, 4:57 am, Brett Davis <gg... at (no spam) yahoo.com> wrote:
Quote:

So now for mainframes we have POWER, a Fujitsu chip and two ASIC based fringe lines at Unisys....

No, for real mainframes IBM has brand new z196 CPU. BTW, it looks
veeery fast.
On the other hand, Fujitsu mainframe-class gear is based on rather old
and slow 180nm ASIC.
Don't know about Unisys. Probably in that regard it is closer to
Fujitsu than to IBM.
That's about "big iron".

In "big tin" class IBM relies on Power7. Once again, brand new CPU
and veeery fast.
Fujitsu answer to that is SPARC64 VII. It is older and slower, but
Fujitsu tries to compensate through tying a lot of them (to be
precise, 64 CPUs, while the biggest announced Power7 system stops at
32 CPUs) into single cache-coherent system image. Still, by now, the
biggest IBM 'big tin' is more than three times faster than the biggest
baddest Fujitsu M9000.
Unisys used to build its big tin servers around Intel Itanium, then
around Intel Tigerton/Dunnington Xeons. With both uncompetitive right
now they are left with pushing the same 8-way Intel Beckton boxen as
the other guys. In effect, they compete head-to-head with the like of
Dell. Too bad, they are not going to win. May be, they should license
the ASIC from SGI/Rackable taht would let them a way into 16-32-way
Beckton servers? As crazy as it sounds it's less crazy than direct
competition with Dells of this world.

Now, Fujitsu "Venus" SPARC64 VIIIfx. that Robert mentioned in its
original post is neither big iron nor big tin. It's a supercomputer
chip with emphasis on floating point performance and throughput.
In theory, it would be possible for Fujitsu to push it into big tin,
but they would need to make significant changes to it. In its current
form, its general-purpose (i.e. non-FP) performance per core is too
low - significantly lower than previous-generation "Jupiter" SPARC64
VII. In the world where price of important software packages is
calculated "per core" such an "upgrade" would be tough to sell. I
suppose, the simplest fix would be cutting the number of cores in half
and using created thermal headroom to push up the frequency.
 
Robert Myers...
Posted: Sat Oct 09, 2010 9:34 pm
 
On Oct 9, 4:34 pm, Michael S <already5cho... at (no spam) yahoo.com> wrote:
Quote:
On Oct 7, 4:57 am, Brett Davis <gg... at (no spam) yahoo.com> wrote:



So now for mainframes we have POWER, a Fujitsu chip and two ASIC based fringe lines at Unisys....

No, for real mainframes IBM has brand new z196 CPU. BTW, it looks
veeery fast.

http://en.wikipedia.org/wiki/Talk:Mainframe_computer#What_is_it_exactly.3F

<quote>

This article meanders on through endless "some say, while others say"
and "a mainframe does this, sometimes, but not always, and are
different from servers, although not necessarily". After reading
through it I don't have any idea what a mainframe actually is due to
all the weasel words and hedging. - Keith D. Tyler ¶ 21:30, 2 January
2010 (UTC)

That's because the article contains so much falsehoods and half truths
that it should be deleted and started all over again, and not by the
IBM sales team. Mainframe is not, and never has been, an IBM only
term. Starfiend (talk) 22:57, 5 March 2010 (UTC)

</quote>

Many apparently now use the term "mainframe" to refer exclusively to
those systems that run Z/OS, even though that usage is historically
inaccurate, and that appears to be how you are using the term. IBM,
however, sells Linux-only "mainframes." Shall we call such systems
big terne-plate?

Quote:
On the other hand, Fujitsu mainframe-class gear is based on rather old
and slow 180nm ASIC.
Don't know about Unisys. Probably in that regard it is closer to
Fujitsu than to IBM.
That's about "big iron".

In "big tin" class IBM relies on Power7. Once again,  brand new CPU
and  veeery fast.

Are you saying that the processor inside new IBM mainframes really
*isn't* Power?

Quote:
Fujitsu answer to that is SPARC64 VII. It is older and slower, but
Fujitsu tries to compensate through tying  a lot of them (to be
precise, 64 CPUs, while the biggest announced Power7 system stops at
32 CPUs) into single cache-coherent system image. Still, by now, the
biggest IBM 'big tin' is more than three times faster than the biggest
baddest Fujitsu M9000.
Unisys used to build its big tin servers around Intel Itanium, then
around Intel Tigerton/Dunnington Xeons. With both uncompetitive right
now they are left with pushing the same 8-way Intel Beckton boxen as
the other guys. In effect, they compete head-to-head with the like of
Dell. Too bad, they are not going to win. May be, they should license
the ASIC from SGI/Rackable taht would let them a way into 16-32-way
Beckton servers? As crazy as it sounds it's less crazy than direct
competition with Dells of this world.

Now, Fujitsu "Venus" SPARC64 VIIIfx. that Robert mentioned in its
original post is neither big iron nor big tin. It's a supercomputer
chip with emphasis on floating point performance and throughput.
In theory, it would be possible for Fujitsu to push it into big tin,
but they would need to make significant changes to it. In its current
form, its general-purpose (i.e. non-FP) performance per core is too
low - significantly lower than previous-generation "Jupiter" SPARC64
VII. In the world where price of important software packages is
calculated "per core" such an "upgrade" would be tough to sell. I
suppose, the simplest fix would be cutting the number of cores in half
and using created thermal headroom to push up the frequency.

IBM probably needs Fujitsu the way Intel needs AMD. No irony. The
existence of nominal competition is probably very handy for IBM.

Robert.
 
Michael S...
Posted: Sat Oct 09, 2010 10:41 pm
 
On Oct 9, 11:34 pm, Robert Myers <rbmyers... at (no spam) gmail.com> wrote:
Quote:
 IBM, however, sells Linux-only "mainframes."  Shall we call such systems
big terne-plate?

There exist IBM mainframes that are prevented from running MVS (a.k.a.
zOS). However, according to my understanding, they all are based on
zVM and as such fully capable of running VM/CMS applications. So,
strictly speaking they are not Linux-only.

Quote:


Are you saying that the processor inside new IBM mainframes really *isn't* Power?

Of course, it isn't. z196 CPU natively executes zArch instruction set,
which is a 64-bit extension of S/370 which, in turn, is an extension
of S/360.
z196 is manufactured by the same IBM 45nm SOI silicon process as
Power7. Also I could be said that two chips share such a volatile
matter as "design philosophy". But that's about it. Other than that
they are substantially different.
Power7: 6-way superscalar OoO, 8 cores per chip, 4 HW threads per
core, up to 4.0 GHz or up to 4.25 GHz for "turbo" 4-core variant.
z196: 3-way superscalar OoO, 4 cores per chip, 1 (???) HW thread per
core, 6 chips are packaged together in MCM book, up to 5.2 GHz.

More details (but not much more) here:
http://www.redbooks.ibm.com/redpieces/abstracts/sg247833.html
http://www.redbooks.ibm.com/abstracts/sg247832.html
There was more detailed representation at hot chips, but I don't have
a link.


Quote:

IBM probably needs Fujitsu the way Intel needs AMD.  No irony.  The
existence of nominal competition is probably very handy for IBM.

Robert.

For "big iron" - yes.
For "big tin" - they have HP and one remaining Oracle SPARC line which
is not based on Fujitsu CPUs (Also this boxen are probably too small
to deserve big tin label). Is HP competition to IBM in this field is
nominal or real is a matter of opinion.
 
Robert Myers...
Posted: Sun Oct 10, 2010 12:21 am
 
On Oct 9, 6:41 pm, Michael S <already5cho... at (no spam) yahoo.com> wrote:
Quote:
On Oct 9, 11:34 pm, Robert Myers <rbmyers... at (no spam) gmail.com> wrote:


Quote:
IBM probably needs Fujitsu the way Intel needs AMD.  No irony.  The
existence of nominal competition is probably very handy for IBM.


For "big iron" - yes.
For "big tin" - they have HP and one remaining Oracle SPARC line which
is not based on Fujitsu CPUs (Also this boxen are probably too small
to deserve big tin label). Is HP competition to IBM in this field is
nominal or real is a matter of opinion.

The issue of interest to me is that the national prestige machines of
both the US and Japan are being built by manufacturers of "mainframes"
whose principal business interests are orthogonal to scientific
computation. The importance of that coincidence for HPC may be
insignificant. I don't even know whether it's a good development or
bad. I can think of arguments both ways. In any case, it's a
departure from what had seemed like the x86/commodity hardware
steamroller that seemed unstoppable in HPC not so very long ago.

Robert.
 
Michael S...
Posted: Sun Oct 10, 2010 11:23 pm
 
On Oct 11, 12:34 am, Brett Davis <gg... at (no spam) yahoo.com> wrote:
Quote:
In article

Opps, blunder there to call POWER a mainframe chip just because IBM currently sells
mainframes based on POWER, which confused the issue.

Please, show me a single IBM-originated quote where they calls POWER-
based computer a mainframe.

Quote:
POWER started as a Workstation CPU line, which points at another blunder in skipping
over an entire class of CPU architectures, short lived as they were.

Corrected list:

Mainframes:
   IBM Z class (harks back to System 360), and three fringe ASIC designs,
   one by Fujitsu, and two by Unisys.
MiniComputers:
   None? A market once dominated by the VAX line.

VAX is dead, but you can run VMS on Itanium. OpenVMS v.8.4 even runs
on newest Tukwilla blades.
Also, unlike IBM, HP does not prohibit people from running VAX/VMS and
AXP/OpenVMS on emulators. In fact, emulators from stromasys running on
HP Proliant are even officially supported by HP.

Quote:
   IBM had at least two major mini lines, do either of these still exist?

S/38->AS/400 still lives in form of IBM System i. The OS and other
software was migrated from original hardware to PPC ~15 years ago. By
now System i is physically identical to System p, the only difference
is software.

Quote:
Workstations:
   *Note: The workstation market no longer exists.
   SPARC, now only available as a server class CPU.
   POWER, which now spans game consoles up to mainframes.
   MIPS, was in consoles, hangs on in networking equipment and other embedded markets.
Servers:
   AMD64 now dominates, some POWER, a few fringe leftovers.

SPARC and Itanium are multi-billion $ businesses. In terms of server
sails both leg POWER by less than 50%.

Quote:
PCs:
   x86/AMD64, complete and utter dominance of one architecture.
   ARM, now in phones and iPads, and innumerable embedded markets.
Consoles:
   Reuse of old workstation CPUs: MIPS, PowerPC, and the embedded SH-2. (Sega Saturn)
Embedded 4:
   Too many architectures to count, and who cares. Wink
Embedded 8:
   Too many architectures to count. 8085 dominate? (Paired 8 bit registers do not count as 16 bit.)

8085 does not exist and never was particularly strong. By now, the
leading 8-bit architecture is Microchip PIC, although it's not easy to
qualify particular PIC variants by bittness. Atmel AVR is likely
second. 8085 probably still third.

Quote:
Embedded 16:
   Too many architectures to count. Largely extinct? (None come to mind, sorry.)

Classic 16-bit MCUs, with exception of TI MPC430, are mostly extinct.
However, few 16-bit chips that originated as DSPs are now sold as
MCUs.

Quote:
Embedded 32:
   Six or so major architectures, dominated by ARM, originally a PC design.

Much more than six architectures. Among relative newcomers Xilinx
Microblaze, Altera NiosII (both MIPS clones) and ADI Blackfin are the
most important.
ARM is strong in mobile and handheld application processors. It does
not dominate other segments although recent ARM-Cortex-M and Cortex-R
offerings does look good for market segments where ARM traditionally
had little presence.

Quote:
Embedded 64:
   MIPS, originally a workstation CPU.

There was attempt by PA Semi to compete against MIPS with 64-bit PPC.
Didn't last for long.

Quote:

Summary:
All that is left of the old computer industry is the IBM Z Class, POWER, x86, and ARM.
Most others went extinct, a few fringe designs hang on, on past life support.
Everyone else left is an also ran and/or embedded.

Corrections welcome.

Brett

Summary:
Brett Davis constantly underestimates the richness and variety of
computing world.
For example, he totally forgot supercomputers, pretty important class
of machines that started the whole discussion.
 
Robert Myers...
Posted: Mon Oct 11, 2010 12:57 am
 
On Oct 10, 7:23 pm, Michael S <already5cho... at (no spam) yahoo.com> wrote:
Quote:
On Oct 11, 12:34 am, Brett Davis <gg... at (no spam) yahoo.com> wrote:

In article

Opps, blunder there to call POWER a mainframe chip just because IBM currently sells
mainframes based on POWER, which confused the issue.

Please, show me a single IBM-originated quote where they calls POWER-
based computer a mainframe.

POWER started as a Workstation CPU line, which points at another blunder in skipping
over an entire class of CPU architectures, short lived as they were.

Corrected list:

Mainframes:
   IBM Z class (harks back to System 360), and three fringe ASIC designs,
   one by Fujitsu, and two by Unisys.
MiniComputers:
   None? A market once dominated by the VAX line.

VAX is dead, but you can run VMS on Itanium. OpenVMS v.8.4 even runs
on newest Tukwilla blades.
Also, unlike IBM, HP does not prohibit people from running VAX/VMS and
AXP/OpenVMS on emulators. In fact, emulators from stromasys running on
HP Proliant are even officially supported by HP.

   IBM had at least two major mini lines, do either of these still exist?

S/38->AS/400 still lives in form of IBM System i. The OS and other
software was migrated from original hardware to PPC ~15 years ago. By
now System i is physically identical to System p, the only difference
is software.

Workstations:
   *Note: The workstation market no longer exists.
   SPARC, now only available as a server class CPU.
   POWER, which now spans game consoles up to mainframes.
   MIPS, was in consoles, hangs on in networking equipment and other embedded markets.
Servers:
   AMD64 now dominates, some POWER, a few fringe leftovers.

SPARC and Itanium are multi-billion $ businesses. In terms of server
sails both leg POWER by less than 50%.


PC's with Xeon are generally marketed and priced like workstations or
servers.

Quote:
PCs:
   x86/AMD64, complete and utter dominance of one architecture.
   ARM, now in phones and iPads, and innumerable embedded markets.
Consoles:
   Reuse of old workstation CPUs: MIPS, PowerPC, and the embedded SH-2. (Sega Saturn)
Embedded 4:
   Too many architectures to count, and who cares. Wink
Embedded 8:
   Too many architectures to count. 8085 dominate? (Paired 8 bit registers do not count as 16 bit.)

8085 does not exist and never was particularly strong. By now, the
leading 8-bit architecture is Microchip PIC, although it's not easy to
qualify particular PIC variants by bittness. Atmel AVR is likely
second. 8085 probably still third.

Embedded 16:
   Too many architectures to count. Largely extinct? (None come to mind, sorry.)

Classic 16-bit MCUs, with exception of TI MPC430,  are mostly extinct.
However, few 16-bit chips that originated as DSPs are now sold as
MCUs.

Embedded 32:
   Six or so major architectures, dominated by ARM, originally a PC design.

Much more than six architectures. Among relative newcomers Xilinx
Microblaze, Altera NiosII (both MIPS clones) and ADI Blackfin are the
most important.
ARM is strong in mobile and handheld application processors. It does
not dominate other segments although recent ARM-Cortex-M and Cortex-R
offerings does look good for market segments where ARM traditionally
had little presence.

Embedded 64:
   MIPS, originally a workstation CPU.

There was attempt by PA Semi to compete against MIPS with 64-bit PPC.
Didn't last for long.



Summary:
All that is left of the old computer industry is the IBM Z Class, POWER, x86, and ARM.
Most others went extinct, a few fringe designs hang on, on past life support.
Everyone else left is an also ran and/or embedded.

Corrections welcome.


Summary:
Brett Davis constantly underestimates the richness and variety of
computing world.
For example, he totally forgot supercomputers, pretty important class
of machines that started the whole discussion.

Supercomputers are easy to overlook from a business point of view,
unless you are in marketing, in which case they box well over their
weight as contributors to company credibility, visibility, image, and
sizzle.

Robert.
 
robertwessel2 at (no spam) yahoo.com...
Posted: Mon Oct 11, 2010 2:33 am
 
On Oct 10, 6:23 pm, Michael S <already5cho... at (no spam) yahoo.com> wrote:
Quote:
On Oct 11, 12:34 am, Brett Davis <gg... at (no spam) yahoo.com> wrote:
Embedded 8:
   Too many architectures to count. 8085 dominate? (Paired 8 bit registers do not count as 16 bit.)

8085 does not exist and never was particularly strong. By now, the
leading 8-bit architecture is Microchip PIC, although it's not easy to
qualify particular PIC variants by bittness. Atmel AVR is likely
second. 8085 probably still third.


I assume you meant 8051? The 8085 was a (slight) 8080 upgrade.
 
Brett Davis...
Posted: Mon Oct 11, 2010 2:34 am
 
In article
<4949974f-e46e-4d6d-a3aa-9f7975f6a805 at (no spam) i21g2000yqg.googlegroups.com>,
Michael S <already5chosen at (no spam) yahoo.com> wrote:
Quote:
On Oct 7, 4:57 am, Brett Davis <gg... at (no spam) yahoo.com> wrote:
So now for mainframes we have POWER, a Fujitsu chip and two ASIC based fringe lines at Unisys....

No, for real mainframes IBM has brand new z196 CPU. BTW, it looks
veeery fast.
On the other hand, Fujitsu mainframe-class gear is based on rather old
and slow 180nm ASIC.
Don't know about Unisys. Probably in that regard it is closer to
Fujitsu than to IBM.
That's about "big iron".

Opps, blunder there to call POWER a mainframe chip just because IBM currently sells
mainframes based on POWER, which confused the issue.
POWER started as a Workstation CPU line, which points at another blunder in skipping
over an entire class of CPU architectures, short lived as they were.

Corrected list:

Mainframes:
IBM Z class (harks back to System 360), and three fringe ASIC designs,
one by Fujitsu, and two by Unisys.
MiniComputers:
None? A market once dominated by the VAX line.
IBM had at least two major mini lines, do either of these still exist?
Workstations:
*Note: The workstation market no longer exists.
SPARC, now only available as a server class CPU.
POWER, which now spans game consoles up to mainframes.
MIPS, was in consoles, hangs on in networking equipment and other embedded markets.
Servers:
AMD64 now dominates, some POWER, a few fringe leftovers.
PCs:
x86/AMD64, complete and utter dominance of one architecture.
ARM, now in phones and iPads, and innumerable embedded markets.
Consoles:
Reuse of old workstation CPUs: MIPS, PowerPC, and the embedded SH-2. (Sega Saturn)
Embedded 4:
Too many architectures to count, and who cares. Wink
Embedded 8:
Too many architectures to count. 8085 dominate? (Paired 8 bit registers do not count as 16 bit.)
Embedded 16:
Too many architectures to count. Largely extinct? (None come to mind, sorry.)
Embedded 32:
Six or so major architectures, dominated by ARM, originally a PC design.
Embedded 64:
MIPS, originally a workstation CPU.

Summary:
All that is left of the old computer industry is the IBM Z Class, POWER, x86, and ARM.
Most others went extinct, a few fringe designs hang on, on past life support.
Everyone else left is an also ran and/or embedded.

Corrections welcome.

Brett
 
Anne & Lynn Wheeler...
Posted: Mon Oct 11, 2010 5:10 am
 
Brett Davis <ggtgp at (no spam) yahoo.com> writes:
Quote:
Mainframes:
IBM Z class (harks back to System 360), and three fringe ASIC designs,
one by Fujitsu, and two by Unisys.
MiniComputers:
None? A market once dominated by the VAX line.
IBM had at least two major mini lines, do either of these still exist?

43xx (360 "mainframe") sold into the same mid-range market as VAX in the
same time frame and in the one-to-few order market, sold similar
volumes. the big difference with total 43xx volumes were large
multi-hundred orders by large corporations ... sort of the leading edge
of distributed computing & departmental servers.

internally, the proliferation of 43xx machines contributed to scarcity
of conference rooms (as they were being taken over for 43xx
machines). the internal network had been larger than the
arpanet/internet from just about the beginning until possibly late '85
or early '86 ... and the large proliferation of 43xx in the early 80s
contributed to internal network passing 1000 nodes when arpanet/internet
wasn't much more than 255 nodes.

misc. old post showing decade of vax sales sliced&diced with year,
model, us/non-us, etc
http://www.garlic.com/~lynn/2002f.html#0

mid-range started to fall to large PCs and workstations in the mid-80s
.... can be seen in the vax volumes. in this time-frame, the 4341
follow-on, ... i.e 4381 was expected to see similar volume growth as
experienced by 4341 in late 70s/early 80s ... but the mid-range market
had already started to move.

circa 1980, there was a program to migrate the large number of different
internal microprocessors to 801/risc (line of iliad chips) ... including
the microprocessor for 4381 and the microprocessor for the s/38
follow-on in the as/400. that effort floundered for various reasons
.... and there was another round of cisc microprocessors (and some number
of the engineers left and show up working on risc efforts at other
vendors).

as it happened ... as/400 eventually did migrate to a 801/risc chip in
the 90s (power/pc).

misc. old email references mentioning 43xx
http://www.garlic.com/~lynn/lhwemail.html#43xx

misc. old eamil references mentioning 801, risc, iliad, etc
http://www.garlic.com/~lynn/lhwemail.html#801

misc. old posts mentioning internal network
http://www.garlic.com/~lynn/subnetwork.html#internalnet

--
virtualization experience starting Jan1968, online at home since Mar1970
 
Anne & Lynn Wheeler...
Posted: Mon Oct 11, 2010 7:18 pm
 
re:
http://www.garlic.com/~lynn/2010n.html#69 Fujitsu starts shipping 800 rack 80,000 chip 'K' supercomputer

a couple supercomputer related announcements from early 90s

"scientific and technical *only*"
http://www.garlic.com/~lynn/2001n.html#6000clusters1 2/17/92 article
"caught by *surprise*"
http://www.garlic.com/~lynn/2001n.html#6000clusters2 06/19/92 article

post referencing meeting early jan92 when it was still both commercial
and scientific/technical
http://www.garlic.com/~lynn/95.html#13

and then old email from the end of jan ... when it was still commercial
and scientific/technical ... but possibly just hrs before it was
transferred and we were told we couldn't work on anything with more than
four processors
http://www.garlic.com/~lynn/2006x.html#email920129

other email regarding cluster scaleup work for both commercial and
scientific/technical (before being told we couldn't work on anything
with more than four processors)
http://www.garlic.com/~lynn/lhwemail.html#medusa

--
virtualization experience starting Jan1968, online at home since Mar1970
 
Owen Shepherd...
Posted: Mon Oct 11, 2010 10:19 pm
 
On 11/10/2010 00:23, Michael S wrote:
Quote:
Embedded 8:
Too many architectures to count. 8085 dominate? (Paired 8 bit registers do not count as 16 bit.)

8085 does not exist and never was particularly strong. By now, the
leading 8-bit architecture is Microchip PIC, although it's not easy to
qualify particular PIC variants by bittness. Atmel AVR is likely
second. 8085 probably still third.

Last time I saw an article on it, it put things as 8051 being in the
lead (primarily due to multiple sources), followed by Microchip PIC
(Microchip being the biggest single manufacturer of 8-bitters). I don't
know where AVR falls on the scale.

There are also a variety of these common in FPGAs, e.g. Mico8 and PicoBlaze

Quote:
Embedded 16:
Too many architectures to count. Largely extinct? (None come to mind, sorry.)

Classic 16-bit MCUs, with exception of TI MPC430, are mostly extinct.
However, few 16-bit chips that originated as DSPs are now sold as
MCUs.

16-bit still holds on (though probably not for too much longer) in
automotive space, e.g. Freescale S12/S12X. The Microchip PIC24s/dsPICs
also pack quite a punch at a very good price point.

Quote:
Embedded 32:
Six or so major architectures, dominated by ARM, originally a PC design.

Much more than six architectures. Among relative newcomers Xilinx
Microblaze, Altera NiosII (both MIPS clones) and ADI Blackfin are the
most important.
ARM is strong in mobile and handheld application processors. It does
not dominate other segments although recent ARM-Cortex-M and Cortex-R
offerings does look good for market segments where ARM traditionally
had little presence.

Shipping architectures I know of (that aren't NRND) in off-the-shelf
devices:
* ARM (ARM7/9/Cortex M/R)
* MIPS M4K (Primarily Microchip PIC32)
* ColdFire (RISCified m68k)
* PowerPC
* AVR32
* SuperH
* BlackFin
* The 5 or so that Renesas has inherited

Others in more niche deployments include SPARC (ESA use SPARCs to
control their spacebound hardware). ARM, NIOS, Microblaze, ZPU and
Mico32 often turn up inside FPGAs and less often ASICs.

I bet one could fit much, much more in here, but those are just the ones
which come to mind

- Owen
 
 
Page 1 of 2    Goto page 1, 2  Next
All times are GMT
The time now is Tue Sep 02, 2014 2:28 pm