64 Cores -- IBM is showing a prototype already
Walter Bushell
some incentive to parallel processing.
<http://news.cnet.com/8301-13924_3-10249366-64.html?tag=newsEditorsPicksA
rea.0>
Ahem A Rivet's Shot
I can hear it already - 64 cores should be enough for anybody.
--
Steve O'Hara-Smith | Directable Mirror Arrays
C:>WIN | A better way to focus the sun
The computer obeys and wins. | licences available see
You lose and Bill collects. | http://www.sohara.org/
jmfbahciv
> On Tue, 26 May 2009 22:48:14 -0400
> Walter Bushell <pr...@panix.com> wrote:
>
>> Eight eight cores processors on a high end server. But, this should give
>> some incentive to parallel processing.
>> <http://news.cnet.com/8301-13924_3-10249366-64.html?tag=newsEditorsPicksA
>> rea.0>
>
> I can hear it already - 64 cores should be enough for anybody.
>
Not the way the discussions are going here.
/BAH
philo
64 cores.
can't wait.
I may finally get a good score in Tetris!!!!
Quadibloc
> I can hear it already - 64 cores should be enough for anybody.
One core that was 64 times as fast would be better.
While one can indeed laugh at statements made in the past... 64 K of
RAM should be enough for anybody... when it comes to computing, a
machine able to run Windows 3.1 probably is enough for what most
desktops are used for, except multimedia.
More processor power will always be useful, even if, in practice, for
many applications, it is difficult to utilize more than 2 or 3 cored
efficiently. There will always be some that will gain from 64 cores or
512 cores if they can have them..
John Savard
jmfbahciv
Man! I'm going have so many communication problems. I still say
core when I talk about physical memory. Why in the whole fucking
world did was the term "core" transformed to mean CPU?
/BAH
Quadibloc
> Quadibloc wrote:
> > More processor power will always be useful, even if, in practice, for
> > many applications, it is difficult to utilize more than 2 or 3 cored
> > efficiently. There will always be some that will gain from 64 cores or
> > 512 cores if they can have them..
>
> Man! I'm going have so many communication problems. I still say
> core when I talk about physical memory. Why in the whole fucking
> world did was the term "core" transformed to mean CPU?
Context usually distinguishes from core as the equivalent to RAM and
core as the equivalent of processor.
Core memory depends on little doughnuts of ferrite that have their
center removed... while the processor is the thing at the heart, the
kernel, the nucleus... and hence the core... of a microprocessor chip.
Now that a microprocessor chip also contains multiple processors in a
parallel sysplex (now called SMP, symmetric multiprocessing)
configuration, and their level 2 shared cache, whereas before a
microprocessor was only a CPU, using the older term, "processor",
might cause confusion over whether you're talking about the chip, or
one of the CPUs inside the chip. That's why they felt they needed a
different word when talking about the heart of the chip. The term
"kernel" has a technical meaning when talking about operating systems.
But there's more to it than core memory being out of date, and "heart"
or "nucleus" being too emotive or something. Originally, the term
"core" was used not by programmers, for whom core was indeed a common
synonym for RAM long into the PC era, but by *chipmakers*. They're the
ones who first spoke in terms of "putting two CPU cores on a die". The
little square things that they cut a wafer into and then put inside a
package with pins are called "dice" when it's necessary to use
something unambiguous (chip could mean the die, or the package
containing a die).
Chipmakers didn't talk about 'core', they only talked about DRAM and
SRAM as kinds of chips they made, so the term, when it originated
among them, caused _them_ no confusion. The term then moved out from
their quarters into the wider world, where it did have the problem you
note.
John Savard
Bernd Felsche
Lack of imagination. :-)
Each "core" is a processor core; i.e. semi-autonomous CPU, though
often sharing one or two levels of cache with other cores on the
same die (chip).
Speaking of core; far more interesting architecturally is the
recent appearance of block-addressable solid-state memory. It
probably won't be very long before systems provide dedicated "memory
core slots" into which several hundred gigabytes of solid-state
memory can be plugged; accessable at near-RAM speeds.
(I'm not aware of how JEDEC is actually progressing on standards
right now.)
Price is still an issue in $/gigabyte but bleeding-edge system are
getting in on the IOPS/$ (IO per second, per dollar) where solid
state is about 1000 times faster than rotating discs - and 10 times
more reliable. This is already providing performance breakthroughs
especially in transaction-processing systems by what is effectively
a stop-gap implementation, usually a PCIe card, as normal hard disc
interfaces don't have the bandwidth.
Such solid state technology forces a major paradigm shift in all
aspects of computer systems design. Hardware and software.
Software designers need to go back to the drawing board in their
design of systems that deal with large amounts of data. It may well
be less efficient to have a buffer cache (as one would for data on
"mechanical" storage devices); than simply go directly to the data.
Software that already supports memory-mapped files could benefit
soonest; without additional application software code needing to be
written. The operating system can manage the block-access
transparently, albeit initially delayed via VM based on a buffer
cache.
Other software which has intrinsic design ties to "mechanical"
storage devices will have to go back to the drawing board,
discarding many assumptions about systems that are no longer valid.
Things that need to be considered are the ability to map all storage
into available address space; which is at present viable for 64-bit
systems; but also extending the paradigm to include all potential
storage outside of the system (i.e. networked or potentially
connected) in the same memory-mapped way to unify the access models.
Does one utilise the momentum of the solid-state-storage paradigm
shift to make that next big leap?
--
/"\ Bernd Felsche - Innovative Reckoning, Perth, Western Australia
\ / ASCII ribbon campaign | When we remember that we are all mad,
X against HTML mail | the mysteries disappear and life stands
/ \ and postings | explained. -- Mark Twain
Christopher C. Stacy
> jmfbahciv <jmfbahciv@aol> wrote:
>>Quadibloc wrote:
>>> On May 27, 1:17 am, Ahem A Rivet's Shot <ste...@eircom.net> wrote:
>
>>>>I can hear it already - 64 cores should be enough for anybody.
>
>>> One core that was 64 times as fast would be better.
>
>>> While one can indeed laugh at statements made in the past... 64 K
>>> of RAM should be enough for anybody... when it comes to
>>> computing, a machine able to run Windows 3.1 probably is enough
>>> for what most desktops are used for, except multimedia.
>
>>> More processor power will always be useful, even if, in practice,
>>> for many applications, it is difficult to utilize more than 2 or
>>> 3 cored efficiently. There will always be some that will gain
>>> from 64 cores or 512 cores if they can have them..
>
>>Man! I'm going have so many communication problems. I still say
>>core when I talk about physical memory. Why in the whole fucking
>>world did was the term "core" transformed to mean CPU?
>
> Lack of imagination. :-)
>
> Each "core" is a processor core; i.e. semi-autonomous CPU, though
> often sharing one or two levels of cache with other cores on the
> same die (chip).
About 10 years ago I had occasion to ask a young woman who was a new
technical documentation writer what she thought "core" was, in the
context of the discussion she was hearing from some older geeks.
She understood that we were referring to memory, and figured we must
be talking about RAM -- but she didn't know about ferrite cores.
In her mind it was obviously the memory on the motherboard tightly
coupled to the CPU, as opposed to the hard drive. Since then,
the word "core" has come to refer to the CPUs, and so "core memory"
makes sense once again.
You should have seen her expression when I explained,
"Oh, no. You see back in the day, memory was made of little tiny
magnetic donuts on a big matrix with these tiny wires wrapped
through them!" And she really thought I was piling it on when
I said, "And they were made by oriental women sitting in factory
rows with magnifying glasses and nimble little fingers."
Then of course I showed her some photos of a core production line,
and a tiny jar full of cores that I had...
Roland Hutchinson
> Originally, the term
> "core" was used not by programmers, for whom core was indeed a common
> synonym for RAM long into the PC era, but by *chipmakers*. They're the
> ones who first spoke in terms of "putting two CPU cores on a die". The
> little square things that they cut a wafer into and then put inside a
> package with pins are called "dice" when it's necessary to use
> something unambiguous (chip could mean the die, or the package
> containing a die).
That use of "die" sounds a little bit funny to this non-chipmaker (though I
have become used to it).
I think of fabs churning out a stream of chips in much the same way that a
mint produces coins. Of course in a mint, the die is not the product
stamped out, but the business end of the machine that does the stamping.
--
Roland Hutchinson
He calls himself "the Garden State's leading violist da gamba,"
... comparable to being ruler of an exceptionally small duchy.
--Newark (NJ) Star Ledger ( http://tinyurl.com/RolandIsNJ )
Anne & Lynn Wheeler
Roland Hutchinson <my.sp...@verizon.net> writes:
> That use of "die" sounds a little bit funny to this non-chipmaker (though I
> have become used to it).
>
> I think of fabs churning out a stream of chips in much the same way that a
> mint produces coins. Of course in a mint, the die is not the product
> stamped out, but the business end of the machine that does the stamping.
fabs turn out wafers ... in this decade there has been moved from
8in/200mm to 12in/300mm wafers. basically manufacturing cost is per
wafer ... move to larger wafers allows cost to be amortized across a
larger number of chips.
wafers are tested and then sliced & diced into individual chips.
in parallel with move to larger wafers ... there has been process
changes that reduce circuit size. given same relatively same number of
circuits per chip ... the number of chips per wafer can significantly
increase (with combination of reduction in physical circuit size and the
larger wafer size).
part of the issue with more processors/cores per chip ... is holding
physical size of chip relatively constant (while the number of circuits
per physical area significantly increases).
in the late 90s ... we faced a problem with drastically reducing the
number of circuits (and physical size) per chip ... because the physical
"sawing" of chips was starting to take up larger physical part of the
wafer (than the chip physical area) ... which placed a limit on the
ability of increasing number of chips per wafer.
turns out that "EPC" RFID chips were running into same limit (i.e.
small chips that were being targeted to replace UPC barcodes on products
... being able to inventory electronicly) ... and there was effort to
develop new technologies for wafer slice&dice that drastically reduced
the size of the cut. Instead of large thousands of chips per wafer,
possibly large tens of thousands of chips per wafer.
we had started with .65 circuit size ... and initial move was to .2
circuit size ... which results in slightly better than factor of ten
times reduction in chip size (for same number of circuits) ... along
with increase in wafer size from 8in to 12in ... a little bit better
than twice the wafer area. Except for the cut size for cutting a lot
larger of number of chips ... it would have represented increase of 30
times in the number of chips per wafer (given same circuits per chip)
... or a theorical reduction in manufacturing cost per chip of 30 times.
We had previously started out by looking at the actual functions
required for the chip ... and had aggresively reduced the number of
circuits required by nearly a factor of ten ... which theoritically
would have represented a (combined) increase of 300 times in number of
chips per wafer.
--
40+yrs virtualization experience (since Jan68), online at home since Mar1970
Anne & Lynn Wheeler
re:
http://www.garlic.com/~lynn/2009h.html#53 64 Cores -- IBM is showing a prototype already
in the mid-90s, I would semi-facetiously comment that I would take a
$500 milspec part, aggresive cost reduce it by 2-3 orders of magnitude
while improving the integrity.
part of the process was slightly tweaking how business work so that
a lot of post-fab chip processing could be liminated (another factor of
four times/chip).
altogether it was something over 1000 times ... but heavily dependent on
new wafer cutting technologies (which was outcome of EPC RFID market
segment).
related AADS chip strawman
http://www.garlic.com/~lynn/x959.html#aads
with some of patents related to process
http://www.garlic.com/~lynn/aadssummary.htm
some of the patent effort discussed in this post:
http://www.garlic.com/~lynn/2009h.html#8 Supercomputers and electronic commerce
Scott Lurndal
>On May 27, 1:17=A0am, Ahem A Rivet's Shot <ste...@eircom.net> wrote:
>
>> =A0 =A0 =A0 =A0 I can hear it already - 64 cores should be enough for any=
>body.
>
>One core that was 64 times as fast would be better.
>
>While one can indeed laugh at statements made in the past... 64 K of
>RAM should be enough for anybody... when it comes to computing, a
>machine able to run Windows 3.1 probably is enough for what most
>desktops are used for, except multimedia.
>
>More processor power will always be useful, even if, in practice, for
>many applications, it is difficult to utilize more than 2 or 3 cored
>efficiently. There will always be some that will gain from 64 cores or
>512 cores if they can have them..
>
>John Savard
Here's my current benchmarking system:
[root@localhost ~]# cat /proc/cpuinfo | grep ^processor | wc -l
128
[root@localhost ~]# cat /proc/meminfo | grep MemTotal
MemTotal: 512092756 kB
(128 2.8Ghz Shanghai Opteron cores - the memory is only half populated).
scott
Rostyslaw J. Lewyckyj
The above was/is certainly confusing enough! :)
Anne & Lynn Wheeler
jmfbahciv <jmfbahciv@aol> writes:
> Man! I'm going have so many communication problems. I still say
> core when I talk about physical memory. Why in the whole fucking
> world did was the term "core" transformed to mean CPU?
one of the things that is people intensive is designing a new
processor. as circuits shrank ... it was possible to get multiple copies
of all the circuits of a single processor chip ... combined together on
single chip ... cut the circuit size in half & quadruple number of
circuits in an area.
with a little fiddling ... take the single-chip, single-core processor
circuit design (w/o some of the support circuit infrastructure) and
replicate it four times ... laying it out in a single chip using circuit
size that has been cut in half (w/o having to do a whole new processor
circuit design).
a enlarged image of the physical circuit layout ... it is even
frequently possible to recognize the replicated cores.
multi-core wiki page
http://en.wikipedia.org/wiki/Multi-core_%28computing%29
the two other issues was that increasing the execution frequency that
had somewhat been following decrease in circuit size ... ran into some
number of problems ... one was resistance, power consumption, & heat as
frequency went up; another was the physical distance that signals had to
travel (in relation to frequence); another was dealing with with
serialization constraints of single processor design with four times (or
8, 16, 64, etc) as many circuits.
in theory ... two processor operation (in single chip) potentially
provides twice as many instructions executed w/o increasing frequency
(and significant power & heat), lot less complexity to do a core
processor design with much smaller number of circuits ... and then
replicate that design multiple times on single chip (attempting to
obtain some benefit from being able to throw significantly larger number
of much smaller circuits into the same physical area). The big increase
in number of circuits in a chip (holding the physical size of chip
relative constaint) is much larger L1 and even L2 (on-chip) caches.
Andrew Swallow
[snip]
Devise a method of mapping block-addressable solid-state memory onto
the data modelling language UML that works both ways. Programmers
will just drift into using the new types of computer.
<http://en.wikipedia.org/wiki/Unified_Modeling_Language>
As for the memory just make sure that it does not make handling the
traditional data formats difficult. These include numeric variables,
string/character variables, arrays, linked lists, serial files, data
base records, data bases, printer spool files, key board input, mouse
input, screen output and bit serial model data streams. With luck the
handling of all these new file formats used on the internet will be
simplified.
Files can have read access, write access, update access, execute
access, delete access, shared access (record locking), access lists
with different access levels, owners, users, grouping of users,
system access, back-up & restore facilities and access protection.
No need to forget something useful.
The hardware does not have to implement all the above since it can
share the work with compilers and the operating system.
Andrew Swallow
Anne & Lynn Wheeler
http://www.garlic.com/~lynn/2009h.html#57 64 Cores -- IBM is showing a prototype already
a more detailed discussion of switch to multi-core processor chips:
http://www.csa.com/discoveryguides/multicore/reviewf.php
one of the comparison is single core processor in 45nm technlogy at 1V,
two core processor at 4GHz requires 107W and single core processor at
7.8GHz requires 430W.
Quadibloc
> I think of fabs churning out a stream of chips in much the same way that a
> mint produces coins. Of course in a mint, the die is not the product
> stamped out, but the business end of the machine that does the stamping.
If you go there, then there is the counterpunch which shapes the punch
so that it may create the matrix, from which the type slug is cast,
that the letter might be printed.
Before, of course, the pantograph, let alone phototypesetting for
offset printing...
John Savard
Lawrence Statton
> one of the comparison is single core processor in 45nm technlogy at 1V,
> two core processor at 4GHz requires 107W and single core processor at
> 7.8GHz requires 430W.
Is this the harbinger of the new quality-metric, "MIP/joule"?
--L
Roland Hutchinson
ITYM MIPS/watt or, equivalently, MI/joule.
Google readily reveals that "MIPS per watt" is already in (you should pardon
the expression) current use.
Charles Richmond
"...and the baby has to come out my *WHAT*???!!!" ;-)
--
+----------------------------------------------------------------+
| Charles and Francis Richmond richmond at plano dot net |
+----------------------------------------------------------------+
Roland Hutchinson
While we're at it, let us -- especially those of us who are fond of paper
money or sheet music -- not forget the intaglio press.
Charles Richmond
>
> [snip...] [snip...] [snip...]
>
> About 10 years ago I had occasion to ask a young woman who was a new
> technical documentation writer what she thought "core" was, in the
> context of the discussion she was hearing from some older geeks.
> She understood that we were referring to memory, and figured we must
> be talking about RAM -- but she didn't know about ferrite cores.
> In her mind it was obviously the memory on the motherboard tightly
> coupled to the CPU, as opposed to the hard drive. Since then,
> the word "core" has come to refer to the CPUs, and so "core memory"
> makes sense once again.
>
> You should have seen her expression when I explained,
> "Oh, no. You see back in the day, memory was made of little tiny
> magnetic donuts on a big matrix with these tiny wires wrapped
> through them!" And she really thought I was piling it on when
> I said, "And they were made by oriental women sitting in factory
> rows with magnifying glasses and nimble little fingers."
> Then of course I showed her some photos of a core production line,
> and a tiny jar full of cores that I had...
<joke>
A mom sent her little 8-year-old to Sunday school. When the boy came
home, the mom asked what he learned.
Boy: "I learned the story of how Moses led the Jews from Egypt."
Mom: "What happened???"
Boy: "Well, Moses was leading the Jews from Egypt. He was being chased
by the Egyptian army. He came to the Red Sea and couldn't get across. So
he got on his walkie-talkie and called in the Corp. of Engineers to
build a pontoon bridge so the Jews could cross."
"After crossing the Red Sea, the Jews looked back and saw the Egyptian
army coming. So Moses got on the walkie-talkie again and called in the
Air Force to bomb out the pontoon bridge. And the Jews were safe."
Mom: "Are you *sure* that this is what your Sunday school teacher told
you???"
Boy: "Well, *not* exactly. But what the teacher told me... you'd
*never* believe!!!"
</joke>
Little Oriental women stringing together tiny rings of ferrous
material... Yeah, right!!! ;-)
Quadibloc
wrote:
> While we're at it, let us -- especially those of us who are fond of paper
> money or sheet music -- not forget the intaglio press.
Oh, dear, yes. And then there was a high-quality magazine about the
art of photography which my local public library carried that used
*gravure*, for that matter.
Alea jacta est.
John Savard
Quadibloc
> Google readily reveals that "MIPS per watt" is already in (you should pardon
> the expression) current use.
Given that transformers for impedance matching are in such common use,
I think the metric MIPS per ampere is an expression that would need to
be pardoned.
John Savard
Quadibloc
> > Is this the harbinger of the new quality-metric, "MIP/joule"?
>
> ITYM MIPS/watt or, equivalently, MI/joule.
>
> Google readily reveals that "MIPS per watt" is already in (you should pardon
> the expression) current use.
While one could indeed refer to millions of instructions per watt-
second, it makes perfect sense that people would think of the unit as
measuring computer performance per unit of power consumption, rather
than unravelling the common computer performance unit to reimagine the
unit as one of computing done per unit of energy consumed.
Engineers and physicists might see the compound nature of the unit as
an opportunity to apply some dimensional analysis, but mere computer
commentators might well just say "Huh?", and I'm not sure I would
blame them.
John Savard
ArarghMai...@not.at.arargh.com
Not Oriental but: http://vt100.net/docs/misc/core/ :-)
--
ArarghMail905 at [drop the 'http://www.' from ->] http://www.arargh.com
BCET Basic Compiler Page: http://www.arargh.com/basic/index.html
To reply by email, remove the extra stuff from the reply address.
Rhialto
> ones who first spoke in terms of "putting two CPU cores on a die". The
> little square things that they cut a wafer into and then put inside a
> package with pins are called "dice" when it's necessary to use
> something unambiguous (chip could mean the die, or the package
> containing a die).
My English dictionary (Collins Concise) doesn't mention "die" having an
irregular plural. Yet this isn't the first time that I have seen people
write "dice" instead of "dies". Is there a trend I'm missing or is it
just a common mistake? (The plural of "pie" isn't "pice" either)
-Olaf
--
___ Olaf 'Rhialto' Seibert -- You author it, and I'll reader it.
\X/ rhialto/at/xs4all.nl -- Cetero censeo "authored" delendum esse.
Peter Flass
> Quadibloc wrote:
>> ones who first spoke in terms of "putting two CPU cores on a die". The
>> little square things that they cut a wafer into and then put inside a
>> package with pins are called "dice" when it's necessary to use
>> something unambiguous (chip could mean the die, or the package
>> containing a die).
>
> My English dictionary (Collins Concise) doesn't mention "die" having an
> irregular plural. Yet this isn't the first time that I have seen people
> write "dice" instead of "dies". Is there a trend I'm missing or is it
> just a common mistake? (The plural of "pie" isn't "pice" either)
>
> -Olaf
The things you use to shoot craps are "dice", and the singular is "die."
I'm not sure if this is the same "die" as used in chip manufacturing,
but if it is you need a new dictionary.
Roland Hutchinson
Nor would I.
On the other hand, for non-time-critical computation, MI/joule might be a
useful measure of cost in some sorts of situations that are not too hard to
imagine.
Roland Hutchinson
It ought to be in the dictionary as the singular of "dice" regardless of
whether it is the same die as in chip manufacturing.
Morten Reistad
Roland Hutchinson <my.sp...@verizon.net> wrote:
>Quadibloc wrote:
>
>> While one could indeed refer to millions of instructions per watt-
>> second, it makes perfect sense that people would think of the unit as
>> measuring computer performance per unit of power consumption, rather
>> than unravelling the common computer performance unit to reimagine the
>> unit as one of computing done per unit of energy consumed.
>>
>> Engineers and physicists might see the compound nature of the unit as
>> an opportunity to apply some dimensional analysis, but mere computer
>> commentators might well just say "Huh?", and I'm not sure I would
>> blame them.
>
>Nor would I.
>
>On the other hand, for non-time-critical computation, MI/joule might be a
>useful measure of cost in some sorts of situations that are not too hard to
>imagine.
I would suggest that it is THE dominating benchmark figure for
more than half of processors sold.
Think "handheld". Mobile phone. Gameboys. Music players. Navigation
devices. Together there are more than half a billion units in sale
for such devices, while peecees struggle to reach a hundred million.
-- mrr
Quadibloc
> Not Oriental but:http://vt100.net/docs/misc/core/
> :-)
While the photos on that page showed what might be a U.S. plant
producing core memories, that production did later get offshored to
places like Indonesia, the Philippines, and Hong Kong because of its
extremely labor-intensive character.
John Savard
jmfbahciv
or crumbs, not core.
/BAH
jmfbahciv
> On May 28, 6:31 am, jmfbahciv <jmfbahciv@aol> wrote:
>> Quadibloc wrote:
>
>>> More processor power will always be useful, even if, in practice, for
>>> many applications, it is difficult to utilize more than 2 or 3 cored
>>> efficiently. There will always be some that will gain from 64 cores or
>>> 512 cores if they can have them..
>> core when I talk about physical memory. Why in the whole fucking
>> world did was the term "core" transformed to mean CPU?
>
> core as the equivalent of processor.
>
> Core memory depends on little doughnuts of ferrite that have their
> center removed... while the processor is the thing at the heart, the
> kernel, the nucleus... and hence the core... of a microprocessor chip.
>
> Now that a microprocessor chip also contains multiple processors in a
> parallel sysplex (now called SMP, symmetric multiprocessing)
> configuration, and their level 2 shared cache, whereas before a
> microprocessor was only a CPU, using the older term, "processor",
> might cause confusion over whether you're talking about the chip, or
> one of the CPUs inside the chip. That's why they felt they needed a
> different word when talking about the heart of the chip. The term
> "kernel" has a technical meaning when talking about operating systems.
>
> But there's more to it than core memory being out of date, and "heart"
> or "nucleus" being too emotive or something. Originally, the term
> "core" was used not by programmers, for whom core was indeed a common
> synonym for RAM long into the PC era, but by *chipmakers*.
right. And they knew better than to call it core.
>They're the
> ones who first spoke in terms of "putting two CPU cores on a die". The
> little square things that they cut a wafer into and then put inside a
> package with pins are called "dice" when it's necessary to use
> something unambiguous (chip could mean the die, or the package
> containing a die).
>
> Chipmakers didn't talk about 'core', they only talked about DRAM and
> SRAM as kinds of chips they made, so the term, when it originated
> among them, caused _them_ no confusion. The term then moved out from
> their quarters into the wider world, where it did have the problem you
> note.
>
But how did the term leak out? The chipmakers had to have doc'ed it
somehow. I still claim that they should have known better than to
say core.
/BAH
Quadibloc
> My English dictionary (Collins Concise) doesn't mention "die" having an
> irregular plural. Yet this isn't the first time that I have seen people
> write "dice" instead of "dies". Is there a trend I'm missing or is it
> just a common mistake? (The plural of "pie" isn't "pice" either)
That is odd.
The plural of a die used for stamping should indeed be "dies". As in
"The Mint commissioned the dies to be used for stamping out the next
year's issue of coinage", or "This tap and die set contains 23 taps
and 23 dies".
The plural of a die which has six faces and is used for gambling - or
of the little square chips named after that object - would be dice,
however.
(Had they called the photomask a die, as someone suggested would be
more appropriate, then its plural would be "dies", since it would be
named after the other sense of the word!)
John Savard
Quadibloc
> It ought to be in the dictionary as the singular of "dice" regardless of
> whether it is the same die as in chip manufacturing.
That's true; one rarely speaks of "throwing a die", so the word might
be listed under the plural which is much more common. Old-fashioned
dictionaries wouldn't do this, but some modern ones might be striving
to be more usable and informal.
John Savard
Charlie Gibbs
<6e63327f-c835-4478...@r3g2000vbp.googlegroups.com>,
jsa...@ecn.ab.ca (Quadibloc) writes:
> On May 29, 6:05�am, Roland Hutchinson <my.spamt...@verizon.net> wrote:
>
>> It ought to be in the dictionary as the singular of "dice" regardless
>> of whether it is the same die as in chip manufacturing.
>
> That's true; one rarely speaks of "throwing a die",
Not as such. The more common phrase is "the die is cast."
--
/~\ cgi...@kltpzyxm.invalid (Charlie Gibbs)
\ / I'm really at ac.dekanfrus if you read it the right way.
X Top-posted messages will probably be ignored. See RFC1855.
/ \ HTML will DEFINITELY be ignored. Join the ASCII ribbon campaign!
Charlie Gibbs
my.sp...@verizon.net (Roland Hutchinson) writes:
> Google readily reveals that "MIPS per watt" is already in (you should
> pardon the expression) current use.
Wrong unit. But it's a powerful statement nonetheless.
Charlie Gibbs
<de5fd4c9-2b0e-4062...@y34g2000prb.googlegroups.com>,
jsa...@ecn.ab.ca (Quadibloc) writes:
> Core memory depends on little doughnuts of ferrite that have their
> center removed...
Redundancy alert! Doughnuts already have their center removed. :-)
Stan Barr
>
> Think "handheld". Mobile phone. Gameboys. Music players. Navigation
> devices. Together there are more than half a billion units in sale
> for such devices, while peecees struggle to reach a hundred million.
3 billion ARM cores alone in one year (2008).
(Some nice looking compact ARM single-boards around, would make a cool
little, quiet, desktop machine - they usually run Linux.)
--
Cheers,
Stan Barr plan.b .at. dsl .dot. pipex .dot. com
The future was never like this!
Mensanator
But that can be misleading also. Some x86 chip may have a seperate
math coprocessor chip and a seperate memory management chip. That's 3
chips, but still only a single processing unit. The use of "core"
seperates the concept of processing units from the physical reality
of chips. And would you care if your machine of 128 processing units
is a single chip or two chips, each a 64-core?
>
> /BAH
Roland Hutchinson
> In article <gvohi8$86i$1...@news.eternal-september.org>,
> Roland Hutchinson <my.sp...@verizon.net> wrote:
>>Quadibloc wrote:
>>
>>> While one could indeed refer to millions of instructions per watt-
>>> second, it makes perfect sense that people would think of the unit as
>>> measuring computer performance per unit of power consumption, rather
>>> than unravelling the common computer performance unit to reimagine the
>>> unit as one of computing done per unit of energy consumed.
>>>
>>> Engineers and physicists might see the compound nature of the unit as
>>> an opportunity to apply some dimensional analysis, but mere computer
>>> commentators might well just say "Huh?", and I'm not sure I would
>>> blame them.
>>
>>Nor would I.
>>
>>On the other hand, for non-time-critical computation, MI/joule might be a
>>useful measure of cost in some sorts of situations that are not too hard
>>to imagine.
>
> I would suggest that it is THE dominating benchmark figure for
> more than half of processors sold.
Well, to a good approximation at least. (I figured we'd be hearing from the
embedded jocks on this one.)
> Think "handheld". Mobile phone. Gameboys. Music players. Navigation
> devices. Together there are more than half a billion units in sale
> for such devices, while peecees struggle to reach a hundred million.
MI/joule is only meaningful to the extent that a MI does a comparable amount
of computational work on the processors being compared. If the processors
have different architectures, of course, an additional fudge factor needs to
be, erm, factored in.
Roland Hutchinson
> In article
> <de5fd4c9-2b0e-4062...@y34g2000prb.googlegroups.com>,
> jsa...@ecn.ab.ca (Quadibloc) writes:
>
>> Core memory depends on little doughnuts of ferrite that have their
>> center removed...
>
> Redundancy alert! Doughnuts already have their center removed. :-)
The ones that are sold abroad as "American doughnuts" do.
Others don't: your basic Dutch oliebolle, f'rex -- to say nothing of the
many types of doughnut popular in the USA that have their center inserted
(jelly doughnuts, Boston creme, etc.).
Perhaps to be clearer we should refer to ferrite bagles.
Mensanator
> Charlie Gibbs wrote:
> > In article
> > jsav...@ecn.ab.ca (Quadibloc) writes:
>
> >> Core memory depends on little doughnuts of ferrite that have their
> >> center removed...
>
> > Redundancy alert! Doughnuts already have their center removed. :-)
>
> The ones that are sold abroad as "American doughnuts" do.
>
> Others don't: your basic Dutch oliebolle, f'rex -- to say nothing of the
> many types of doughnut popular in the USA that have their center inserted
> (jelly doughnuts, Boston creme, etc.).
>
> Perhaps to be clearer we should refer to ferrite bagles.
You get my vote, as bagels usually have a consistency not
unlike ferrite.
Morten Reistad
Stan Barr <pla...@dsl.pipex.com> wrote:
>On Fri, 29 May 2009 14:40:47 +0200, Morten Reistad <fi...@last.name> wrote:
>>
>> Think "handheld". Mobile phone. Gameboys. Music players. Navigation
>> devices. Together there are more than half a billion units in sale
>> for such devices, while peecees struggle to reach a hundred million.
>
>3 billion ARM cores alone in one year (2008).
Yep. New statistics is in. For the first time humanity produced one
computer for every human on earth during 2008.
But the growth was "only" 15 percent or so. But MMUs are up
radically, from 620M to over 800M. MMU means "real computer", just
like the PDP10s needs the paging box to run tops20; the MMUs
means we have 800M computers with a real OS. We cannot see such a
boom in the sales figures for Windows, Symbian or BEos; so it is
bound to be BSDs and Linuxen in droves out there.
If this interpretation is correct, more than a third of all "real computers"
run BSD or Linux. (MS has a slight increase to 150M units, symbian is
up to 230, all the other commercial units are around 100M combined.
Still less than half a billion. And the 800M figure is a low estimate.
>(Some nice looking compact ARM single-boards around, would make a cool
>little, quiet, desktop machine - they usually run Linux.)
You are not the only one. ARM cores are slow, but you can easily
have 16 or more on a small die.
And the number of these ARM cores produced is just staggering.
-- mrr
Quadibloc
> And would you care if your machine of 128 processing units
> is a single chip or two chips, each a 64-core?
Yes, unless Microsoft changes its licensing policy.
John Savard
Eugene Miya
Walter Bushell <pr...@panix.com> wrote:
>Eight eight cores processors on a high end server. But, this should give
>some incentive to parallel processing.
Depends on the software.
"Otherwise, it's just a door stop."
><http://news.cnet.com/8301-13924_3-10249366-64.html?tag=newsEditorsPicksArea.0>
I thought someone had 80.
These are all behind other projects in the 90s.
--
Looking for an H-912 (container).
Quadibloc
> In article <slrnh2015r.2c8.pla...@rotwang.metropolis.local>,
> Stan Barr <pla...@dsl.pipex.com> wrote:
This is true. But until major advances are made in parallel computing,
most programs will run at a speed not much above what *one* of those
cores would give them.
This is why it's hard for me to get excited about the smaller CPUs,
even though they're plenty big enough for a lot of real work.
John Savard
Eugene Miya
Roland Hutchinson <my.sp...@verizon.net> wrote:
>Charlie Gibbs wrote:
>> In article
>> <de5fd4c9-2b0e-4062...@y34g2000prb.googlegroups.com>,
>> jsa...@ecn.ab.ca (Quadibloc) writes:
>>> Core memory depends on little doughnuts of ferrite that have their
>>> center removed...
>> Redundancy alert! Doughnuts already have their center removed. :-)
>
>The ones that are sold abroad as "American doughnuts" do.
>
>Others don't: your basic Dutch oliebolle, f'rex -- to say nothing of the
>many types of doughnut popular in the USA that have their center inserted
>(jelly doughnuts, Boston creme, etc.).
Psycho Donuts' Massive Head Trama is a filled jelly lacking a center.
I must still check out Voodoo Donuts for centerless donuts.
>Perhaps to be clearer we should refer to ferrite bagles.
Do you mean ferrite beagles?
Eugene Miya
Quadibloc <jsa...@ecn.ab.ca> wrote:
>On May 27, 1:17=A0am, Ahem A Rivet's Shot <ste...@eircom.net> wrote:
>> =A0 =A0 =A0 =A0 I can hear it already - 64 cores should be enough for any=
>body.
>
>One core that was 64 times as fast would be better.
Why settle for 64?
Nick Metropolis wanted 10,000 times as fast (on a VAX no less).
>While one can indeed laugh at statements made in the past... 64 K of
>RAM should be enough for anybody... when it comes to computing, a
>machine able to run Windows 3.1 probably is enough for what most
>desktops are used for, except multimedia.
>
>More processor power will always be useful, even if, in practice, for
>many applications, it is difficult to utilize more than 2 or 3 cored
>efficiently. There will always be some that will gain from 64 cores or
>512 cores if they can have them..
Well efficiency in the von Neumann bottleneck sense is best done
minimizing memory/processor ratios but this falls apart with more
complex calculations. Fewer people would know how to do say Nick's
Monte Carlo in various sequential languages on symmetric machines.
Eugene Miya
>Man! I'm going have so many communication problems. I still say
>core when I talk about physical memory. Why in the whole fucking
>world did was the term "core" transformed to mean CPU?
%A Ad Emmen
%A Jaap Hollenberg
%T Supercomputer is just an advertisement word, an interview with
^^^^^^^^^^^^^^^^^^^^^
Enrico Clementi
^^^^^^^^^^^^^^^ <- just ask.
%J Supercomputer
%C Amsterdam
%D July-September 1986
%P 24-33
%K marketing, policy/politics, IBM Kingston Engineering & Science Center,
%X Never very technical, but interesting reading.
Eugene Miya
Anne & Lynn Wheeler <ly...@garlic.com> wrote:
>Roland Hutchinson <my.sp...@verizon.net> writes:
>> That use of "die" sounds a little bit funny to this non-chipmaker (though I
>> have become used to it).
>>
>> I think of fabs churning out a stream of chips in much the same way that a
>> mint produces coins. Of course in a mint, the die is not the product
>> stamped out, but the business end of the machine that does the stamping.
>
>fabs turn out wafers ... in this decade there has been moved from
>8in/200mm to 12in/300mm wafers. basically manufacturing cost is per
>wafer ... move to larger wafers allows cost to be amortized across a
>larger number of chips.
Over simple.
The amortization has become a nail baiting issue for firms like Applied
Materials, Lam, etc. who provide equipment for fab lines. They want to
stay in business. Intel/AMD/IBM/others want to recoup purchase cost.
The problem also holds back warm superconductor applications.
>wafers are tested and then sliced & diced into individual chips.
>
>in parallel with move to larger wafers ... there has been process
>changes that reduce circuit size. given same relatively same number of
feature size
>circuits per chip ... the number of chips per wafer can significantly
>increase (with combination of reduction in physical circuit size and the
>larger wafer size).
>
>part of the issue with more processors/cores per chip ... is holding
>physical size of chip relatively constant (while the number of circuits
>per physical area significantly increases).
Depends on 8 inch development or merger into 12 inch (which WILL be less
trivial).
>in the late 90s ... we faced a problem with drastically reducing the
>number of circuits (and physical size) per chip ... because the physical
>"sawing" of chips was starting to take up larger physical part of the
>wafer (than the chip physical area) ... which placed a limit on the
>ability of increasing number of chips per wafer.
Buy finer slice and dice hardware.
Adapt or let your competitors do it.
>turns out that "EPC" RFID chips were running into same limit (i.e.
>small chips that were being targeted to replace UPC barcodes on products
>... being able to inventory electronicly) ... and there was effort to
>develop new technologies for wafer slice&dice that drastically reduced
>the size of the cut. Instead of large thousands of chips per wafer,
>possibly large tens of thousands of chips per wafer.
>
>we had started with .65 circuit size ... and initial move was to .2
>circuit size ... which results in slightly better than factor of ten
>times reduction in chip size (for same number of circuits) ... along
>with increase in wafer size from 8in to 12in ... a little bit better
>than twice the wafer area. Except for the cut size for cutting a lot
>larger of number of chips ... it would have represented increase of 30
>times in the number of chips per wafer (given same circuits per chip)
>... or a theorical reduction in manufacturing cost per chip of 30 times.
>
>We had previously started out by looking at the actual functions
>required for the chip ... and had aggresively reduced the number of
>circuits required by nearly a factor of ten ... which theoritically
>would have represented a (combined) increase of 300 times in number of
>chips per wafer.
Roland Hutchinson
That dog won't hunt.
Patrick Scheible
> Yep. New statistics is in. For the first time humanity produced one
> computer for every human on earth during 2008.
>
> But the growth was "only" 15 percent or so. But MMUs are up
> radically, from 620M to over 800M. MMU means "real computer", just
> like the PDP10s needs the paging box to run tops20; the MMUs
> means we have 800M computers with a real OS. We cannot see such a
> boom in the sales figures for Windows, Symbian or BEos; so it is
> bound to be BSDs and Linuxen in droves out there.
>
> If this interpretation is correct, more than a third of all "real computers"
> run BSD or Linux.
This is the best news I've heard in a while! Thanks.
-- Patrick
Walter Bushell
<fbf6c727-14c6-4144...@q2g2000vbr.googlegroups.com>,
Quadibloc <jsa...@ecn.ab.ca> wrote:
> While one could indeed refer to millions of instructions per watt-
> second, it makes perfect sense that people would think of the unit as
> measuring computer performance per unit of power consumption, rather
> than unravelling the common computer performance unit to reimagine the
> unit as one of computing done per unit of energy consumed.
These are equivalent measures, of course.
Walter Bushell
Roland Hutchinson <my.sp...@verizon.net> wrote:
> Nor would I.
>
> On the other hand, for non-time-critical computation, MI/joule might be a
> useful measure of cost in some sorts of situations that are not too hard to
> imagine.
Heat dissipation is also measured here. I suppose you included that in
your cost dimension. Apple with the dual G5 line got to the point it
required a special power circuit of like 20 Amps to itself, making it a
rather expensive computer to run.
Walter Bushell
Morten Reistad <fi...@last.name> wrote:
> In article <gvohi8$86i$1...@news.eternal-september.org>,
> Roland Hutchinson <my.sp...@verizon.net> wrote:
> >Quadibloc wrote:
> >
> >> While one could indeed refer to millions of instructions per watt-
> >> second, it makes perfect sense that people would think of the unit as
> >> measuring computer performance per unit of power consumption, rather
> >> than unravelling the common computer performance unit to reimagine the
> >> unit as one of computing done per unit of energy consumed.
> >>
> >> Engineers and physicists might see the compound nature of the unit as
> >> an opportunity to apply some dimensional analysis, but mere computer
> >> commentators might well just say "Huh?", and I'm not sure I would
> >> blame them.
> >
> >Nor would I.
> >
> >On the other hand, for non-time-critical computation, MI/joule might be a
> >useful measure of cost in some sorts of situations that are not too hard to
> >imagine.
>
> I would suggest that it is THE dominating benchmark figure for
> more than half of processors sold.
>
> Think "handheld". Mobile phone. Gameboys. Music players. Navigation
> devices. Together there are more than half a billion units in sale
> for such devices, while peecees struggle to reach a hundred million.
>
> -- mrr
And large server farms.
Walter Bushell
<53cc43c7-d61c-4b41...@s20g2000vbp.googlegroups.com>,
Quadibloc <jsa...@ecn.ab.ca> wrote:
You can always run more that one program at a time, and if you have
enough cores, it works for that.
Walter Bushell
<08bdd0fa-7893-4732...@o30g2000vbc.googlegroups.com>,
Mensanator <mensa...@aol.com> wrote:
> On May 29, 2:27�pm, Roland Hutchinson <my.spamt...@verizon.net> wrote:
> > Charlie Gibbs wrote:
> > > In article
> > > <de5fd4c9-2b0e-4062-96eb-7ea7efab6...@y34g2000prb.googlegroups.com>,
> > > jsav...@ecn.ab.ca (Quadibloc) writes:
> >
> > >> Core memory depends on little doughnuts of ferrite that have their
> > >> center removed...
> >
> > > Redundancy alert! �Doughnuts already have their center removed. �:-)
> >
> > The ones that are sold abroad as "American doughnuts" do.
> >
> > Others don't: your basic Dutch oliebolle, f'rex -- to say nothing of the
> > many types of doughnut popular in the USA that have their center inserted
> > (jelly doughnuts, Boston creme, etc.).
> >
> > Perhaps to be clearer we should refer to ferrite bagles.
>
> You get my vote, as bagels usually have a consistency not
> unlike ferrite.
>
And I've seen a product referred to as "a brown rice bagel". But
seriously a bagel is made with white flour as white flour was expensive
and a special treat.
Bernd Felsche
>In article
><de5fd4c9-2b0e-4062...@y34g2000prb.googlegroups.com>,
>jsa...@ecn.ab.ca (Quadibloc) writes:
>> Core memory depends on little doughnuts of ferrite that have their
>> center removed...
>Redundancy alert! Doughnuts already have their center removed. :-)
Not the Berliner. Their centre is filled with jam.
--
/"\ Bernd Felsche - Innovative Reckoning, Perth, Western Australia
\ / ASCII ribbon campaign | When we remember that we are all mad,
X against HTML mail | the mysteries disappear and life stands
/ \ and postings | explained. -- Mark Twain
Bernd Felsche
>On Fri, 29 May 2009 14:40:47 +0200, Morten Reistad <fi...@last.name> wrote:
>>
>> Think "handheld". Mobile phone. Gameboys. Music players. Navigation
>> devices. Together there are more than half a billion units in sale
>> for such devices, while peecees struggle to reach a hundred million.
>3 billion ARM cores alone in one year (2008).
>(Some nice looking compact ARM single-boards around, would make a cool
>little, quiet, desktop machine - they usually run Linux.)
Alas, I'd venture to say that more than half of those are in land-fill.
Disposable computers. Too cheap to repair.
Bernd Felsche
One should. Those systems are usually NUMA and each chip can
physically access directly only the memory attached to that chip.
An external "hyper-transport" or data channel of identical
functionality, connects between multiple processor chips to transfer
data from "another chip's memory" to the local memory. Cache
coherence is required between the memory pages/lines on all the
registered chips, slowing access until the non-local chips all
release that physical memory reference.
Walter Bushell
Bernd Felsche <ber...@innovative.iinet.net.au> wrote:
> "Charlie Gibbs" <cgi...@kltpzyxm.invalid> wrote:
>
> >In article
> ><de5fd4c9-2b0e-4062...@y34g2000prb.googlegroups.com>,
> >jsa...@ecn.ab.ca (Quadibloc) writes:
>
> >> Core memory depends on little doughnuts of ferrite that have their
> >> center removed...
>
> >Redundancy alert! Doughnuts already have their center removed. :-)
>
> Not the Berliner. Their centre is filled with jam.
The late President Kennedy was a jelly doughnut?
Joe Pfeiffer
> "Charlie Gibbs" <cgi...@kltpzyxm.invalid> wrote:
>
>>In article
>><de5fd4c9-2b0e-4062...@y34g2000prb.googlegroups.com>,
>>jsa...@ecn.ab.ca (Quadibloc) writes:
>
>>> Core memory depends on little doughnuts of ferrite that have their
>>> center removed...
>
>>Redundancy alert! Doughnuts already have their center removed. :-)
>
> Not the Berliner. Their centre is filled with jam.
The default donut has the center removed. Others are special cases.
Gene Wirchenko
>On May 29, 6:05�am, Roland Hutchinson <my.spamt...@verizon.net> wrote:
>
>> It ought to be in the dictionary as the singular of "dice" regardless of
>> whether it is the same die as in chip manufacturing.
>
>That's true; one rarely speaks of "throwing a die", so the word might
>be listed under the plural which is much more common. Old-fashioned
>dictionaries wouldn't do this, but some modern ones might be striving
>to be more usable and informal.
How often do you throw a die? Usually, it is a pair of them. In
RPGs, "die roll" is used quite heavily.
Sincerely,
Gene Wirchenko
Computerese Irregular Verb Conjugation:
I have preferences.
You have biases.
He/She has prejudices.
Gene Wirchenko
[snip]
>About 10 years ago I had occasion to ask a young woman who was a new
>technical documentation writer what she thought "core" was, in the
>context of the discussion she was hearing from some older geeks.
>She understood that we were referring to memory, and figured we must
>be talking about RAM -- but she didn't know about ferrite cores.
>In her mind it was obviously the memory on the motherboard tightly
>coupled to the CPU, as opposed to the hard drive. Since then,
>the word "core" has come to refer to the CPUs, and so "core memory"
>makes sense once again.
>
>You should have seen her expression when I explained,
>"Oh, no. You see back in the day, memory was made of little tiny
>magnetic donuts on a big matrix with these tiny wires wrapped
>through them!" And she really thought I was piling it on when
>I said, "And they were made by oriental women sitting in factory
>rows with magnifying glasses and nimble little fingers."
>Then of course I showed her some photos of a core production line,
>and a tiny jar full of cores that I had...
The best B.S. stories are the ones that are true.
Charles Richmond
> On May 29, 3:21 am, ArarghMail905NOS...@NOT.AT.Arargh.com wrote:
>
>> Not Oriental but:http://vt100.net/docs/misc/core/
>
>> :-)
>
> While the photos on that page showed what might be a U.S. plant
> producing core memories, that production did later get offshored to
> places like Indonesia, the Philippines, and Hong Kong because of its
> extremely labor-intensive character.
>
*and* ... some of those female folk were *smaller* than your average
Western person and *did* have smaller fingers.
--
+----------------------------------------------------------------+
| Charles and Francis Richmond richmond at plano dot net |
+----------------------------------------------------------------+
Charles Richmond
> In article
> jsa...@ecn.ab.ca (Quadibloc) writes:
>
>> On May 29, 6:05 am, Roland Hutchinson <my.spamt...@verizon.net> wrote:
>>
>>> It ought to be in the dictionary as the singular of "dice" regardless
>>> of whether it is the same die as in chip manufacturing.
>> That's true; one rarely speaks of "throwing a die",
>
>
Yeah, yeah... and:
"The slow one now
Will later be fast
As the present now
Will later be past
The order is
Rapidly fadin'.
And the first one now
Will later be last
For the times they are a-changin'."
Charles Richmond
> On May 29, 4:47 am, Rhialto <rhia...@falu.nl> wrote:
>
>> My English dictionary (Collins Concise) doesn't mention "die" having an
>> irregular plural. Yet this isn't the first time that I have seen people
>> write "dice" instead of "dies". Is there a trend I'm missing or is it
>> just a common mistake? (The plural of "pie" isn't "pice" either)
>
> That is odd.
>
> The plural of a die used for stamping should indeed be "dies". As in
> "The Mint commissioned the dies to be used for stamping out the next
> year's issue of coinage", or "This tap and die set contains 23 taps
> and 23 dies".
>
> The plural of a die which has six faces and is used for gambling - or
> of the little square chips named after that object - would be dice,
> however.
>
> (Had they called the photomask a die, as someone suggested would be
> more appropriate, then its plural would be "dies", since it would be
> named after the other sense of the word!)
>
> John Savard
Well, if the plural of "mouse" is "mice", then the *singular* of "dice"
should be "douse". ;-)
(The plural of "die" is really "mega-death". ;-)
Mensanator
> In article <oog6f6x19p....@innovative.iinet.net.au>,
> �Bernd Felsche <ber...@innovative.iinet.net.au> wrote:
>
> > "Charlie Gibbs" <cgi...@kltpzyxm.invalid> wrote:
>
> > >In article
> > >jsav...@ecn.ab.ca (Quadibloc) writes:
>
> > >> Core memory depends on little doughnuts of ferrite that have their
> > >> center removed...
>
> > >Redundancy alert! �Doughnuts already have their center removed. �:-)
>
> > Not the Berliner. Their centre is filled with jam.
>
> The late President Kennedy was a jelly doughnut?
That's what he said. The Germans thought it was
funny.
Charles Richmond
> In article
> jsa...@ecn.ab.ca (Quadibloc) writes:
>
>> Core memory depends on little doughnuts of ferrite that have their
>> center removed...
>
> Redundancy alert! Doughnuts already have their center removed. :-)
>
But sometimes you just have to repeat it again one more time... ;-)
Charles Richmond
And Mi$uck OS's that work at all are special cases... ;-)
Roland Hutchinson
Not in all cultures!
Roland Hutchinson
> Joe Pfeiffer wrote:
>> Bernd Felsche <ber...@innovative.iinet.net.au> writes:
>>
>>> "Charlie Gibbs" <cgi...@kltpzyxm.invalid> wrote:
>>>
>>>> In article
>>>> <de5fd4c9-2b0e-4062...@y34g2000prb.googlegroups.com>,
>>>> jsa...@ecn.ab.ca (Quadibloc) writes:
>>>>> Core memory depends on little doughnuts of ferrite that have their
>>>>> center removed...
>>>> Redundancy alert! Doughnuts already have their center removed. :-)
>>> Not the Berliner. Their centre is filled with jam.
>>
>> The default donut has the center removed. Others are special cases.
>
> And Mi$uck OS's that work at all are special cases... ;-)
Maybe if they took the jam out of the center...
Gene Wirchenko
[snip]
>MI/joule is only meaningful to the extent that a MI does a comparable amount
>of computational work on the processors being compared. If the processors
>have different architectures, of course, an additional fudge factor needs to
>be, erm, factored in.
Lie, statistics, and CPU benchmarks?
Gene Wirchenko
>In article <gvnlqp$nmr$1...@news.eternal-september.org>,
>my.sp...@verizon.net (Roland Hutchinson) writes:
>
>> Google readily reveals that "MIPS per watt" is already in (you should
>> pardon the expression) current use.
>
>Wrong unit. But it's a powerful statement nonetheless.
Watt you say?
Morten Reistad
Quadibloc <jsa...@ecn.ab.ca> wrote:
>On May 29, 2:57�pm, Morten Reistad <fi...@last.name> wrote:
>> In article <slrnh2015r.2c8.pla...@rotwang.metropolis.local>,
>> Stan Barr �<pla...@dsl.pipex.com> wrote:
>
>> >little, quiet, desktop machine - they usually run Linux.)
>>
>> have 16 or more on a small die.
>
>This is true. But until major advances are made in parallel computing,
>most programs will run at a speed not much above what *one* of those
>cores would give them.
Linux runs well on 64+ cores, but cache thrashing may become an issue.
Mysql, apache too. PHP and web scripts are usually pretty small, so
as long as linux, apache and mysql can handle parallell threads you
are home free. Firfox handles threads acceptably well for a 16-wide
set of cores. X and gnome/KDE needs work.
Systems such as asterisk, ser, dns, ntp, sendmail, media players,
transcoders and mixers already handle multiple threads very well.
>This is why it's hard for me to get excited about the smaller CPUs,
>even though they're plenty big enough for a lot of real work.
They can also become a totally new device; a mini-laptop less than
a cm high, no moving parts, rugged, battery for a week, and
fits it a purse. As the economist said, "forget that it is
running a different OS. This is a new type of device, with
new apps. As long as it interfaces well with the old documents,
all is fine."
-- mrr
Morten Reistad
Smaller than you may think, although they are tuned to performance.
Dedicated IT-based servers number less than 14M processors per year.
Microsoft license around 7M multiprocessor server licenses per year,
once you correct for duplicates. (there is a LOT of migration in
MS server licenses, which inflate sales numbers).
They have a market share of slightly above half.
Washing machines use 8M processors, and they can save a lot
of water and electricity by having intelligent processors.
The whole desktop/laptop market is well below 100M units/year, and
has leveled off. The only boost has come from the micro-laptops.
Microsoft has hit a brick wall in sales at around 140M units. This
counts everything; servers, mobile, desktop.
Symbian is thriving at ~230M. And growing. And the process control
BSD and Linux market is booming. The alarm industry alone can
absorb 30M units a year.
-- mrr
Morten Reistad
Dhrystones per watt, anyone? (or is the dhrystone possibly not
per second, so it is dhrystones/joule?)
The VIA Eden series should score high here. 600k ds per 8 watts.
-- mrr
Morten Reistad
Bernd Felsche <ber...@innovative.iinet.net.au> wrote:
>Stan Barr <pla...@dsl.pipex.com> wrote:
>>On Fri, 29 May 2009 14:40:47 +0200, Morten Reistad <fi...@last.name> wrote:
>>>
>>> Think "handheld". Mobile phone. Gameboys. Music players. Navigation
>>> devices. Together there are more than half a billion units in sale
>>> for such devices, while peecees struggle to reach a hundred million.
>
>>3 billion ARM cores alone in one year (2008).
>
>>(Some nice looking compact ARM single-boards around, would make a cool
>>little, quiet, desktop machine - they usually run Linux.)
>
>Alas, I'd venture to say that more than half of those are in land-fill.
No, they are in gadgets. Like disks, tapes, cd and dvd devices,
car engines, instrumentation of all types, etc etc. A modern PC
probably contains a two-digit number of ARM cores. A modern car
too.
It is all part of the gadgetisation of computing. The desktop
is not dead, but it is rapidly becoming irrelevant. All the
cool stuff happens elsewhere.
>Disposable computers. Too cheap to repair.
But we are unlikely to dispose of these for another decade or
more.
-- mrr
Peter Flass
>
> How often do you throw a die? Usually, it is a pair of them. In
> RPGs, "die roll" is used quite heavily.
>
Usually in pairs, but not always. I've seen games where the
instructions start out "roll one die to determine who moves first." Of
course this was before video games.
Peter Flass
>
> It is all part of the gadgetisation of computing. The desktop
> is not dead, but it is rapidly becoming irrelevant. All the
> cool stuff happens elsewhere.
Maybe for you, but most of us will never get to interact with these
gadgets as "computers." Who cares what CPU(s) your refrigerator is
running, except the guy who programs them?
jmfbahciv
> Christopher C. Stacy wrote:
>>
>> [snip...] [snip...] [snip...]
>>
>> About 10 years ago I had occasion to ask a young woman who was a new
>> technical documentation writer what she thought "core" was, in the
>> context of the discussion she was hearing from some older geeks.
>> She understood that we were referring to memory, and figured we must
>> be talking about RAM -- but she didn't know about ferrite cores.
>> In her mind it was obviously the memory on the motherboard tightly
>> coupled to the CPU, as opposed to the hard drive. Since then,
>> the word "core" has come to refer to the CPUs, and so "core memory"
>> makes sense once again.
>>
>> You should have seen her expression when I explained, "Oh, no. You see
>> back in the day, memory was made of little tiny
>> magnetic donuts on a big matrix with these tiny wires wrapped
>> through them!" And she really thought I was piling it on when
>> I said, "And they were made by oriental women sitting in factory
>> rows with magnifying glasses and nimble little fingers."
>> Then of course I showed her some photos of a core production line,
>> and a tiny jar full of cores that I had...
>
>
> A mom sent her little 8-year-old to Sunday school. When the boy came
> home, the mom asked what he learned.
>
> Boy: "I learned the story of how Moses led the Jews from Egypt."
>
> Mom: "What happened???"
>
> Boy: "Well, Moses was leading the Jews from Egypt. He was being chased
> by the Egyptian army. He came to the Red Sea and couldn't get across. So
> he got on his walkie-talkie and called in the Corp. of Engineers to
> build a pontoon bridge so the Jews could cross."
>
> "After crossing the Red Sea, the Jews looked back and saw the Egyptian
> army coming. So Moses got on the walkie-talkie again and called in the
> Air Force to bomb out the pontoon bridge. And the Jews were safe."
>
> Mom: "Are you *sure* that this is what your Sunday school teacher told
> you???"
>
> Boy: "Well, *not* exactly. But what the teacher told me... you'd
> *never* believe!!!"
>
> </joke>
>
>
> Little Oriental women stringing together tiny rings of ferrous
> material... Yeah, right!!! ;-)
>
Based on the quality of the crocheting I saw in China, those
women are amazing. The loops on the doilies that are placed on
all the tray tables on the trains were so tiny, I marvelled for
a long time. And these couldn't have been their best stuff.
/BAH
jmfbahciv
> In article
> jsa...@ecn.ab.ca (Quadibloc) writes:
>
>> On May 29, 6:05 am, Roland Hutchinson <my.spamt...@verizon.net> wrote:
>>
>>> It ought to be in the dictionary as the singular of "dice" regardless
>>> of whether it is the same die as in chip manufacturing.
>> That's true; one rarely speaks of "throwing a die",
>
> Not as such. The more common phrase is "the die is cast."
>
thought that it had to do with a metal die which had
already been formed (which I thought was the word for cast).
/BAH
jmfbahciv
> In article <gvnlqp$nmr$1...@news.eternal-september.org>,
> my.sp...@verizon.net (Roland Hutchinson) writes:
>
>> Google readily reveals that "MIPS per watt" is already in (you should
>> pardon the expression) current use.
>
> Wrong unit. But it's a powerful statement nonetheless.
>
C? Or c?
/BAH
jmfbahciv
> On May 29, 8:23 am, jmfbahciv <jmfbahciv@aol> wrote:
>> Anne & Lynn Wheeler wrote:
>>
>>
>>
>>> jmfbahciv <jmfbahciv@aol> writes:
>>>> core when I talk about physical memory. Why in the whole fucking
>>>> world did was the term "core" transformed to mean CPU?
>>> one of the things that is people intensive is designing a new
>>> processor. as circuits shrank ... it was possible to get multiple copies
>>> of all the circuits of a single processor chip ... combined together on
>>> single chip ... cut the circuit size in half & quadruple number of
>>> circuits in an area.
>>> with a little fiddling ... take the single-chip, single-core processor
>>> circuit design (w/o some of the support circuit infrastructure) and
>>> replicate it four times ... laying it out in a single chip using circuit
>>> size that has been cut in half (w/o having to do a whole new processor
>>> circuit design).
>>> a enlarged image of the physical circuit layout ... it is even
>>> frequently possible to recognize the replicated cores.
>>> multi-core wiki page
>>> http://en.wikipedia.org/wiki/Multi-core_%28computing%29
>>> the two other issues was that increasing the execution frequency that
>>> had somewhat been following decrease in circuit size ... ran into some
>>> number of problems ... one was resistance, power consumption, & heat as
>>> frequency went up; another was the physical distance that signals had to
>>> travel (in relation to frequence); another was dealing with with
>>> serialization constraints of single processor design with four times (or
>>> 8, 16, 64, etc) as many circuits.
>>> in theory ... two processor operation (in single chip) potentially
>>> provides twice as many instructions executed w/o increasing frequency
>>> (and significant power & heat), lot less complexity to do a core
>>> processor design with much smaller number of circuits ... and then
>>> replicate that design multiple times on single chip (attempting to
>>> obtain some benefit from being able to throw significantly larger number
>>> of much smaller circuits into the same physical area). The big increase
>>> in number of circuits in a chip (holding the physical size of chip
>>> relative constaint) is much larger L1 and even L2 (on-chip) caches.
>> If I were the Namer of Things, I'd have been calling the CPUs chips
>> or crumbs, not core.
>
> But that can be misleading also. Some x86 chip may have a seperate
> math coprocessor chip and a seperate memory management chip. That's 3
> chips, but still only a single processing unit.
Sigh! You mean central procession unit.
>The use of "core"
> seperates the concept of processing units from the physical reality
> of chips. And would you care if your machine of 128 processing units
> is a single chip or two chips, each a 64-core?
I would care depending on what each processor was supposed to do.
/BAH
jmfbahciv
> Morten Reistad wrote:
>
>> In article <gvohi8$86i$1...@news.eternal-september.org>,
>> Roland Hutchinson <my.sp...@verizon.net> wrote:
>>> Quadibloc wrote:
>>>
>>>> While one could indeed refer to millions of instructions per watt-
>>>> second, it makes perfect sense that people would think of the unit as
>>>> measuring computer performance per unit of power consumption, rather
>>>> than unravelling the common computer performance unit to reimagine the
>>>> unit as one of computing done per unit of energy consumed.
>>>>
>>>> an opportunity to apply some dimensional analysis, but mere computer
>>>> commentators might well just say "Huh?", and I'm not sure I would
>>>> blame them.
>>> Nor would I.
>>>
>>> On the other hand, for non-time-critical computation, MI/joule might be a
>>> useful measure of cost in some sorts of situations that are not too hard
>>> to imagine.
>> I would suggest that it is THE dominating benchmark figure for
>> more than half of processors sold.
>
> embedded jocks on this one.)
>
>> Think "handheld". Mobile phone. Gameboys. Music players. Navigation
>> devices. Together there are more than half a billion units in sale
>> for such devices, while peecees struggle to reach a hundred million.
>
> of computational work on the processors being compared. If the processors
> have different architectures, of course, an additional fudge factor needs to
> be, erm, factored in.
>
/BAH
jmfbahciv
> Eugene Miya wrote:
>
>> In article <gvpcue$4lb$2...@news.eternal-september.org>,
>>>> jsa...@ecn.ab.ca (Quadibloc) writes:
>>>>> Core memory depends on little doughnuts of ferrite that have their
>>>>> center removed...
>>>> Redundancy alert! Doughnuts already have their center removed. :-)
>>>
>>> Others don't: your basic Dutch oliebolle, f'rex -- to say nothing of the
>>> many types of doughnut popular in the USA that have their center inserted
>>> (jelly doughnuts, Boston creme, etc.).
>>
>> Psycho Donuts' Massive Head Trama is a filled jelly lacking a center.
>> I must still check out Voodoo Donuts for centerless donuts.
>>
>>
>>> Perhaps to be clearer we should refer to ferrite bagles.
>> Do you mean ferrite beagles?
>
> That dog won't hunt.
>
Unless it bytes.
/BAH
Mensanator
Ok, but what would you call it, Miss Namer of Things?
A two-chip, 128-crumb processor?
>
> /BAH
Walter Bushell
Morten Reistad <fi...@last.name> wrote:
And how is that on a dollar basis, many of these applications are not
compute intensive.
Walter Bushell
<94746f2d-7bf3-4348...@x3g2000yqa.googlegroups.com>,
Mensanator <mensa...@aol.com> wrote:
> On May 29, 9:39?pm, Walter Bushell <pr...@panix.com> wrote:
> > In article <oog6f6x19p....@innovative.iinet.net.au>,
> > ?Bernd Felsche <ber...@innovative.iinet.net.au> wrote:
> >
> > > "Charlie Gibbs" <cgi...@kltpzyxm.invalid> wrote:
> >
> > > >In article
> > > ><de5fd4c9-2b0e-4062-96eb-7ea7efab6...@y34g2000prb.googlegroups.com>,
> > > >jsav...@ecn.ab.ca (Quadibloc) writes:
> >
> > > >> Core memory depends on little doughnuts of ferrite that have their
> > > >> center removed...
> >
> > > >Redundancy alert! ?Doughnuts already have their center removed. ?:-)
> >
> > > Not the Berliner. Their centre is filled with jam.
> >
> > The late President Kennedy was a jelly doughnut?
>
> That's what he said. The Germans thought it was
> funny.
But they cheered it first.
Walter Bushell
Charles Richmond <fri...@tx.rr.com> wrote:
> Joe Pfeiffer wrote:
> > Bernd Felsche <ber...@innovative.iinet.net.au> writes:
> >
> >> "Charlie Gibbs" <cgi...@kltpzyxm.invalid> wrote:
> >>
> >>> In article
> >>> <de5fd4c9-2b0e-4062...@y34g2000prb.googlegroups.com>,
> >>> jsa...@ecn.ab.ca (Quadibloc) writes:
> >>>> Core memory depends on little doughnuts of ferrite that have their
> >>>> center removed...
> >>> Redundancy alert! Doughnuts already have their center removed. :-)
> >> Not the Berliner. Their centre is filled with jam.
> >
> > The default donut has the center removed. Others are special cases.
>
> And Mi$uck OS's that work at all are special cases... ;-)
The "jam" in the center of the Microsoft OSen is very gooey.
Scott Lurndal
>>> Anne & Lynn Wheeler wrote:
>>> > jmfbahciv <jmfbahciv@aol> writes:
>
>>> > in theory ... two processor operation (in single chip)
>>> > potentially provides twice as many instructions executed w/o
>>> > increasing frequency (and significant power & heat), lot less
>>> > complexity to do a core processor design with much smaller
>>> > number of circuits ... and then replicate that design multiple
>>> > times on single chip (attempting to obtain some benefit from
>>> > being able to throw significantly larger number of much smaller
>>> > circuits into the same physical area). The big increase in
>>> > number of circuits in a chip (holding the physical size of chip
>>> > relative constaint) is much larger L1 and even L2 (on-chip)
>>> > caches.
>
>>> If I were the Namer of Things, I'd have been calling the CPUs chips
>>> or crumbs, not core.
>
>>But that can be misleading also. Some x86 chip may have a seperate
>>math coprocessor chip and a seperate memory management chip. That's 3
>>seperates the concept of processing units from the physical reality
>>of chips. And would you care if your machine of 128 processing units
>>is a single chip or two chips, each a 64-core?
>
>physically access directly only the memory attached to that chip.
>An external "hyper-transport" or data channel of identical
>functionality, connects between multiple processor chips to transfer
>data from "another chip's memory" to the local memory. Cache
>coherence is required between the memory pages/lines on all the
>registered chips, slowing access until the non-local chips all
>release that physical memory reference.
Pretty much every x86 server going forward will be ccNUMA. AMD's
Hypertransport based servers have been ccNUMA for a few years now,
and with Quickpath Interconnect (QPI), Intel has also gone the
NUMA route for all server processors.
Unlike traditional frontside bus snoop schemes, direct connect busses use
a broadcast paradigm (with or without directories) to handle
coherency. The socket is considered a node, so with Istanbul, for
example, you'll have 6 processing cores per node with 12 on a two-socket
(two NUMA node) system.
Some of the added latency is covered by larger L3 caches (or remote caches
for larger directory-based ccNUMA systems).
scott
Stan Barr
>
> They can also become a totally new device; a mini-laptop less than
> a cm high, no moving parts, rugged, battery for a week, and
> fits it a purse. As the economist said, "forget that it is
> running a different OS. This is a new type of device, with
> new apps. As long as it interfaces well with the old documents,
> all is fine."
Sorta like a Nokia N810??
--
Cheers,
Stan Barr plan.b .at. dsl .dot. pipex .dot. com
The future was never like this!
Roland Hutchinson
> "Charlie Gibbs" <cgi...@kltpzyxm.invalid> wrote:
>
>>In article <gvnlqp$nmr$1...@news.eternal-september.org>,
>>my.sp...@verizon.net (Roland Hutchinson) writes:
>>
>>> Google readily reveals that "MIPS per watt" is already in (you should
>>> pardon the expression) current use.
>>
>>Wrong unit. But it's a powerful statement nonetheless.
>
> Watt you say?
How much is that in teraflops per horsepower?
Ahem A Rivet's Shot
Wow! I never thought of that interpretation but it makes good
sense, I'm pretty sure it's about the gaming variety though.
--
Steve O'Hara-Smith | Directable Mirror Arrays
C:>WIN | A better way to focus the sun
The computer obeys and wins. | licences available see
You lose and Bill collects. | http://www.sohara.org/
Ahem A Rivet's Shot
Si!
Morten Reistad
Exactly. 90% of computer-exposed people have a so firm association
between the "microsoft desktop" and "computer" that it is hard
to fit anything else. They are very sceptical to performance,
having learnt to live with the MS quirks and bugs, but see no
feasible alternative inside the computer domain.
So solutions step out of the "computer" domain and into
the "gadget" domain. Only a small step for a programmer, but
a giant leap for the marketeer.
See all the Ipods, Nokia 770s/8xx, gameboys, etc. We don't
interact with them as computers, that is the whole idea of
gadgetisation. You interact with the GPS as a Gps, but the
menues may be a graphics system form some well known
graphics system from an academic institution in the 1980s.
Microsoft has crashed and burned the idea of a computer
as a general respository of usuful stuff that works. So
we settle for the next best thing, gadgets that do specialised
things, and work.
We have only seen the enbryonic beginnings of this.
-- mrr
Quadibloc
> Charlie Gibbs wrote:
> > Not as such. The more common phrase is "the die is cast."
>
> That phrase was talking about gambling dice?!!!! I always
> thought that it had to do with a metal die which had
> already been formed (which I thought was the word for cast).
Yes. When Gaius Julius took his soldiers across the Rubicon, because
he was going to overthrow the Roman Republic, and make himself Caesar,
he knew the outcome was unpredictable when he said "Alea jacta est".
The Latin word for "jacta" is, of course, cognate to "eject", so it
means to throw.
John Savard