Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]
Groups > comp.lang.forth > #15795 > unrolled thread
| Started by | "B.Person" <bruce.person@gmail.com> |
|---|---|
| First post | 2012-09-28 07:16 -0700 |
| Last post | 2012-10-27 08:54 -0700 |
| Articles | 20 on this page of 97 — 19 participants |
Back to article view | Back to comp.lang.forth
Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) "B.Person" <bruce.person@gmail.com> - 2012-09-28 07:16 -0700
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) jacko <jackokring@gmail.com> - 2012-09-28 10:44 -0700
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) jacko <jackokring@gmail.com> - 2012-09-28 10:45 -0700
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) Mark Wills <forthfreak@gmail.com> - 2012-09-28 11:42 -0700
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) Jecel <jecel@merlintec.com> - 2012-09-28 11:39 -0700
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) Paul Rubin <no.email@nospam.invalid> - 2012-09-28 11:44 -0700
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) "Rod Pemberton" <do_not_have@notemailnotz.cmm> - 2012-09-29 14:55 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) Paul Rubin <no.email@nospam.invalid> - 2012-09-29 12:13 -0700
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) Bernd Paysan <bernd.paysan@gmx.de> - 2012-09-29 23:13 +0200
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) Paul Rubin <no.email@nospam.invalid> - 2012-09-29 17:30 -0700
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) "Elizabeth D. Rather" <erather@forth.com> - 2012-09-29 17:59 -1000
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. Anonymous <nobody@remailer.paranoici.org> - 2012-09-30 13:50 +0000
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. "Ed" <invalid@nospam.com> - 2012-10-02 23:53 +1000
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. Fritz Wuehler <fritz@spamexpire-201210.rodent.frell.theremailer.net> - 2012-10-03 19:10 +0200
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. "Elizabeth D. Rather" <erather@forth.com> - 2012-10-03 07:41 -1000
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. vandys@vsta.org - 2012-10-03 18:06 +0000
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. Bernd Paysan <bernd.paysan@gmx.de> - 2012-10-03 20:22 +0200
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. rickman <gnuarm@gmail.com> - 2012-10-04 08:39 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. "Elizabeth D. Rather" <erather@forth.com> - 2012-10-04 07:51 -1000
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. rickman <gnuarm@gmail.com> - 2012-10-04 14:05 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. "Elizabeth D. Rather" <erather@forth.com> - 2012-10-04 09:26 -1000
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. rickman <gnuarm@gmail.com> - 2012-10-04 19:23 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. vandys@vsta.org - 2012-10-05 00:41 +0000
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. rickman <gnuarm@gmail.com> - 2012-10-05 15:43 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. vandys@vsta.org - 2012-10-05 21:14 +0000
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. rickman <gnuarm@gmail.com> - 2012-10-07 17:58 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. "Ed" <invalid@nospam.com> - 2012-10-04 13:40 +1000
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. "Elizabeth D. Rather" <erather@forth.com> - 2012-10-03 21:34 -1000
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. Anonymous <nobody@remailer.paranoici.org> - 2012-10-04 09:49 +0000
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. "Ed" <invalid@nospam.com> - 2012-10-10 00:10 +1000
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. Bernd Paysan <bernd.paysan@gmx.de> - 2012-10-09 16:42 +0200
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. rickman <gnuarm@gmail.com> - 2012-10-09 14:01 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. Paul Rubin <no.email@nospam.invalid> - 2012-10-09 11:43 -0700
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. "Elizabeth D. Rather" <erather@forth.com> - 2012-10-09 09:02 -1000
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. rickman <gnuarm@gmail.com> - 2012-10-09 15:34 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. "Elizabeth D. Rather" <erather@forth.com> - 2012-10-09 09:53 -1000
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. rickman <gnuarm@gmail.com> - 2012-10-09 16:06 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. "Elizabeth D. Rather" <erather@forth.com> - 2012-10-09 17:00 -1000
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. rickman <gnuarm@gmail.com> - 2012-10-10 20:54 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. "Elizabeth D. Rather" <erather@forth.com> - 2012-10-11 19:26 -1000
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. rickman <gnuarm@gmail.com> - 2012-10-12 23:52 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. "Elizabeth D. Rather" <erather@forth.com> - 2012-10-13 07:56 -1000
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. rickman <gnuarm@gmail.com> - 2012-10-13 15:06 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. Bernd Paysan <bernd.paysan@gmx.de> - 2012-10-10 01:29 +0200
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. rickman <gnuarm@gmail.com> - 2012-10-09 15:31 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. Paul Rubin <no.email@nospam.invalid> - 2012-10-09 14:14 -0700
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. Bernd Paysan <bernd.paysan@gmx.de> - 2012-10-09 21:37 +0200
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. Paul Rubin <no.email@nospam.invalid> - 2012-10-09 15:26 -0700
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. Bernd Paysan <bernd.paysan@gmx.de> - 2012-10-10 02:21 +0200
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. Paul Rubin <no.email@nospam.invalid> - 2012-10-09 19:40 -0700
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. "Rod Pemberton" <do_not_have@notemailnotz.cmm> - 2012-10-04 17:07 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) "Rod Pemberton" <do_not_have@notemailnotz.cmm> - 2012-09-29 22:12 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) "Elizabeth D. Rather" <erather@forth.com> - 2012-09-29 18:38 -1000
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) Paul Rubin <no.email@nospam.invalid> - 2012-09-29 23:38 -0700
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) Jecel <jecel@merlintec.com> - 2012-09-30 11:47 -0700
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) rickman <gnuarm@gmail.com> - 2012-10-02 00:19 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) Bernd Paysan <bernd.paysan@gmx.de> - 2012-10-02 16:06 +0200
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) gavino_himself <visploveslisp@gmail.com> - 2012-10-05 14:50 -0700
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) Bernd Paysan <bernd.paysan@gmx.de> - 2012-10-06 03:14 +0200
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) Bernd Paysan <bernd.paysan@gmx.de> - 2012-09-30 16:34 +0200
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) "Rod Pemberton" <do_not_have@notemailnotz.cmm> - 2012-10-01 02:31 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) Paul Rubin <no.email@nospam.invalid> - 2012-10-01 01:27 -0700
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) Bernd Paysan <bernd.paysan@gmx.de> - 2012-10-01 16:57 +0200
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) "Rod Pemberton" <do_not_have@notemailnotz.cmm> - 2012-10-02 16:51 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) Bernd Paysan <bernd.paysan@gmx.de> - 2012-10-03 02:48 +0200
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) "Rod Pemberton" <do_not_have@notemailnotz.cmm> - 2012-10-04 17:47 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) Bernd Paysan <bernd.paysan@gmx.de> - 2012-10-05 00:39 +0200
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) "Rod Pemberton" <do_not_have@notemailnotz.cmm> - 2012-10-06 18:47 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) Paul Rubin <no.email@nospam.invalid> - 2012-10-02 19:23 -0700
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) gavino_himself <visploveslisp@gmail.com> - 2012-09-30 17:05 -0700
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) "Rod Pemberton" <do_not_have@notemailnotz.cmm> - 2012-10-01 02:26 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) albert@spenarnc.xs4all.nl (Albert van der Horst) - 2012-10-05 19:45 +0000
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) albert@spenarnc.xs4all.nl (Albert van der Horst) - 2012-10-04 13:22 +0000
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) awegel@arcor.de (Alex Wegel) - 2012-10-06 02:37 +0200
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) Mark Wills <forthfreak@gmail.com> - 2012-10-08 01:38 -0700
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) Chris Hinsley <chris.hinsley@gmail.com> - 2012-10-08 20:43 +0100
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) Chris Hinsley <chris.hinsley@gmail.com> - 2012-10-08 20:56 +0100
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) rickman <gnuarm@gmail.com> - 2012-10-08 16:00 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) Chris Hinsley <chris.hinsley@gmail.com> - 2012-10-08 21:05 +0100
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) rickman <gnuarm@gmail.com> - 2012-10-08 16:40 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) Chris Hinsley <chris.hinsley@gmail.com> - 2012-10-08 23:16 +0100
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) Mark Wills <forthfreak@gmail.com> - 2012-10-09 00:19 -0700
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) Chris Hinsley <chris.hinsley@gmail.com> - 2012-10-09 18:43 +0100
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) albert@spenarnc.xs4all.nl (Albert van der Horst) - 2012-10-09 09:59 +0000
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) Chris Hinsley <chris.hinsley@gmail.com> - 2012-10-09 18:46 +0100
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) marko <marko@marko.marko> - 2012-10-10 07:29 +1100
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) rickman <gnuarm@gmail.com> - 2012-10-09 17:11 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) Bernd Paysan <bernd.paysan@gmx.de> - 2012-10-10 01:11 +0200
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) rickman <gnuarm@gmail.com> - 2012-10-09 20:46 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) Bernd Paysan <bernd.paysan@gmx.de> - 2012-10-10 14:48 +0200
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) gavino_himself <visploveslisp@gmail.com> - 2012-10-12 13:35 -0700
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) rickman <gnuarm@gmail.com> - 2012-10-13 00:01 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) "Rod Pemberton" <do_not_have@notemailnotz.cnm> - 2012-10-13 20:51 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) rickman <gnuarm@gmail.com> - 2012-10-14 17:36 -0400
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) gavino_himself <visploveslisp@gmail.com> - 2012-10-12 13:34 -0700
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) "B.Person" <bruce.person@gmail.com> - 2012-10-26 13:35 -0700
Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) "B.Person" <bruce.person@gmail.com> - 2012-10-27 08:54 -0700
Page 4 of 5 — ← Prev page 1 2 3 [4] 5 Next page →
| From | "Rod Pemberton" <do_not_have@notemailnotz.cmm> |
|---|---|
| Date | 2012-10-01 02:31 -0400 |
| Message-ID | <k4bd63$da0$1@speranza.aioe.org> |
| In reply to | #15826 |
"Bernd Paysan" <bernd.paysan@gmx.de> wrote in message news:10709389.OQkU8GVx6R@sunwukong.fritz.box... > Rod Pemberton wrote: ... > And FYI: We are discussion a project that is mostly > competing with GPGPUs, not at all > with a typical general purpose microprocessor. I see nothing in either of the two articles related to "GPGPUs" or GPUs. So, where do you get that? Or, why do you think that? > It's also not using 16 ARM cores. True. However, that's an error by the authors of second article, at the link I posted. The wording of the first article is not especially clear that's the case either. It implies the RISC cores are ARM, not Adapteva's own RISC cores. It's poorly phrased. And, stop ignoring the point. It's irrelevant if the cores are ARM cores, or x86 cores, or even 6502 cores. The point is that it's going to be difficult to maximize usage of the multiple cores for many tasks. A novice home user will never be able accomplish it. The cores are only programmable in three languages: C, C++, and something no one knows ... So, that limits the ability to develop for the processor even further. The language no one knows: OpenCL, is C derived and the only one of the three that supports parallel processing. This is even more limiting. I.e., programming this thing requires expert programming skills in C or C++ and use of OpenCL. Yet, the project is supposedly developing the processor for the novices in the general public and home hobbyists. Does that really make sense to you? This combination is doomed. > You deliberately do the best you can to misunderstand people. No, I don't. > >> But as Moore's law has an > >> exponential cost in fabs, they get fewer and fewer. Therefore, > >> fabless is the future. > > > > Fabless is a _temporary_ solution driven by cost pressures of building > > a new fab. No one wants to pay the large upfront cost anymore. > > But fabs will become more expensive with every generation. True and still irrelevant. > How is that going ot be "temporary"? No, I expect one or two companies will eventually realize that the entire industry outsourced their fabs to eliminate the upfront cost of manufacturing a fab. They'll also realize that if their competitors are outsourcing the fab work and if they build and own their own fab, that their competitors will be at a competitive disadvantage financially. They'll be able to generate more profit, since they aren't outsourcing. Of course, the nature of financial competition for profit requires a company to not be at a competitive disadvantage financially relative to their competitors. Once they learn that their competition is making more profit by owning their own fab, they'll be forced to build a fab too or risk going out of business. Over time, the industry will slowly shift from fabless to owning fabs. If you don't understand this, you need to read about the American automotive industry sometime. They went through the same process. The cost of building a manufacturing plant became "astronomical". They outsourced. They sold off old plants. Then, one or two started building plants again to boost profits and disadvantage their competition. Now, they're all building plants. The plants are smaller, use less labor, use lower cost labor, have more robotics, are more efficient. In fact, the automotive industry has gone through just about everything that has ever happened to any major industry. I.e., what's happening to the semiconductor industry today has all happened before, but for different industries. > Do you expect fabs to become cheap again? It has nothing to do with the cost of the fab. It has to do with the total cost of the fab versus total cost outsourcing. The latter is more expensive. > If you can sell enough of your chips to fill your fab, > it is worth the cost. If they can't, they overbuilt. They should build smaller fabs next time. > AMD can't - fab costs are essentially up-front costs, so you have > to utilize it fully. So, they pay more. > Even Intel now started to let others use their > fab; these projects are still tiny, and probably more about learning how > to become a fab than to utilizing the fabs completely. Intel will face > this problem in future, too. The output per fab grows faster than the > total output Intel produces, and once they only need one fab for > producing all their stuff, they will have to open up. They better start > learning now how to do that; Intel is not stupid, they won't risk their > business by doing it too late. If they don't need the volume that a fab can produce for their own production, they're better off selling it. They can recover more capital that way. When opening up production to outside, the plant owner still has to pay fixed costs and plant operating costs, and labor costs. When they sell it, they don't. Rod Pemberton
[toc] | [prev] | [next] | [standalone]
| From | Paul Rubin <no.email@nospam.invalid> |
|---|---|
| Date | 2012-10-01 01:27 -0700 |
| Message-ID | <7xvceu7g2g.fsf@ruckus.brouhaha.com> |
| In reply to | #15841 |
"Rod Pemberton" <do_not_have@notemailnotz.cmm> writes: > The point is that it's going to be difficult to maximize usage of the > multiple cores for many tasks. A novice home user will never be able > accomplish it. I didn't get the sense this board was intended for "novice home users" to program. It could be that some home users would buy the boards to run programs written by hypothetical developers, or else that programmers would buy the boards as cheap development systems to write code for embedded targets. > The cores are only programmable in three languages: C, C++, and > something no one knows ... OpenCL is a well-accepted GPU language that's a less proprietary alternative to CUDA. Anyone who knows anything about GPU programming knows what it is. > It has nothing to do with the cost of the fab. It has to do with the total > cost of the fab versus total cost outsourcing. The latter is more > expensive. The fab is an awfully large cost up front. The company making this chip is roughly in the same boat as Green Arrays. They want to make this 64 core thing in 28nm. That's a very high tech, expensive process. While they are working on getting a fab, maybe they can also get the company cafeteria to stop buying vegetables from vendors, and instead grow them themselves to save money. That's less crazy than what you're suggesting.
[toc] | [prev] | [next] | [standalone]
| From | Bernd Paysan <bernd.paysan@gmx.de> |
|---|---|
| Date | 2012-10-01 16:57 +0200 |
| Message-ID | <2175391.upJk9XtueH@sunwukong.fritz.box> |
| In reply to | #15841 |
Rod Pemberton wrote:
> "Bernd Paysan" <bernd.paysan@gmx.de> wrote in message
> news:10709389.OQkU8GVx6R@sunwukong.fritz.box...
>> Rod Pemberton wrote:
> ...
>
>> And FYI: We are discussion a project that is mostly
>> competing with GPGPUs, not at all
>> with a typical general purpose microprocessor.
>
> I see nothing in either of the two articles related to "GPGPUs" or
> GPUs.
>
> So, where do you get that? Or, why do you think that?
I tried to find more meat, and downloaded an adapteva_mpr.pdf paper
(from MPRonline.com). There, they compare it with other cores, both a
Coretex A9 and a Vivante GC2000. The latter compares better than the
A9, and it is a GPU.
> I.e., programming this thing requires expert
> programming skills in C or C++ and use of OpenCL. Yet, the project is
> supposedly developing the processor for the novices in the general
> public and home hobbyists. Does that really make sense to you? This
> combination is doomed.
This is exactly the same combination people use to program big
supercomputers (ok, add Fortran into that mix ;-). And if you actually
bother to read the title of the thread, you see the word "supercomputer"
in it.
The thing is: Hobbyists have been trying a number of quite weird things
in the past, if these toys are just cheap enough. They have bought
Arduino boards to figure out what this embedded 8 bit microcontroller
thingy is; they are even willing to look at Forth on an Arduino.
You misunderstand what sort of "home hobbyist" this device is targetting
now: Not the people who are in the iPhone 5 queue. It's more about
geeks who want a new toy, to learn OpenCL. Novices in parallel
computing, not novices in programming.
The student (he's no longer student, and has a job now) in our Munich
Forth group, who plays with the GA144 is clearly also a home hobbyist.
There are hobbyists who do these strange things ;-).
>> You deliberately do the best you can to misunderstand people.
>
> No, I don't.
Yes, you do. Whatever word has more than one meaning, you always pick
the wrong one. I'll attribute it to incompetence instead of maliciousy
any day, and accept that you don't even realize what you are doing.
> Once they learn that their competition is making more profit by owning
> their own fab, they'll be forced to build a fab too or risk going out
> of business. Over time, the industry will slowly shift from fabless to
> owning fabs.
Well, we have these few companies who have their own fabs, and
especially Intel has always made more profit than anybody else. The
rest just struggles, and as consequence either opens up their fab to
others, or become fabless.
It's another misunderstanding of you: You absolutely have to fill your
fab. For memory makers, it's cut-throat; they have been dying like
flies in the past years.
> I.e., what's happening to the semiconductor industry today
> has all happened before, but for different industries.
Detroit is pretty special, with its combination of American management
("outsource everything") and strong unions. The semiconductor industry
did something completely different: They separated making and designing.
This is very natural, as the semiconductor industry is a "printer"
industry. The making is litography, i.e. printing. There are some
businesses in the printing industry where you need the press in-house
(newspapers, e.g.), but usually, the design (the writing) is separated
from the replication (the printing).
Maybe there will be some disruptive technology in future that allows
fabs to become much smaller again - as was with printing, when the laser
printer allowed everybody again to have his own "printing press".
>> Do you expect fabs to become cheap again?
>
> It has nothing to do with the cost of the fab. It has to do with the
> total cost of the fab versus total cost outsourcing. The latter is
> more expensive.
You can make smaller fabs, but these are less efficient - older
processes, smaller wavers, less output. The people who have small fabs
use old equipment and have their niche where this is profitable -
discretes, small analog components.
>> If you can sell enough of your chips to fill your fab,
>> it is worth the cost.
>
> If they can't, they overbuilt. They should build smaller fabs next
> time.
With smaller wavers? The fabs are "overbuild", because there still is a
big economy of scale there, and the money for the equipment maker comes
from big companies like Intel, TSMC, and Samsung. They want really big
fabs.
>> AMD can't - fab costs are essentially up-front costs, so you have
>> to utilize it fully.
>
> So, they pay more.
Yes. They did pay considerably more than they do now. It was killing
them.
> If they don't need the volume that a fab can produce for their own
> production, they're better off selling it.
Yes, so you finally got it. That's why they went fabless. And when you
start as a relatively small design house, you will be fabless from day
one, and never consider building your own fab. That's just a completely
different business.
--
Bernd Paysan
"If you want it done right, you have to do it yourself"
http://bernd-paysan.de/
[toc] | [prev] | [next] | [standalone]
| From | "Rod Pemberton" <do_not_have@notemailnotz.cmm> |
|---|---|
| Date | 2012-10-02 16:51 -0400 |
| Message-ID | <k4fjtf$so6$1@speranza.aioe.org> |
| In reply to | #15846 |
"Bernd Paysan" <bernd.paysan@gmx.de> wrote in message
news:2175391.upJk9XtueH@sunwukong.fritz.box...
> Rod Pemberton wrote:
> > "Bernd Paysan" <bernd.paysan@gmx.de> wrote in message
> > news:10709389.OQkU8GVx6R@sunwukong.fritz.box...
> >> Rod Pemberton wrote:
...
> > I.e., programming this thing requires expert
> > programming skills in C or C++ and use of OpenCL. Yet, the project is
> > supposedly developing the processor for the novices in the general
> > public and home hobbyists. Does that really make sense to you? This
> > combination is doomed.
>
> This is exactly the same combination people use to program big
> supercomputers (ok, add Fortran into that mix ;-). And if you actually
> bother to read the title of the thread, you see the word "supercomputer"
> in it.
>
If you had bothered to read the title of the thread, you'd see it also says
"your", as in me or you or anyone else. So, the point remains valid. It
says "next" too, as if everyone reading or posting here already has a
supercomputer crushing the concrete in their basement ... ;-)
> You misunderstand what sort of "home hobbyist" this device is targetting
> now: Not the people who are in the iPhone 5 queue. It's more about
> geeks who want a new toy, to learn OpenCL. Novices in parallel
> computing, not novices in programming.
I see no such "restriction".
Why did you both introduce "home hobbyist" and define what it means?
If you had read the article at the first link, you'd see that Adapteva
doesn't define who uses it as "home hobbyist" either. It lists
"enthusiasts", "hobbyists", and "developers". If you've got $99 and buy one
for yourself, you're one of the three. Your choice.
> >> You deliberately do the best you can to misunderstand people.
> >
> > No, I don't.
>
> Yes, you do. Whatever word has more than one meaning, you
> always pick the wrong one.
I'm a native speaker of English. I have a large vocabulary. I can speak
and write at an advanced level. So, how can you legitimately say that? In
a group like this, I prefer to write at a level similar to what would be
spoken, but a bit more fluently, precisely, accurately.
I don't translate to English through an intermediate language. I don't
mentally process concepts in a non-English language. Many of your
statements remind me of a few students I once worked with who spoke Chinese
and were learning English. They had a few pronunciation problems and
problems with pronouns. You have a few spelling problems and a few
comprehension issues. I've mentioned one of your misspellings more than a
few times. They understood the literal meanings of words, but had problems
grasping the concept of the words together, such as a phrase, sentence, or
paragraph. You do the same, but at a higher, more functional level. I.e.,
it's less noticeable, but still present. They mentally translated from
English to Chinese, processed the idea, translated from Chinese back to
English. Something becomes lost when that's done. The nuances of English
meaning are glossed over or ignored. They needed to learn to think in
English.
Actually, talking to you sometimes is a bit like talking to two people: one
who is fairly adept at English, and one who learned English as a second
language. Do you have someone helping you?
> It's another misunderstanding of you: You absolutely have to
> fill your fab. For memory makers, it's cut-throat; they have
> been dying like flies in the past years.
What's wrong with making a smaller fab? It's one way to fill your fab.
> > I.e., what's happening to the semiconductor industry today
> > has all happened before, but for different industries.
>
> Detroit is pretty special, with its combination of American management
> ("outsource everything") and strong unions. The semiconductor industry
> did something completely different: They separated making and designing.
So did the automotive industry, circa the early 1990's. Many of them moved
design from Detroit to California, Italy, Taiwan, Korea, China, etc.
> This is very natural, as the semiconductor industry is a "printer"
> industry. The making is litography, i.e. printing.
You mean lithography, not "litography". Specifically, it's
photolithography.
> There are some businesses in the printing industry where you need
> the press in-house (newspapers, e.g.), but usually, the design
> (the writing) is separated from the replication (the printing).
>
There you go again with the "lump and dump" - conflating lithography with
photolithography.
Photolithography is too highly specialized to compare it to the printing of
book and newspapers. Basically, you're conflating a simple process of
placing ink on paper with a complex process such as chemical etching and
vapor deposition of elements, like silicon, gallium, or arsenic, and also
re-classifying the work numerous Electrical Engineers as that of mere
wordsmiths.
> >> Do you expect fabs to become cheap again?
> >
> > It has nothing to do with the cost of the fab. It has to do with the
> > total cost of the fab versus total cost outsourcing. The latter is
> > more expensive.
>
> You can make smaller fabs, but these are less efficient - older
> processes, smaller wavers, less output. The people who have small fabs
> use old equipment and have their niche where this is profitable -
> discretes, small analog components.
>
Small doesn't mean less output. It depends on your rate of production.
Fast machines and three shifts a day can produce as much as a single shift
plant with large, slow machines. Most large factories use increased scale
as a replacement for speed, i.e., large, slow machines instead of fewer
smaller, faster machines.
Do you know what a related-rate problem is?
> >> If you can sell enough of your chips to fill your fab,
> >> it is worth the cost.
> >
> > If they can't, they overbuilt. They should build smaller fabs next
> > time.
>
> With smaller wavers? The fabs are "overbuild", because there still is a
> big economy of scale there, and the money for the equipment maker comes
> from big companies like Intel, TSMC, and Samsung. They want really big
> fabs.
>
These machines are all custom. There is nothing requiring them to be
constructed for large capacities. A bread baking machine in bread factory
or a sheet glass machine in a glass factory are sometimes over thousands of
feet long and hundreds of feet wide. But, nothing requires that they be
made that large. The can be made for whatever size is required. The
process and machine size are not linked. It's just a matter of planning,
design, and money.
> > If they don't need the volume that a fab can produce for their own
> > production, they're better off selling it.
>
> Yes, so you finally got it. That's why they went fabless.
That just indicates they overbuilt. There is still a financial advantage to
owning your own fab. They can make them smaller or more efficient, if they
want.
Rod Pemberton
[toc] | [prev] | [next] | [standalone]
| From | Bernd Paysan <bernd.paysan@gmx.de> |
|---|---|
| Date | 2012-10-03 02:48 +0200 |
| Message-ID | <8596609.knU0DjP0OM@sunwukong.fritz.box> |
| In reply to | #15868 |
Rod Pemberton wrote:
> I don't translate to English through an intermediate language. I
> don't mentally process concepts in a non-English language.
I don't say so. You just want to misunderstand people and therefore
choose the most inapropriate meaning of an ambiguous term in the next
round.
> You have a few spelling problems and a few
> comprehension issues. I've mentioned one of your misspellings more
> than a few times.
And you even needed to correct your own spelling... anal-retentive
stuff. Nobody else bothers here about the occasional mis-spelling. I
occasionally misspell in my own language, too, we all do, we respond to
that by doing best effort to try to understand what was written. This
is not a problem, apart for anal-retentive spelling Nazis, as we call
them in Germany. Why do you do it? It contributes *nothing* to the
discussion.
> They understood the literal meanings of words, but had
> problems grasping the concept of the words together, such as a phrase,
> sentence, or
> paragraph. You do the same, but at a higher, more functional level.
> I.e.,
> it's less noticeable, but still present.
I don't translate English into something else, particularly not into
Chinese. When I speak Chinese (my fourth language), I try not to
translate, either, it's a rather futile thing to do (Chinese words are
about as ambiguous as English words, but the different meanings they
cover is completely orthogonal - one English word translates to 5
Chinese words, and one of these Chinese words translate to 5 different
English words, which clearly aren't synonyms). I try to think in
Chinese. It's one of the things I would encourage every language
student: Try to think in the target language, don't try to translate.
However, I think you don't follow Postel's advice for communication: If
you speak to someone else, try to understand, not try to misunderstand.
If there's an interpretation that fits your understanding of things,
it's probably that you and I are in agreement. There's no need to find
the interpretation where we disagree.
Maybe you are not aware of the ambiguousities, because you don't know
any other language. Particularly Chinese is an eye-opener for that,
because you suddenly realize how context-sensitive the words in your own
language are, and how much you need to know the context to understand
Chinese, because the words there simply don't map to our words. Words
have a meaning only in context.
> Actually, talking to you sometimes is a bit like talking to two
> people: one who is fairly adept at English, and one who learned
> English as a second language. Do you have someone helping you?
No. I have definitely learned English as a second language, and don't
have someone helping me. It's just sometimes you don't understand me;
and I don't think in these cases it is my fault.
>> It's another misunderstanding of you: You absolutely have to
>> fill your fab. For memory makers, it's cut-throat; they have
>> been dying like flies in the past years.
>
> What's wrong with making a smaller fab? It's one way to fill your
> fab.
It's less efficient, i.e. the price per die will be higher. They don't
build bigger fabs for the fun of it, but because they want to be cheaper
than the competition. Furthermore, there aren't many companies who make
equipment; for leading edge processes, you essentially have to buy from
one company in the Netherlands (ASML); they have a sort-of (natural)
monopoly on the most important component - the stepper. That however
means that the equipment is designed for those who are willing to pay
the price - and those are the big manufacturers who want big fabs.
>> The semiconductor
>> industry did something completely different: They separated making
>> and designing.
>
> So did the automotive industry, circa the early 1990's. Many of them
> moved design from Detroit to California, Italy, Taiwan, Korea, China,
> etc.
Stop saying "automotive industry" when you want to say "Detroit". I
applied Postel's principle, and I clearly figure out that you use the
words "automotive industry" as synonym for "Detroit", and you ignore the
rest of the world, which also has their automotive industries.
Only Detroit destroyed itself by outsourcing everything to some other
place. Furthermore, when you say something like "moved design to
Italy", it sounds as if Italians haven't made cars before. FIAT is 9
years older than GM (the old GM, the one founded in 1908 which went
bankrupt 100 years later). You probably wouldn't say GM "moved design
to Germany", because, yes, they do (Opel), but that would make you sound
silly. Outsource it to the country where the car was invented? Maybe a
good idea (but Opels are rather lousy cars for German standards, though
they are clearly better than GM cars in America).
>> This is very natural, as the semiconductor industry is a "printer"
>> industry. The making is litography, i.e. printing.
>
> You mean lithography, not "litography". Specifically, it's
> photolithography.
See, you are doing it again. You perfectly understood me, but you think
it's worth to show off that your English is better than mine. What's
the point? It's supposed to pronounce "lit^{h}ography", but you
probably pronounce it with a fricative "th" (as in modern greek and
English, but this is supposed to be pronounced as ancient greek, where
theta was a plosive).
>> There are some businesses in the printing industry where you need
>> the press in-house (newspapers, e.g.), but usually, the design
>> (the writing) is separated from the replication (the printing).
>>
>
> There you go again with the "lump and dump" - conflating lithography
> with photolithography.
You apparently have an abstraction/analogies problem, you see the
differences but miss the similarities. The essence of lithography and
photolithography is that both make cheap replications possible -
replications of drawings, of books, of integrated circuits, and that the
replication process is independent of the creation process.
> Photolithography is too highly specialized to compare it to the
> printing of
> book and newspapers. Basically, you're conflating a simple process of
> placing ink on paper with a complex process such as chemical etching
> and vapor deposition of elements, like silicon, gallium, or arsenic,
> and also re-classifying the work numerous Electrical Engineers as that
> of mere wordsmiths.
Uh-oh. A writer or a painter who creates an *original*, which later is
replicated by printing, is an artist, his profession is not to do the
replication, but to create the original. As is the EE's task, who
creates a circuit, which is then "printed" in the fab. Both professions
require skills, quite different skills, but good writers are even rarer
than good EEs.
The fab however is not doing EE. They are doing lithography and
chemistry, their expertise is to control the process and keep the yield
high - they essentially operate really, really high-tech printers. It's
a well more advanced way of doing it than printing a newspaper, but the
essence is doing many, many cheap replications.
I've designed chips from top to botton, including (one time) device
engineering, which means creating a new type of transistor that wasn't
available in the design kit before.
However, I never was directly involved in *any* sort of actual
processing of the silicon in the fab, even during the years at Zetex,
which had their own fab (used e.g. for the low-scale integration analog
compagnion of the digital audio amplifier I developed there). Fab
people are not EEs. You confuse a printer with a writer. We visited
fabs to see how they do it and of course we talked to the fab people, so
we need some understanding of what they do. As in writing, between the
EE and the fab, there is a layouter - named the same way as the layouter
who does the typesetting of the words the writer has written down.
>> You can make smaller fabs, but these are less efficient - older
>> processes, smaller wavers, less output. The people who have small
>> fabs use old equipment and have their niche where this is profitable
>> - discretes, small analog components.
>>
>
> Small doesn't mean less output. It depends on your rate of
> production. Fast machines and three shifts a day can produce as much
> as a single shift plant with large, slow machines.
Rest assured that all high-end fabs around the world have the fastest
machines available, and are operated 24/7.
> Most large factories use increased
> scale as a replacement for speed, i.e., large, slow machines instead
> of fewer smaller, faster machines.
Have you ever been inside a fab? The machines are not particularly
large, they are room-height, the steppers are 2-3x5-10m or such, the
rest of the equipment is slightly smaller. In the free space, they have
robots to carry wafer trays from one machine to the next (the robots
don't look like R2D2, they are just a cross between cranes and small
railroads).
> Do you know what a related-rate problem is?
Yes. But this is completely irrelevant here.
> These machines are all custom.
Actually not. They are mass produced. For a not that big amount of
"mass", but they are definitely not built to order. Just look up
asml.com, they have nice photos, where you also can look inside the
machines. The steppers are the most essential machines for a fab, there
are other machines for etching and CVD, which however look similar from
the outside (they are all sealed to keep the dust out). The stepper is
doing the actual copying: mask to wafer. They can expose >100 wafers
per hour, so they aren't really slow. The rest of the equipment, the
CVD and etching, have comparable throughput rates, but you should
understand that each individual wafer stays for hours in any such
machine. The size of each of these things is similar to a stepper, see
for example this one, which set a record of 120wph:
http://www.novellus.com/products/product_lines/altus/
> There is nothing requiring them to be
> constructed for large capacities. A bread baking machine in bread
> factory or a sheet glass machine in a glass factory are sometimes over
> thousands of
> feet long and hundreds of feet wide. But, nothing requires that they
> be
> made that large. The can be made for whatever size is required. The
> process and machine size are not linked. It's just a matter of
> planning, design, and money.
Well, the semiconductor industry buys standard off-the-shelf products.
Lego bricks to build your fab, so to speak. The bricks get bigger and
bigger, though. When I visited the .35µm fab of AMS 10 years ago, the
stuff was a lot smaller than it is now for 32nm and below.
>> > If they don't need the volume that a fab can produce for their own
>> > production, they're better off selling it.
>>
>> Yes, so you finally got it. That's why they went fabless.
>
> That just indicates they overbuilt.
They have to buy and install new equipment every two years, so many
chances to correct that mistake. I don't think they overbuilt.
Let's look at the data: TSMC's 28nm fab has now 24k wafer starts per
month (and that's actually rather small, this is still in learning and
growing mode; GlobalFoundries combined 32/28nm fab has 50k wafer starts
per month and wants to grow to 80k, because now they have the customers,
not just AMD). ASML's twinscan XL, which is the appropriate stepper for
that fab, has 150wph, or 108k per month. It takes around 40 masks to
process a complete wafer (20 of them are metals and vias), so TSMC needs
only 10 of these steppers, if they are 100% utilized.
There isn't much room for less. Several process steps like the tungsten
contacts (done with the Altus linked above) are only done once per wafer
(contacts and local interconnects), and you want to utilize the entire
fab 100%. The single Altus with its 120wph is good for more than 80k
wafers per month. They won't sell that many of them; low two-digits, I
guess. I hope this illustrates the problem: The machines are quite
small and fast, but really, really expensive.
> There is still a financial advantage to
> owning your own fab. They can make them smaller or more efficient, if
> they want.
It's a fairly trivial calculation: Suppose your external fab has a cost
disadvantage of 30% per wafer - that's the profit they want to make.
Because they fab for everybody, they can fill their big, big fab, and
make it very efficient, by having 100% utilization for everything, and a
good yield. If they succeed to be 30% more efficient than your small
fab, the deal is done: Go fabless.
There was a time where you did need your own fab, that was during the
GHz race. Fabs like TSMC didn't offer the special processes you needed
for high-end processors; their processes were tuned for general purpose
ASICs, which didn't need that speed.
This has changed since. TSMC still has slower transistors, but Samsung,
GlobalFoundries, and IBM can provide you with the right processes to
build your supercomputer. And they are all open to the public, only
Intel's one half step ahead process isn't. And that lead is in danger:
ASML has troubles to make the EUV steppers production ready, and now all
the big fab companies and Intel did a joint investment into ASML, and
the result will be that everybody will get the new steppers at the same
time - Intel always paid a premium to get them first.
IMHO, apart from Intel and the memory makers, nobody today benefits from
having their own fab. The memory makers benefit, because they have
special needs (separate steps for the capacitors, which nobody else
makes, and still only two metal layers), and cut-throat competition.
Maybe with RRAM, that will go away - apparently it is possible to make
resistive RAMs with standard processes, i.e. without any special
equipment, and you also benefit from the many metal layers.
--
Bernd Paysan
"If you want it done right, you have to do it yourself"
http://bernd-paysan.de/
[toc] | [prev] | [next] | [standalone]
| From | "Rod Pemberton" <do_not_have@notemailnotz.cmm> |
|---|---|
| Date | 2012-10-04 17:47 -0400 |
| Message-ID | <k4kvuk$ms4$1@speranza.aioe.org> |
| In reply to | #15871 |
"Bernd Paysan" <bernd.paysan@gmx.de> wrote in message
news:8596609.knU0DjP0OM@sunwukong.fritz.box...
> Rod Pemberton wrote:
...
> And you even needed to correct your own spelling...
Once I'm aware a word is misspelled, I attempt to spell the word correctly
the next time and thereafter. I know I misspell when I haven't had
sufficient sleep. I know I have problems with certain words. I know I edit
and rewrite my posts and so end up with incorrect words or phrases
sometimes.
> Nobody else bothers here about the occasional mis-spelling.
Yours isn't occasional. For some words, it's permanent.
> Why do you do it? It contributes *nothing* to the discussion.
You responded to my post, but with an included insult. Why do you
insult people? "A typical RP argument ..." To use your words, "It
contributes *nothing* to the discussion."
Should I take it that you don't like valid criticism?
Should I take it that you don't like your faults exposed?
> However, I think you don't follow Postel's advice for communication: If
> you speak to someone else, try to understand, not try to misunderstand.
That's something you ignore in it's entirety.
The fact that you posted it to me is ironic.
> Words have a meaning only in context.
I agree. Can you please attempt to comprehend the context?
I ate a green apple in the morning, and I ate a green apple in the evening.
I salted the apple I ate in the evening. Which apple is the unripe red
apple? It could be either or neither or both, but it's more likely one is
green in color and the other is unripe. Salted or unsalted?
> It's just sometimes you don't understand me;
> and I don't think in these cases it is my fault.
Are you serious?
> >> It's another misunderstanding of you: You absolutely have to
> >> fill your fab. For memory makers, it's cut-throat; they have
> >> been dying like flies in the past years.
> >
> > What's wrong with making a smaller fab? It's one way to fill your
> > fab.
>
> It's less efficient, i.e. the price per die will be higher. They don't
> build bigger fabs for the fun of it, but because they want to be cheaper
> than the competition.
You're attempting to claim that the higher price per die over the operating
lifetime of a smaller fab will be substantially more than the upfront cost
of a much larger fab distributed on a per die basis? Doubtful.
> Furthermore, there aren't many companies who make
> equipment; for leading edge processes, you essentially have to buy from
> one company in the Netherlands (ASML); they have a sort-of (natural)
> monopoly on the most important component - the stepper. That however
> means that the equipment is designed for those who are willing to pay
> the price - and those are the big manufacturers who want big fabs.
Equipment manufacturers have to go where the money is too. If their
customers no longer want large equipment, they'll design smaller equipment.
They'll take custom orders too. They're designing very specialized
machines. If no one buys them, they're out of business.
> >> The semiconductor industry did something completely different:
> >> They separated making and designing.
> >
> > So did the automotive industry, circa the early 1990's. Many of them
> > moved design from Detroit to California, Italy, Taiwan, Korea, China,
> > etc.
>
> Stop saying "automotive industry" when you want to say "Detroit".
No, I don't mean Detroit. I mean the major automotive companies. At one
time, that meant the Detroit based Big-3. Today, I'd define it as General
Motors, Ford, Toyota, Daimler-Chrysler. Even Toyota designs in Michigan
and California.
> I applied Postel's principle, and I clearly figure out that you use the
> words "automotive industry" as synonym for "Detroit", and you ignore the
> rest of the world, which also has their automotive industries.
Clearly, it was a waste of your time to use "Postel's principle".
> Only Detroit destroyed itself by outsourcing everything to some other
> place. Furthermore, when you say something like "moved design to
> Italy", it sounds as if Italians haven't made cars before.
That's the way it sounds to you. That's not what was said.
> FIAT is 9 years older than GM (the old GM, the one founded in 1908
> which went bankrupt 100 years later).
The "new" GM is the "old" GM without the unused assets
and financial liabilities.
> You probably wouldn't say GM "moved design to Germany", because,
> yes, they do (Opel), but that would make you sound silly.
No, it would be inaccurate. AFAIK, nothing was moved there.
> >> This is very natural, as the semiconductor industry is a "printer"
> >> industry. The making is litography, i.e. printing.
> >
> > You mean lithography, not "litography". Specifically, it's
> > photolithography.
>
> See, you are doing it again. You perfectly understood me, but you think
> it's worth to show off that your English is better than mine. What's
> the point?
I just thought you misspelled it because you don't know how to spell it in
English, or you used a non-English spelling.
Just typing a few letters: "litog" into Google or Yahoo finds the correct
spelling. That takes you what, two seconds? Two seconds to not have
someone point it out. Is that worth it? Even simpler, enable your
spelling checker.
> It's supposed to pronounce "lit^{h}ography", but you
> probably pronounce it with a fricative "th" (as in modern greek and
> English, but this is supposed to be pronounced as ancient greek,
> where theta was a plosive).
In English, it's pronounced with a 'th':
http://www.merriam-webster.com/dictionary/lithography
http://dictionary.reference.com/browse/lithography?s=t
http://www.thefreedictionary.com/lithography
> >> There are some businesses in the printing industry where you need
> >> the press in-house (newspapers, e.g.), but usually, the design
> >> (the writing) is separated from the replication (the printing).
> >>
> >
> > There you go again with the "lump and dump" - conflating lithography
> > with photolithography.
>
> You apparently have an abstraction/analogies problem, you see the
> differences but miss the similarities. The essence of lithography and
> photolithography is that both make cheap replications possible -
> replications of drawings, of books, of integrated circuits, and that the
> replication process is independent of the creation process.
Mass production, injection molding, electronic component insertion, CAD/CAM,
lost foam casting, etc also make cheap replications possible. Do you intend
to classify all of them, and numerous other technologies and business
processes, as "'printer' industry" also?
> > Photolithography is too highly specialized to compare it to the
> > printing of book and newspapers. Basically, you're conflating a
> > simple process of placing ink on paper with a complex process
> > such as chemical etching and vapor deposition of elements, like
> > silicon, gallium, or arsenic, and also re-classifying the work
> > numerous Electrical Engineers as that of mere wordsmiths.
>
> Uh-oh. A writer or a painter who creates an *original*, which later
> is replicated by printing, is an artist, his profession is not to do the
> replication, but to create the original.
I should apologize to the artists for lumping them in with writers.
> As is the EE's task, who creates a circuit, which is then "printed"
> in the fab. Both professions require skills, quite different skills,
[...]
So?
> [...] but good writers are even rarer than good EEs.
Opinion. Many EEs come from degree "mills".
> The fab however is not doing EE. They are doing lithography and
> chemistry, their expertise is to control the process and keep the yield
> high - they essentially operate really, really high-tech printers. It's
> a well more advanced way of doing it than printing a newspaper, but the
> essence is doing many, many cheap replications.
True.
> > Do you know what a related-rate problem is?
>
> Yes. But this is completely irrelevant here.
Why? You seem to not understand that a process can operate at different
speeds and volumes.
> Actually not. They are mass produced. For a not that big amount of
> "mass", but they are definitely not built to order. Just look up
> asml.com, they have nice photos, where you also can look inside the
> machines.
Just because a machine is part of a company's product line doesn't mean it's
not custom. Highly specialized machines are custom.
> Well, the semiconductor industry buys standard off-the-shelf products.
That is true, but only to a certain extent. At some point, they need
equipment modified to their needs, or to fit the plant, etc.
> They have to buy and install new equipment every two years, so many
> chances to correct that mistake. I don't think they overbuilt.
...
> Let's look at the data: TSMC's 28nm fab has now 24k wafer starts per
> month (and that's actually rather small, this is still in learning and
> growing mode; GlobalFoundries combined 32/28nm fab has 50k wafer starts
> per month and wants to grow to 80k, because now they have the customers,
> not just AMD). ASML's twinscan XL, which is the appropriate stepper for
> that fab, has 150wph, or 108k per month. It takes around 40 masks to
> process a complete wafer (20 of them are metals and vias), so TSMC needs
> only 10 of these steppers, if they are 100% utilized.
>
> There isn't much room for less. Several process steps like the tungsten
> contacts (done with the Altus linked above) are only done once per wafer
> (contacts and local interconnects), and you want to utilize the entire
> fab 100%. The single Altus with its 120wph is good for more than 80k
> wafers per month. They won't sell that many of them; low two-digits, I
> guess. I hope this illustrates the problem: The machines are quite
> small and fast, but really, really expensive.
>
You say they want to utilize their entire fab. However, you cite examples
where they aren't, and you still claim they didn't overbuild. How does your
reasoning make sense to you? If they have any spare capacity for growth,
didn't they overbuild? Of course, they did. Any spare capacity indicates
they overbuilt. It's irrelevant that they intended for that capacity to be
used for growth. There is no guaranty they'll ever see any of the expected
growth. This is the same logic that caused the automotive industry to
design massive plants that eventually operated at only 30% of capacity for
decades.
"The expected growth didn't materialize."
"The market is in a downturn."
...
> > There is still a financial advantage to owning your own fab. They
> > can make them smaller or more efficient, if they want.
>
> It's a fairly trivial calculation: Suppose your external fab has a cost
> disadvantage of 30% per wafer - that's the profit they want to make.
> Because they fab for everybody, they can fill their big, big fab, and
> make it very efficient, by having 100% utilization for everything, and a
> good yield. If they succeed to be 30% more efficient than your small
> fab, the deal is done: Go fabless.
If they don't operate at or above a certain capacity levels, they're
operating at a loss until bankruptcy, or operating at break-even essentially
forever. If the capacity level is larger than break-even, they make a
profit.
The argument that semiconductor companies couldn't keep their fabs full,
i.e., lack of demand, was your prior argument for the industry selling their
fabs in the first place. Why do you expect outsourced fabs to be able to
fill their fab? Why do you expect that there is sufficient demand for that?
Consumed capacity is a function of supply and demand. There was
insufficient market demand in the first place to fill the fab for the fab's
original owner. So, they sold it. Why do you think there will be more than
enough demand for the fab's second owner if there wasn't for the first?
Rod Pemberton
[toc] | [prev] | [next] | [standalone]
| From | Bernd Paysan <bernd.paysan@gmx.de> |
|---|---|
| Date | 2012-10-05 00:39 +0200 |
| Message-ID | <1874283.RdYtElfWLd@sunwukong.fritz.box> |
| In reply to | #15931 |
Rod Pemberton wrote: >> It's just sometimes you don't understand me; >> and I don't think in these cases it is my fault. > > Are you serious? Yes, of course. >> It's less efficient, i.e. the price per die will be higher. They >> don't build bigger fabs for the fun of it, but because they want to >> be cheaper than the competition. > > You're attempting to claim that the higher price per die over the > operating lifetime of a smaller fab will be substantially more than > the upfront cost > of a much larger fab distributed on a per die basis? Doubtful. I've been working in that industry and I'm trying to explain it to you, but it is futile. > Equipment manufacturers have to go where the money is too. If their > customers no longer want large equipment, they'll design smaller > equipment. > They'll take custom orders too. They're designing very specialized > machines. If no one buys them, they're out of business. And that's why they make the big ones. The big semiconductor manufacturer drive the others out of the market, because being big is an advantage, they buy first, they buy most of the equipment. The most cut-throat business there is memories, and there aren't many left. >> Stop saying "automotive industry" when you want to say "Detroit". > > No, I don't mean Detroit. I mean the major automotive companies. > At one > time, that meant the Detroit based Big-3. Today, I'd define it as > General > Motors, Ford, Toyota, Daimler-Chrysler. Even Toyota designs in > Michigan and California. Daimler removed the Chrysler from its name several years ago. It's most popular brand is by far Mercedes, not Chrysler. And Volkswagen is one of the biggest automotive companies in the world, though it didn't buy into Detroit. If Daimler wouldn't have bought Chrysler or Toyota wouldn't have bought remainders of Detroit, you wouldn't even know about them. m( You don't mean "major automotive companies". You mean "Detroit". I feel like Alice talking to Humpty Dumpty. > Clearly, it was a waste of your time to use "Postel's principle". Ok, I stop now, delete the rest and put you back on my killfile. Unfortunately, it's by default a nice killfile and lets people pop up again after some time. Just one more thing: > The argument that semiconductor companies couldn't keep their fabs > full, i.e., lack of demand, was your prior argument for the industry > selling their fabs in the first place. Why do you expect outsourced > fabs to be able to fill their fab? Why do you expect that there is > sufficient demand for that? I've already told you that, several times. Such a spinned off fab is open for everyone. GlobalFoundry has several other design houses as customers. GlobalFoundries is a consolidation of a fab (UMC) and a processor maker (AMD), and together, they can fill this fab. This consolidation means that there are now less fabs than before (at least for the more advanced processes). -- Bernd Paysan "If you want it done right, you have to do it yourself" http://bernd-paysan.de/
[toc] | [prev] | [next] | [standalone]
| From | "Rod Pemberton" <do_not_have@notemailnotz.cmm> |
|---|---|
| Date | 2012-10-06 18:47 -0400 |
| Message-ID | <k4qc89$e6v$1@speranza.aioe.org> |
| In reply to | #15932 |
"Bernd Paysan" <bernd.paysan@gmx.de> wrote in message news:1874283.RdYtElfWLd@sunwukong.fritz.box... > Rod Pemberton wrote: ... > >> It's less efficient, i.e. the price per die will be higher. They > >> don't build bigger fabs for the fun of it, but because they want to > >> be cheaper than the competition. > > > > You're attempting to claim that the higher price per die over the > > operating lifetime of a smaller fab will be substantially more than > > the upfront cost > > of a much larger fab distributed on a per die basis? Doubtful. > > I've been working in that industry and I'm trying to explain it to you, > but it is futile. It's futile because you're going around and around in circles. You're not accepting the truth as such. You state one thing with an illogical conclusion. I bring up the illogical conclusion. You state another with an illogical conclusion. Eventually, you restate the your original premise. I.e., circle. You seem to think that repeating circle makes it true. > > Equipment manufacturers have to go where the money is too. If their > > customers no longer want large equipment, they'll design smaller > > equipment. They'll take custom orders too. They're designing very > > specialized machines. If no one buys them, they're out of business. > > And that's why they make the big ones. The big semiconductor > manufacturer drive the others out of the market, because being big is an > advantage, they buy first, they buy most of the equipment. The most > cut-throat business there is memories, and there aren't many left. This makes no sense. I know you don't understand why. The key word is "today". What you said is true only when semiconductor factories build their own fabs. That's not the market of "today". As you've noted, they've all sold their fabs. I.e., there are no big semiconductor manufacturers buying equipment for fabs. It's only new fabs buying equipment. If the new fabs choose to buy small, then the equipment manufacturers have to follow the market. > >> Stop saying "automotive industry" when you want to say "Detroit". > > > > No, I don't mean Detroit. I mean the major automotive companies. > > At one time, that meant the Detroit based Big-3. Today, I'd define it > > as General Motors, Ford, Toyota, Daimler-Chrysler. Even Toyota designs > > in Michigan and California. > > Daimler removed the Chrysler from its name several years ago. True, but I figured you wouldn't know who I was talking about. > It's most > popular brand is by far Mercedes, not Chrysler. And Volkswagen is one > of the biggest automotive companies in the world, though it didn't buy > into Detroit. If Daimler wouldn't have bought Chrysler or Toyota > wouldn't have bought remainders of Detroit, you wouldn't even know > about them. m( > I know about many of them. I very likely even know of companies you've probably never heard of. > You don't mean "major automotive companies". You mean "Detroit". I > feel like Alice talking to Humpty Dumpty. No, I didn't. It's still not Big-3. Toyota replacing one of the Detroit three, doesn't change the validity of what I said. > > The argument that semiconductor companies couldn't keep their fabs > > full, i.e., lack of demand, was your prior argument for the industry > > selling their fabs in the first place. Why do you expect outsourced > > fabs to be able to fill their fab? Why do you expect that there is > > sufficient demand for that? > > I've already told you that, several times. You still don't comprehend that your description violates economics, history, and logic. > Such a spinned off fab is > open for everyone. GlobalFoundry has several other design houses as > customers. GlobalFoundries is a consolidation of a fab (UMC) and a > processor maker (AMD), and together, they can fill this fab. This > consolidation means that there are now less fabs than before (at least > for the more advanced processes). > Selling off a fab doesn't change market demand. As you noted, the original fab owner could've opened their fab to outsiders, and as you noted, some did. Yet, the original fab owners still sold because of lack of demand. So, where's the extra demand? Did it just appear out of nowhere? Rod Pemberton
[toc] | [prev] | [next] | [standalone]
| From | Paul Rubin <no.email@nospam.invalid> |
|---|---|
| Date | 2012-10-02 19:23 -0700 |
| Message-ID | <7x7gr870pv.fsf@ruckus.brouhaha.com> |
| In reply to | #15868 |
"Rod Pemberton" <do_not_have@notemailnotz.cmm> writes: > Photolithography is too highly specialized to compare it to the printing of > book and newspapers. Basically, you're conflating a simple process of > placing ink on paper with a complex process such as chemical etching and > vapor deposition of elements "Printing" is a term I've heard for chip manufacture several times in the past few years, from people in the industry. I thought it was a one-off simplification the first time I heard it, but then I heard other people say it too. I think it may be a fairly recent idiom.
[toc] | [prev] | [next] | [standalone]
| From | gavino_himself <visploveslisp@gmail.com> |
|---|---|
| Date | 2012-09-30 17:05 -0700 |
| Message-ID | <742fcec0-5ec7-4eac-84de-35b5879ca22a@googlegroups.com> |
| In reply to | #15795 |
On Friday, September 28, 2012 7:16:25 AM UTC-7, B.Person wrote: > Posted 16 hours ago. Time to get into the ARM world; IMHO. > > > > http://arstechnica.com/information-technology/2012/09/99-raspberry-pi-sized-supercomputer-touted-in-kickstarter-project/ When $99 forth pc with a web browser? I will buy one.
[toc] | [prev] | [next] | [standalone]
| From | "Rod Pemberton" <do_not_have@notemailnotz.cmm> |
|---|---|
| Date | 2012-10-01 02:26 -0400 |
| Message-ID | <k4bcs9$cik$1@speranza.aioe.org> |
| In reply to | #15838 |
"gavino_himself" <visploveslisp@gmail.com> wrote in message news:742fcec0-5ec7-4eac-84de-35b5879ca22a@googlegroups.com... > On Friday, September 28, 2012 7:16:25 AM UTC-7, B.Person wrote: > > Posted 16 hours ago. Time to get into the ARM world; IMHO. > > > > When $99 forth pc with a web browser? I will buy one. You can by any x86 computer or laptop with a 486DX2 or more recent processor that will execute a web browser. Some are likely less than $99 today. Why does it need to be a Forth computer or laptop for you to buy it? Rod Pemberton
[toc] | [prev] | [next] | [standalone]
| From | albert@spenarnc.xs4all.nl (Albert van der Horst) |
|---|---|
| Date | 2012-10-05 19:45 +0000 |
| Message-ID | <506f38d9$0$3477$e4fe514c@dreader37.news.xs4all.nl> |
| In reply to | #15840 |
In article <k4bcs9$cik$1@speranza.aioe.org>, Rod Pemberton <do_not_have@notemailnotz.cmm> wrote: >"gavino_himself" <visploveslisp@gmail.com> wrote in message >news:742fcec0-5ec7-4eac-84de-35b5879ca22a@googlegroups.com... >> On Friday, September 28, 2012 7:16:25 AM UTC-7, B.Person wrote: >> > Posted 16 hours ago. Time to get into the ARM world; IMHO. >> > >> >> When $99 forth pc with a web browser? I will buy one. > >You can by any x86 computer or laptop with a 486DX2 or more recent processor >that will execute a web browser. Some are likely less than $99 today. Why >does it need to be a Forth computer or laptop for you to buy it? A raspberry pi will serve as a tv setup box to browse the net.(35 euros). It is not a Forth computer, but it will run gforth just fine. (It even runs mina in a Dosbox, and even my "full screen" editor works that patches characters at 0xB000 ). > > >Rod Pemberton > > Groetjes Albert
[toc] | [prev] | [next] | [standalone]
| From | albert@spenarnc.xs4all.nl (Albert van der Horst) |
|---|---|
| Date | 2012-10-04 13:22 +0000 |
| Message-ID | <506d8dac$0$3169$e4fe514c@dreader35.news.xs4all.nl> |
| In reply to | #15838 |
In article <742fcec0-5ec7-4eac-84de-35b5879ca22a@googlegroups.com>, gavino_himself <visploveslisp@gmail.com> wrote: >On Friday, September 28, 2012 7:16:25 AM UTC-7, B.Person wrote: >> Posted 16 hours ago. Time to get into the ARM world; IMHO. >> >> >> >> >http://arstechnica.com/information-technology/2012/09/99-raspberry-pi-sized-supercomputer-touted-in-kickstarter-project/ > >When $99 forth pc with a web browser? I will buy one. Yesterday I helped Leon Konings out with installing a 35 euro Raspberry. He has a high definition TV, cables laying around, an old wall wart for a cell phone, an unused SMD card, an USB keyboard from an old Apple. Bottom line he didn't have any extra costs for a linux computer with a web browser (and a couple of Forths, even chforth, running in a DOS box.) Groetjes Albert --
[toc] | [prev] | [next] | [standalone]
| From | awegel@arcor.de (Alex Wegel) |
|---|---|
| Date | 2012-10-06 02:37 +0200 |
| Message-ID | <1kriswi.krreno1j79ns8N%awegel@arcor.de> |
| In reply to | #15838 |
gavino_himself <visploveslisp@gmail.com> wrote: > When $99 forth pc with a web browser? I will buy one. Sounds pretty much like the description of a used powermac ;-)
[toc] | [prev] | [next] | [standalone]
| From | Mark Wills <forthfreak@gmail.com> |
|---|---|
| Date | 2012-10-08 01:38 -0700 |
| Subject | Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) |
| Message-ID | <971ca4fc-0ec5-457b-b941-c268c2cc48fe@z8g2000yql.googlegroups.com> |
| In reply to | #15795 |
On Sep 28, 3:16 pm, "B.Person" <bruce.per...@gmail.com> wrote: > Posted 16 hours ago. Time to get into the ARM world; IMHO. > > http://arstechnica.com/information-technology/2012/09/99-raspberry-pi... More on this. An article on The Register today: "The Epiphany core has a mere 35 instructions – yup, that is RISC alright – and the current Epiphany-IV has a dual-issue core with 64 registers and delivers 50 gigaflops per watt. It has one arithmetic logic unit (ALU) and one floating point unit and a 32KB static RAM on the other side of those registers." "Each core also has a router that has four ports that can be extended out to a 64x64 array of cores for a total of 4,096 cores. The currently shipping Epiphany-III chip is implemented in 65 nanometer processors and sports 16 cores, and the Epiphany-IV is implemented in 28 nanometer processes and offers 64 cores." "The secret sauce in the Epiphany design is the memory architecture, which allows any core to access the SRAM of any other core on the die. This SRAM is mapped as a single address space across the cores, greatly simplifying memory management. Each core has a direct memory access (DMA) unit that can prefetch data from external flash memory." Oh. That is seriously nice. See here: http://www.theregister.co.uk/2012/10/07/adapteva_epiphany_parallel_computing_kickstarter/
[toc] | [prev] | [next] | [standalone]
| From | Chris Hinsley <chris.hinsley@gmail.com> |
|---|---|
| Date | 2012-10-08 20:43 +0100 |
| Subject | Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) |
| Message-ID | <2012100820430026080-chrishinsley@gmailcom> |
| In reply to | #16041 |
On 2012-10-08 08:38:34 +0000, Mark Wills said: > "The secret sauce in the Epiphany design is the memory architecture, > which allows any core to access the SRAM of any other core on the die. > This SRAM is mapped as a single address space across the cores, > greatly simplifying memory management. Each core has a direct memory > access (DMA) unit that can prefetch data from external flash memory." > > Oh. That is seriously nice. > > See here: > http://www.theregister.co.uk/2012/10/07/adapteva_epiphany_parallel_computing_kickstarter/ > The hardware and SDK refernce manuals are avaiable for download now BTW. Check the kickstarter announcement, it has the links to download. Wish it had a bit more local memory, but yep, it looks quite nice. Chris
[toc] | [prev] | [next] | [standalone]
| From | Chris Hinsley <chris.hinsley@gmail.com> |
|---|---|
| Date | 2012-10-08 20:56 +0100 |
| Subject | Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) |
| Message-ID | <2012100820560326642-chrishinsley@gmailcom> |
| In reply to | #16062 |
On 2012-10-08 19:43:00 +0000, Chris Hinsley said: > On 2012-10-08 08:38:34 +0000, Mark Wills said: > >> "The secret sauce in the Epiphany design is the memory architecture, >> which allows any core to access the SRAM of any other core on the die. >> This SRAM is mapped as a single address space across the cores, >> greatly simplifying memory management. Each core has a direct memory >> access (DMA) unit that can prefetch data from external flash memory." >> >> Oh. That is seriously nice. >> >> See here: >> http://www.theregister.co.uk/2012/10/07/adapteva_epiphany_parallel_computing_kickstarter/ >> > > The hardware and SDK refernce manuals are avaiable for download now > BTW. Check the kickstarter announcement, it has the links to download. > > Wish it had a bit more local memory, but yep, it looks quite nice. > > Chris Actually the memory architecture is very nice, you can even run instructions from the memory of external cores. Esentially the memory address top 6 bits are an core (x,y) address, the lower bits the dram address. The scoreboarding of registers in the dependancy logic can cope fine with instructions and register reads/writes from/to external core addresses. Each hop of a core adds 1 cycle to the stall if you try to use a register prior to the data being available. You can think of the mesh network as 3 seperate 2D token rings, a core can fill in an empty slot as it goes by etc (arbitration logic round robbins the slots between CPU, DMA and the neibouring cores). 2 DMA channels per core that can read/write 2D memory regions, plus chained DMA descriptors so you can do nice Amiga style DMA blit lists etc. Plus I belive the DMA can do remote read/writes, so you can get a cores DMA engine to do transfers from a remote core to another remote core… The 32KB SRAM is organised as 4 8KB banks, each can be accsessed every cycle by various bits of the hardware, instruction read, DMA, etc etc. You need to keep instructions, data in, data out, buffers in seperate banks to hit maximum speed. Probably best for compute kernels ATM till more local memory is available, but there is room in the memory map for more per core ! Bring on the version with 64 cores and 1MB SRAM per core and things will get very interesting ! Chris
[toc] | [prev] | [next] | [standalone]
| From | rickman <gnuarm@gmail.com> |
|---|---|
| Date | 2012-10-08 16:00 -0400 |
| Subject | Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) |
| Message-ID | <k4vbco$v5m$1@dont-email.me> |
| In reply to | #16062 |
On 10/8/2012 3:43 PM, Chris Hinsley wrote: > > Wish it had a bit more local memory, but yep, it looks quite nice. > > Chris > WTF?!!! This thing has more memory on each processor than the GA144 has in total! Rick
[toc] | [prev] | [next] | [standalone]
| From | Chris Hinsley <chris.hinsley@gmail.com> |
|---|---|
| Date | 2012-10-08 21:05 +0100 |
| Subject | Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) |
| Message-ID | <2012100821052179545-chrishinsley@gmailcom> |
| In reply to | #16064 |
On 2012-10-08 20:00:22 +0000, rickman said: > On 10/8/2012 3:43 PM, Chris Hinsley wrote: >> >> Wish it had a bit more local memory, but yep, it looks quite nice. >> >> Chris >> > > WTF?!!! This thing has more memory on each processor than the GA144 > has in total! > > Rick :) Well I do like the idea of having a system like 64 super T800 trams on a single chip… call me old fationed but I liked the 1MB tram cards... Chris
[toc] | [prev] | [next] | [standalone]
| From | rickman <gnuarm@gmail.com> |
|---|---|
| Date | 2012-10-08 16:40 -0400 |
| Subject | Re: Your next supercomputer! Just $99 for 16 ARM cores on a chip. (link supplied) |
| Message-ID | <k4vdnt$em5$1@dont-email.me> |
| In reply to | #16065 |
On 10/8/2012 4:05 PM, Chris Hinsley wrote: > On 2012-10-08 20:00:22 +0000, rickman said: > >> On 10/8/2012 3:43 PM, Chris Hinsley wrote: >>> >>> Wish it had a bit more local memory, but yep, it looks quite nice. >>> >>> Chris >>> >> >> WTF?!!! This thing has more memory on each processor than the GA144 >> has in total! >> >> Rick > > :) > > Well I do like the idea of having a system like 64 super T800 trams on a > single chip… call me old fationed but I liked the 1MB tram cards... > > Chris I don't remember the name of the company, but someone was supposed to be working on a combined processor/memory which was more like a memory chip with an embedded processor than a processor with memory. Once they do that, there is no reason why it couldn't be an array of processors. Personally I never quite got the hang of Transputers. I did work with them for a while though. Not sure if the system ever got built. It was for a spook agency. Massively parallel. The software people felt the need to implement a full blown networking scheme on the links. I'm not sure how well that worked. Rick
[toc] | [prev] | [next] | [standalone]
Page 4 of 5 — ← Prev page 1 2 3 [4] 5 Next page →
Back to top | Article view | comp.lang.forth
csiph-web