I'm glad you corrected the previous poster's mistake about pal code.
> (IA-64's instructions are, as I remember, 41 bits or so, with 3 of them packed in a 128-bit bundle.)
Speaking of instruction sizes, I think packing three instructions in 128 bit was a very bad design decision for Itanium. Given identical CPU-memory bandwidth, what is the advantage over packing four 32-bit instructions in 128 bit?
The answer is "none." VLIW touts high IPC but once you have the 4 instructions into the CPU, you can have as many instruction decoders and execution units as you wanted to, and there is nothing you can't do with superscalar machine that VLIW does. In fact, tightly coupling instructions is (1) a hindrance for oo and
(2) cause for code bloat.
(1) The problem with a company taking 7+ years for designing a chip is that by that time they FRS, the name of the game has changed. In early 90's, people were interested in solving the problem of executing highly linear single-thread code fast. In the 2000's, the people are interested in things like how fast a CPU can process multiple web requests and TPC. Which is why SMT is being looked at as the next natural step in the evolution of CPU's. Ironically VLIW makes oo and SMT very difficult to implement.
(2) A badly designed IA64 compiler can produce code that has over 50% nops. And if your code has short runs between branches, even a good compiler will need to stuff the edges with nops. And one must remember that each nops that a cpu fetches is taking away from bandwidth.
I have very little knowledge of Yamhill but insofar as it is unencumbered by VLIW, it seems a more promising architecture than Itanium.
I agree with your explanation of MIS. But the term Computer "Science" has been my pet peeve. Science means Physics, Chemistry, Earth science, Biology etc. Science is a field that seeks to better understand the workings of the nature. But when we deal with computer, everything is man-made. That's why I think "Computer Science" is a misnomer. I think Universities should stop using the term "Computer Science" since it degrades the term "science." "Software Engineering", "Applied Computation methods", "Software Art", etc will do.
From my 10+ years of selecting chips to build into products, the number one factor determining the long-term affordability is popularity, not material cost. In other words, chips that are easy to design with and provide a lot of functionality will be purchased by many companies and will go down the price curve faster than the chips that are supposedly cheaper to manufacture but have limited use.
Matrix semi says their product will be low-cost. I will believe them only when they publish the actual price per bit.
I've built a battery backed SRAM-based microcomputer in 1979 using chips called 5101, back in the days when "building a PC by yourself" actually meant soldering chips on circuit boards or wire-wrapping them together.
BTW, you do realize that srams take 6 transistors per cell whereas rom takes one.
This is old technology with new marketing. The technology is called OTP ROM and has been available for over 20 years. During development time of a product, engineers want to use UV-erasable PROMs. But once you release it to manufacturing, there is no need to erase the ROM. So the idea is to put an identical die with the same electrical and speed characteristics in two different packages, ceramic, and plastic. The UV-erasable ones have a crystal window and are packaged in ceramic enclosure, which is very expensive. The one-time programmable ones don't have a window and are packaged in plastic, which are very low-cost.
Fund it with advertisement
on
Lunar Lasers
·
· Score: 1, Funny
A trillion dollars? Easy! Fund it by selling ad space on the moon to the highest bidder. For example, if it were McDonald's, draw a big golden arch on the moon with laser. Every night when people look up the sky, they will feel like eating hamburger.
> Ask any merchant that takes credit cards and they'll tell you it's not even worth their effort to take the cards for transactions less than $20
Humm. . . everything I sell on the web is less than $20.
Micropayment doesn't work.
Several years ago I tried to sell a game program for 50 cents using the First Virtual's system.
Nobody bought.
Then I
started accepting credit card
and raised the price to $2
Then many people started buying.
If anybody reading this is a shareware author trying to sell through the web, I highly recommend that they apply for a credit card merchant account.
Masatoshi Shima is the inventor of microprocessor
on
Intel 4004 Turns 30
·
· Score: 1
Note that the IEEE interview of Shima says:
> But he [Hoff] jumped into the room and said, "I found some ideas." But it looks to me there were almost no new ideas . . .
But the Hoff interview says:
> And again, the idea [of a general purpose computer] was basically rejected by the Japanese engineers but . . .
I'm very surprised to learn that there is a big discrepancy between the US version of history and the Japanese version regarding who invented the first microprocessor. For all these years I thought that Mr. Masatoshi Shima, formerly of Busicom designed 4004 inspite of lack of understanding and support from Intel.
For example, link [1] writes (my translation): When Busicom approached Intel with the idea of designing an LSI to be used in calculators, Intel managers Ted Hoff and Stan Mazur showed little interest. Discussion went nowhere for many months. Then one day Hoff stormed into the room excited about "his" idea of creating a 4-bit ALU with 16 registers and 4 stacks, which was basically a rehash of Shimafs original proposal to create a chip based on the concept of "stored program." Moreover, Hoff designed only the arithmetic unit, which is the core of the microprocessor, so Shima had to write the spec and do the logic design for the rest of the chip all by himself. RAM, ROM, and I/O expander were designed by Federico Faggin, who had just joined Intel.
Slightly off-topic -- In general, I don't recommend PC-based measurement equipment. A company I used to work for bought a PC-based Oscilloscope trying to save some money. It was a cheap data acquisition card you plug in to an ISA slot and used software to look at waveforms. But in actuality the capture and rendering was ridiculously slow, and the controls hard to operate, so nobody was able to perform any real work with it. Also one must take into consideration the additional trouble of having to set up the PC near the lab bench, waiting for the PC to boot up, and the lack of isolation between the probe and PC.
In the end, the purchase was a short-lived expensive toy.
Speaking of GIF, does anybody know if it is legal to use GIF in our web pages? Today I came across an article that said GIF is a proprietary format of some company and anybody who wants to use GIF-encoded graphics in their web page must pay this company $5000 or face lawsuit? I appreciate any information concerning the current state of this issue.
I suspect that when people say they like or dislike a particular language, they often mean they like or dislike the entire development environment associated with that language. LISP needs a powerful IDE like MSCV++ with syntax-sensitive indenting, keyword auto-completion, and chroma-coding (perhaps they already do?)
(2) Non-intrusive garbage collection.
One thing that makes me shy away from using LISP for commercial product is the long pauses of the garbage collector that kicks in at unpredictable times. I would like to have a guaranteed maximum pause time.
(3) Efficient recursion
I would like to have an implementation that does not use up too much stack space with non-tail recursion. For example, the return path for a case like A calls B calls A.... probably can be compressed to 1 or 2 bits per call.
(4) Better handling of large projects
I don't have a specific suggestion on this but just a vague notion that large projects in LISP are hard to manage. Granted, part of the charm of LISP is that you can quickly write and run code without much planning, i.e., without declaring variables, and the binding is non-lexical i.e., last-visible. So my question is, is manageability and writability mutually exclusive?
Thank you for your reply. My scheme is different from the conventional CDR coding such as the one used in the Symbolics 3600 machine where the emphasis is on _reducing_ memory space. Rather, my scheme _wastes_ memory space in order to attain higher execution speed.
Part of the strength of LISP lies in the ability to execute data using the EVAL function. There are many cases where you want to do so. For example, you can easily write a theorem prover by calling EVAL on a prefix Boolean expression with exhaustive combination of the input variables.
Several years ago I wrote a term paper on my idea of a LISP machine in which CONS cells were verctorized in the CDR direction. For example, 8 cells will be stored in consecutive memory locations and the chunk will be handled by the processor as a single data unit. The hardware would be a VLIW architecture where operations will be performed simultaneously on all 8 cells. The simulation of such hypothetical machine exhibited an 8-fold speedup for list operations such as REVERSE and APPEND which encoded data in the CDR direction. There was no functional difference with conventional LISP at the user level except that REPLCD was disallowed. Also memory efficiency was somewhat sacrificed. Another conclusion from my study was that the width of the VLIW did not have to exactly match the width of the chunk in order to reap the benefit of this architecture. For example, a 4-wide VLIW could be used to process chunks consisting of 16 CONS cells with a performance comparable to this implemented with a 16-wide VLIW.
I wrote a game called SNUG, which is a puzzle of fitting random shapes into a box. There is no time limit to encourage deep thinking. This game improves children's spatial visualization skills. I received emails from teachers asking if they could use it in their schools. The answer is yes, because it's freeware. If you want to download it, here is where I put it:
http://www.puz.com/sw/freeware/page3.htm
I have a better idea. It's called a one-row keyboard. It only has the keys QWERTYUIOP.
The letter Q actually stands for 3 letters Q, A, and Z. W for W, S, and X, and so on. The computer tries all possible combinations and deduces the most appropriate word for the sentence that I am typing. For example, if I type Declaration of IYEEPEYEEYEE, it guesses that I was typing "Independence."
Confession of an Intel engineer
on
Pentium IV study
·
· Score: 1
Also there was an EET article that came out last December in which an Intel engineer confessed that the released Pentium 4 falls short of what they originally envisioned it to be. They scrapped the L3 cache, cut the L1 size in half, and gdumbed downh the FPUfs.
What most Americans don't realize is that there were many Japanese companies such as NEC, Fujitsu, Hitachi, Oki, Toshiba, etc that developed their own 4, 8 and 12 bit CPU's in the early 70's based on proprietary architecture and ISA, most of which were intended to be used in calculators and domestic consumption. None of the proprietary CPUs were successful as a general purpose CPU in the US market, but several microcontrollers (e.g. H8) and variants of Intel and Zilog architectures (e.g. V-series) were moderately successful in the US.
I have a theory why text messaging in Japan caught on earlier than the US. PC's were overpriced, (comparable units used to cost twice as much), typing is not taught in school, ISP and land-based telephone connection are overpriced (many areas still charge by the minute for local calls), which all contributed to using a home PC to exchange email a very cumbersome experience. So, it it's cumbersome anyway, then the inconvenience of entering text with one finger on small buttons is tolerable.
I think you are confused. Manufacturing wafer is different from "fabbing" wafer. Wafer means wafer. The pure, undoped silicon. The substrate. The fresh ingot sliced and pollished perfectly to the orientation of crystal lattice. What I said was Japan dominates the world in the production of raw material from which IC's are made.
Female CEOs do exist in Japan. I've seen a TV program featuring women who preside over nation-wide food chains, big hotels, Jewelry store, etc. (that they are report-worthy does admit to the scarcity). Also women control the finance in most Japanese families. A husband must ask his wife for monthly allowances. Also there was a period in Japan thousands of years ago in which apparently a woman ruled the country.
Microsoft is partially to blame for the reason you don't see any Japanese-written Windows software sold in the US. The Japanese version of Visual C++ links to the Japanese version of the MFC library, which most US users don't have installed in their PC.
Also their VC++ creates an application that uses Japanese fonts, which US machines cannot render. All Japanese PCs can render English fonts, plus Japanese consumers will make effort to read English in the English applications whereas American consumers don't make effort to read Japanese.
Microsoft probably has a good reason to try to shut out the Japanese from the US software market.
Slashdot Mirror
I'm glad you corrected the previous poster's mistake about pal code.
> (IA-64's instructions are, as I remember, 41 bits or so, with 3 of them packed in a 128-bit bundle.)
Speaking of instruction sizes, I think packing three instructions in 128 bit was a very bad design decision for Itanium. Given identical CPU-memory bandwidth, what is the advantage over packing four 32-bit instructions in 128 bit?
The answer is "none." VLIW touts high IPC but once you have the 4 instructions into the CPU, you can have as many instruction decoders and execution units as you wanted to, and there is nothing you can't do with superscalar machine that VLIW does. In fact, tightly coupling instructions is (1) a hindrance for oo and
(2) cause for code bloat.
(1) The problem with a company taking 7+ years for designing a chip is that by that time they FRS, the name of the game has changed. In early 90's, people were interested in solving the problem of executing highly linear single-thread code fast. In the 2000's, the people are interested in things like how fast a CPU can process multiple web requests and TPC. Which is why SMT is being looked at as the next natural step in the evolution of CPU's. Ironically VLIW makes oo and SMT very difficult to implement.
(2) A badly designed IA64 compiler can produce code that has over 50% nops. And if your code has short runs between branches, even a good compiler will need to stuff the edges with nops. And one must remember that each nops that a cpu fetches is taking away from bandwidth.
I have very little knowledge of Yamhill but insofar as it is unencumbered by VLIW, it seems a more promising architecture than Itanium.
I agree with your explanation of MIS. But the term Computer "Science" has been my pet peeve. Science means Physics, Chemistry, Earth science, Biology etc. Science is a field that seeks to better understand the workings of the nature. But when we deal with computer, everything is man-made. That's why I think "Computer Science" is a misnomer. I think Universities should stop using the term "Computer Science" since it degrades the term "science." "Software Engineering", "Applied Computation methods", "Software Art", etc will do.
From my 10+ years of selecting chips to build into products, the number one factor determining the long-term affordability is popularity, not material cost. In other words, chips that are easy to design with and provide a lot of functionality will be purchased by many companies and will go down the price curve faster than the chips that are supposedly cheaper to manufacture but have limited use.
Matrix semi says their product will be low-cost. I will believe them only when they publish the actual price per bit.
I've built a battery backed SRAM-based microcomputer in 1979 using chips called 5101, back in the days when "building a PC by yourself" actually meant soldering chips on circuit boards or wire-wrapping them together.
BTW, you do realize that srams take 6 transistors per cell whereas rom takes one.
Yes, I did read the article before posting. I was just objecting to the way people were focusing on the write-once aspect as if it were something new.
This is old technology with new marketing. The technology is called OTP ROM and has been available for over 20 years. During development time of a product, engineers want to use UV-erasable PROMs. But once you release it to manufacturing, there is no need to erase the ROM. So the idea is to put an identical die with the same electrical and speed characteristics in two different packages, ceramic, and plastic. The UV-erasable ones have a crystal window and are packaged in ceramic enclosure, which is very expensive. The one-time programmable ones don't have a window and are packaged in plastic, which are very low-cost.
A trillion dollars? Easy! Fund it by selling ad space on the moon to the highest bidder. For example, if it were McDonald's, draw a big golden arch on the moon with laser. Every night when people look up the sky, they will feel like eating hamburger.
> Ask any merchant that takes credit cards and they'll tell you it's not even worth their effort to take the cards for transactions less than $20
Humm. . . everything I sell on the web is less than $20.
Micropayment doesn't work. Several years ago I tried to sell a game program for 50 cents using the First Virtual's system. Nobody bought. Then I started accepting credit card and raised the price to $2 Then many people started buying. If anybody reading this is a shareware author trying to sell through the web, I highly recommend that they apply for a credit card merchant account.
Note that the IEEE interview of Shima says:
> But he [Hoff] jumped into the room and said, "I found some ideas." But it looks to me there were almost no new ideas . . .
But the Hoff interview says:
> And again, the idea [of a general purpose computer] was basically rejected by the Japanese engineers but . . .
I'm very surprised to learn that there is a big discrepancy between the US version of history and the Japanese version regarding who invented the first microprocessor. For all these years I thought that Mr. Masatoshi Shima, formerly of Busicom designed 4004 inspite of lack of understanding and support from Intel. For example, link [1] writes (my translation): When Busicom approached Intel with the idea of designing an LSI to be used in calculators, Intel managers Ted Hoff and Stan Mazur showed little interest. Discussion went nowhere for many months. Then one day Hoff stormed into the room excited about "his" idea of creating a 4-bit ALU with 16 registers and 4 stacks, which was basically a rehash of Shimafs original proposal to create a chip based on the concept of "stored program." Moreover, Hoff designed only the arithmetic unit, which is the core of the microprocessor, so Shima had to write the spec and do the logic design for the rest of the chip all by himself. RAM, ROM, and I/O expander were designed by Federico Faggin, who had just joined Intel.
[2] [3]
Slightly off-topic -- In general, I don't recommend PC-based measurement equipment. A company I used to work for bought a PC-based Oscilloscope trying to save some money. It was a cheap data acquisition card you plug in to an ISA slot and used software to look at waveforms. But in actuality the capture and rendering was ridiculously slow, and the controls hard to operate, so nobody was able to perform any real work with it. Also one must take into consideration the additional trouble of having to set up the PC near the lab bench, waiting for the PC to boot up, and the lack of isolation between the probe and PC.
In the end, the purchase was a short-lived expensive toy.
Speaking of GIF, does anybody know if it is legal to use GIF in our web pages? Today I came across an article that said GIF is a proprietary format of some company and anybody who wants to use GIF-encoded graphics in their web page must pay this company $5000 or face lawsuit? I appreciate any information concerning the current state of this issue.
(1) Better IDE
I suspect that when people say they like or dislike a particular language, they often mean they like or dislike the entire development environment associated with that language. LISP needs a powerful IDE like MSCV++ with syntax-sensitive indenting, keyword auto-completion, and chroma-coding (perhaps they already do?)
(2) Non-intrusive garbage collection.
One thing that makes me shy away from using LISP for commercial product is the long pauses of the garbage collector that kicks in at unpredictable times. I would like to have a guaranteed maximum pause time.
(3) Efficient recursion
I would like to have an implementation that does not use up too much stack space with non-tail recursion. For example, the return path for a case like A calls B calls A.... probably can be compressed to 1 or 2 bits per call.
(4) Better handling of large projects
I don't have a specific suggestion on this but just a vague notion that large projects in LISP are hard to manage. Granted, part of the charm of LISP is that you can quickly write and run code without much planning, i.e., without declaring variables, and the binding is non-lexical i.e., last-visible. So my question is, is manageability and writability mutually exclusive?
Thank you for your reply. My scheme is different from the conventional CDR coding such as the one used in the Symbolics 3600 machine where the emphasis is on _reducing_ memory space. Rather, my scheme _wastes_ memory space in order to attain higher execution speed.
Part of the strength of LISP lies in the ability to execute data using the EVAL function. There are many cases where you want to do so. For example, you can easily write a theorem prover by calling EVAL on a prefix Boolean expression with exhaustive combination of the input variables.
(1) And there was the Xerox 1108 workstation.
(2) If the FGCS had embraced LISP instead of Prolog, it probably would not have been such a colossal failure, IMO.
Several years ago I wrote a term paper on my idea of a LISP machine in which CONS cells were verctorized in the CDR direction. For example, 8 cells will be stored in consecutive memory locations and the chunk will be handled by the processor as a single data unit. The hardware would be a VLIW architecture where operations will be performed simultaneously on all 8 cells. The simulation of such hypothetical machine exhibited an 8-fold speedup for list operations such as REVERSE and APPEND which encoded data in the CDR direction. There was no functional difference with conventional LISP at the user level except that REPLCD was disallowed. Also memory efficiency was somewhat sacrificed. Another conclusion from my study was that the width of the VLIW did not have to exactly match the width of the chunk in order to reap the benefit of this architecture. For example, a 4-wide VLIW could be used to process chunks consisting of 16 CONS cells with a performance comparable to this implemented with a 16-wide VLIW.
What do you think of my idea?
I wrote a game called SNUG, which is a puzzle of fitting random shapes into a box. There is no time limit to encourage deep thinking. This game improves children's spatial visualization skills. I received emails from teachers asking if they could use it in their schools. The answer is yes, because it's freeware. If you want to download it, here is where I put it: http://www.puz.com/sw/freeware/page3.htm
I have a better idea. It's called a one-row keyboard. It only has the keys QWERTYUIOP. The letter Q actually stands for 3 letters Q, A, and Z. W for W, S, and X, and so on. The computer tries all possible combinations and deduces the most appropriate word for the sentence that I am typing. For example, if I type Declaration of IYEEPEYEEYEE, it guesses that I was typing "Independence."
Also there was an EET article that came out last December in which an Intel engineer confessed that the released Pentium 4 falls short of what they originally envisioned it to be. They scrapped the L3 cache, cut the L1 size in half, and gdumbed downh the FPUfs.
The company that dug the long hole connecting France to England was a Japanese company (I think Kawasaki heavy industry).
What most Americans don't realize is that there were many Japanese companies such as NEC, Fujitsu, Hitachi, Oki, Toshiba, etc that developed their own 4, 8 and 12 bit CPU's in the early 70's based on proprietary architecture and ISA, most of which were intended to be used in calculators and domestic consumption. None of the proprietary CPUs were successful as a general purpose CPU in the US market, but several microcontrollers (e.g. H8) and variants of Intel and Zilog architectures (e.g. V-series) were moderately successful in the US.
I have a theory why text messaging in Japan caught on earlier than the US. PC's were overpriced, (comparable units used to cost twice as much), typing is not taught in school, ISP and land-based telephone connection are overpriced (many areas still charge by the minute for local calls), which all contributed to using a home PC to exchange email a very cumbersome experience. So, it it's cumbersome anyway, then the inconvenience of entering text with one finger on small buttons is tolerable.
I think you are confused. Manufacturing wafer is different from "fabbing" wafer. Wafer means wafer. The pure, undoped silicon. The substrate. The fresh ingot sliced and pollished perfectly to the orientation of crystal lattice. What I said was Japan dominates the world in the production of raw material from which IC's are made.
Female CEOs do exist in Japan. I've seen a TV program featuring women who preside over nation-wide food chains, big hotels, Jewelry store, etc. (that they are report-worthy does admit to the scarcity). Also women control the finance in most Japanese families. A husband must ask his wife for monthly allowances. Also there was a period in Japan thousands of years ago in which apparently a woman ruled the country.
Microsoft is partially to blame for the reason you don't see any Japanese-written Windows software sold in the US. The Japanese version of Visual C++ links to the Japanese version of the MFC library, which most US users don't have installed in their PC.
Also their VC++ creates an application that uses Japanese fonts, which US machines cannot render. All Japanese PCs can render English fonts, plus Japanese consumers will make effort to read English in the English applications whereas American consumers don't make effort to read Japanese.
Microsoft probably has a good reason to try to shut out the Japanese from the US software market.