16 bit audio has a S/N (signal/noise) ratio of 96 dB, which some 30 dB to much for the human ear. In other words, it is perfect for humans.
Adding extra bits improves the S/N ratio with 6 dB per bit, but this does not add more value.
Your perfect signal must go through an AD converter, then through wires, then through pre-amplification, power amplification and the loudspeakers
All these stages add estra noise which lessen your 96 dB S/N ratio.
Add to that also equalizing, which introduces still extra noise, and there is not a whole lot left.
So, no, Ozzy is not going to sound better on DVD-Audio, considering he has probably only a S/N ratio of 50 dB by himself.
Investing in extra bits above 16 for end-users, is just money thrown away. For people who need to record to separata channels and afterwards mix things, some extra bits might be worthwile, but probably not any more than four extra, ie. 20 bits in total.
You say you are mainly a Linux/Unix based operation ? Then why haven't you tried DOSEMU yet with your DOS based Canon software ? At least this could give you access to your documents on CD.
FWIW, electronic spreadsheets where created from paper spreadsheets, a technique which is used in theoretical mechanics to compute resulting vectors in space.
If you lookup the story about spreadsheets from Dan Bricklin, you will find that they really started from this concept.
With computers, you can automate many things which where previously cumbersome.
However, with computers you can do things that are impossible in reality : magic, anti-gravity, maybe we should go for patents on these because we can implement them on a computer...
It is easy to step up the clock cycle when you add more and shorter pipeline stages, but what about the interlocking dependencies.
One fact that many people do not seem to realize is that the current crop of CPU's is optimized for linear vector computations and to push as much bytes through as possible.
However, scientific computing and graphical applications are still in the minority of applications.
Everything that has to do with business or AI, which is highly complex and irregular, does not benefit from the current architectures. This is code with an enormous amount of jumps.
Superscalar or superpipelined architectures do not provide the necessary horsepower to process those applications.
Anyone knows of any benchmarks which do basic tests for such kind of code ?
Re:The thing I don't get about VLIW is this...
on
Itanium Problems
·
· Score: 1
I think I understand this point. The libraries themselves can be optimised and the compiled program can be optimised, but you cannot optimise the current program around the subroutine call, if you do not know the contents of the call.
To really reach the fastest program possible, you should inline all functions calls, then the optimiser could do really its best.
Re:It's not because of the economy
on
Itanium Problems
·
· Score: 1
I can applaud the decision to make a break from the old X86 architecture, but why did they design it as structurally complex bloatware?
First they head into the direction of more simplicity (switch to RISC core inside the CISC Pentiums) and then they double back into the complexity trap with Itanium.
Intel did only break out of the old X86 architecture after some of their competitors built Pentium class systems with a RISC core architecture. I think that you should even be able to find a document on the Internet from Intel, which argues about the benefits from CISC over RISC.
I think that people here are somewhat mixing up design and engineering.
You could say that design is coming up with something, with an idea, something that has a new quality, but no quantity yet.
To assign quantity to the idea, you need math. What math teaches you, is not only computation, but also model building.
Exam questions for calculus when studying did not concentrate on the numbers, but on the solutions. If you got the solution right, then you got the numbers right. We were allowed to use a calculator, but it did not help solving the problems.
I think of engineering more as implementing and refining the creation, but this is also in phases. At certain points you will need new (smaller) ideas, which will have to be implemented again.
The trajectory between a qualitative idea and a quantitative implementation can not be done by calculators, programs or computers. They can help speed up and manage the quantitative aspects better, but they are absolutely no substitute for our creativity.
There was someone who mentioned neural nets in the future. I think we can envision them with two possible outcomes : they are exact, in which case they stay computers which will not have any advantages over the human brain, or they should be trained and behave like other brains, in which case they will also have personalities and maybe personality deficiencies and so on, so they will probably also not have any advantages over the human brain.
A transaction can and will act as a lock, but a lock will absolutely not act as a transaction.
Transactions are used to protect the consistency of the database in case of a system failure. When the system would be up again, uncommitted transactions are rolled back and the system is consistent again. This usually does not take too long.
When using only locks, in case of a system failure, all your indices could be inconsistent and this takes a serious rebuild of them all, plus finding out which added records have garbage in them.
And the early nineties. I used to administer a small Novell 3.11 server, based upon a 33 MHz 386 processor, 8 Mb of RAM and 200 Mb of duplexed storage (raid-1). We served some 15 workstations from it and ran FoxPro across the network.
A few years ago I worked with Oracle and it was only easy to drop a column which was at the end of a row.
If we wanted to drop a column somewhere in the middle, we also had to write a script to query the database minus the deleted column, then delete the old table and rename the new table.
Novell used to offer Btrieve if one wanted to implement a transactional system. It worked together with their transactional system, which was an add-on.
I think that offering a real RDBMS on their servers, instead of indexed-sequential storage, should provide access to more developers.
Where they electronic, electric or mechanical failures ?
Mechanical failures are really a sign of bad quality from the manufacturer.
Electric failures (drive does not spin) could be a sign of bad quality from the manufacturer, but could also be related to heat.
Electronic failures are almost always a sign of bad handling by the company which builds them into the system. ESD precautions should be taken. However, ESD precautions should also be taken on the site where the systems are deployed. You will have more drive failures on sites with carpets than on sites with floors.
A year or six ago, Philips had such plans to create an add-on for motherboards to off-load video and audio processing from the CPU, and I think that there were other companies with the same plans.
It did not go through, because this would break the revenue model for Intel, which is based upon Moore's law : every eighteen months faster CPU's to create more demand.
So what did Intel do ? They created the MMX multimedia extensions and convinced several software publishers to support them. This way, they could say that their processors could handle all this processing and the Philips plans were almost doomed from the start.
Now they have MMX and fast processors, but some people still do not have enough processing power. The fact that 3D acceleration boards have success must mean that MMX is not really what it is supposed to be.
It seems to me that there is a cyclic movement between doing more things in software and then adding more specialised functions in hardware. I have seen this since the first articles on microcomputers appeared.
For the above suggestions (Speech recognition, database and network accelerator), I suggest you get "Computer architecture : A Quantitative Approach". Your suggestions have a far lower value than you might expect, because they would be costly to implement, and not used very much. I think that in that book there is even an example of a database accelerator on a mainframe system.
As for the network accelerator, "Computer architecture" also states that you can buy bandwidth, but you have to design for latency. One uses a network card to transfer data between computers. If you were to implement the things you said here in a network card, it would be a computer in itself. You would need to implement the communication between your computer and the network card as a network connection itself.
e.g. take a DNS on board your network card. Domain and host names are not a property of datagrams, but are things used the application to find hosts. This means that the computer should contact the network card as a DNS host, for which it needs a network stack.
If you want to simplify this, then you will have to create a new protocol between your computer and the network card, which is a waste of time I think.
Also, putting protocol stack functions on the network card effectively limits the card to the protocol chosen unless you implement it in software of course.
I know that here in Belgium amortisation of such high-tech stuff is three years. However, smaller companies tend to not replace PC's when they are still capable of doing all the work that they are needed for.
Mine is almost 6 years old. PPro 200 MHz. Since I upgraded my 2.4.16 kernel with preemptive and low-latency patches, I find that do not have the need for a faster processor at the moment.
How do artists get paid in America ?
on
RIAA Headway Dwindling
·
· Score: 2, Interesting
Here in Belgium we have SABAM, which I think should be kind of neutral because it groups artists, authors and publishers.
This means that it should maybe be possible to co-operate with such an agency and an online download system, which makes music available for the price that the artist wants (and should get). Downloads and payments on that system could be transferred to the agencies' database for payment to the artist, while at the same time the artist could get weekly and monthly reports about the amount of money he should get.
What could be needed to offer artists freedom of corporations ?
Digital recording software, in several levels of complexity, for home-made recordings up to real recording studio possibilities
A safe digital storage network for copyrighted works
A safe online payment system, which not only offers credit card payments, but also the possibility of subscriptions, which can be payed using all possible payment methods
Possibilities to download content
Possibilities to order pressed content (CD)
The biggest problem I see here is to know what the overhead costs are in setting up and maintaining this system, cost which of course should be spread over all the artists and works. Thus, the more artists and works there are on the system, the lower the cost per item, which means that people would probably pay a much fairer price.
Today, I found a link to a paper 'Thirty years later : Lessons from the Multics Security Evaluation'.
In it, a whole study is done about Multics and its security features. Multics was designed from the ground up with security in mind, and yet, on several points, didn't pass the test.
You can almost be sure that even with DRM and Palladium, it will not take a long time to breach all these things based upon them. Why ? Just because of the sloppy programming going on at MS.
Search the document and read it.
It also shows that for good security, you need some hardware features, but the necessary features have always been available in modern-day microprocessors. No need for something extra. It is just an extra factor which even could complicate things more from a security perspective.
My conclusion on Palladium ? Always the same about things Microsoft : we are not here to investigate if we are paranoid, but to investigate if we are paranoid enough.
Thus, people who fool themselves into thinking that their computing experience will be more secure due to Palladium, will be in for a nasty surprise.
People who think that Palladium will not be misused by Microsoft should show me something that has not been misused by them.
Slashdot Mirror
16 bit audio has a S/N (signal/noise) ratio of 96 dB, which some 30 dB to much for the human ear. In other words, it is perfect for humans.
Adding extra bits improves the S/N ratio with 6 dB per bit, but this does not add more value.
Your perfect signal must go through an AD converter, then through wires, then through pre-amplification, power amplification and the loudspeakers
All these stages add estra noise which lessen your 96 dB S/N ratio.
Add to that also equalizing, which introduces still extra noise, and there is not a whole lot left.
So, no, Ozzy is not going to sound better on DVD-Audio, considering he has probably only a S/N ratio of 50 dB by himself.
Investing in extra bits above 16 for end-users, is just money thrown away. For people who need to record to separata channels and afterwards mix things, some extra bits might be worthwile, but probably not any more than four extra, ie. 20 bits in total.
Do not know if you could call it on- or offtopic, but a nice story based upon misuse of patents is "The Iron Standard", by Henri Kuttner in 1943.
You say you are mainly a Linux/Unix based operation ? Then why haven't you tried DOSEMU yet with your DOS based Canon software ? At least this could give you access to your documents on CD.
FWIW, electronic spreadsheets where created from paper spreadsheets, a technique which is used in theoretical mechanics to compute resulting vectors in space.
If you lookup the story about spreadsheets from Dan Bricklin, you will find that they really started from this concept.
With computers, you can automate many things which where previously cumbersome.
However, with computers you can do things that are impossible in reality : magic, anti-gravity, maybe we should go for patents on these because we can implement them on a computer...
It is easy to step up the clock cycle when you add more and shorter pipeline stages, but what about the interlocking dependencies.
One fact that many people do not seem to realize is that the current crop of CPU's is optimized for linear vector computations and to push as much bytes through as possible.
However, scientific computing and graphical applications are still in the minority of applications.
Everything that has to do with business or AI, which is highly complex and irregular, does not benefit from the current architectures. This is code with an enormous amount of jumps.
Superscalar or superpipelined architectures do not provide the necessary horsepower to process those applications.
Anyone knows of any benchmarks which do basic tests for such kind of code ?
I think I understand this point. The libraries themselves can be optimised and the compiled program can be optimised, but you cannot optimise the current program around the subroutine call, if you do not know the contents of the call.
To really reach the fastest program possible, you should inline all functions calls, then the optimiser could do really its best.
I can applaud the decision to make a break from the old X86 architecture, but why did they design it as structurally complex bloatware? First they head into the direction of more simplicity (switch to RISC core inside the CISC Pentiums) and then they double back into the complexity trap with Itanium.
Intel did only break out of the old X86 architecture after some of their competitors built Pentium class systems with a RISC core architecture. I think that you should even be able to find a document on the Internet from Intel, which argues about the benefits from CISC over RISC.
Win95 still has no pre-emptive scheduling, only collaborative.
Yeah, I have still binary images, read with Linux and burned on CD-ROM, of MS-DOS 3.21, Compaq-DOS 3.31 and IBM-DOS 3.30. Why shouldn't he ?
In 1990 I have tested Windows 2.0... and I seem to remember that it included Solitaire.
Funny that they helped the businessman. I thought that ties were to be used to help them out of their agony.
I think that people here are somewhat mixing up design and engineering.
You could say that design is coming up with something, with an idea, something that has a new quality, but no quantity yet.
To assign quantity to the idea, you need math. What math teaches you, is not only computation, but also model building.
Exam questions for calculus when studying did not concentrate on the numbers, but on the solutions. If you got the solution right, then you got the numbers right. We were allowed to use a calculator, but it did not help solving the problems.
I think of engineering more as implementing and refining the creation, but this is also in phases. At certain points you will need new (smaller) ideas, which will have to be implemented again.
The trajectory between a qualitative idea and a quantitative implementation can not be done by calculators, programs or computers. They can help speed up and manage the quantitative aspects better, but they are absolutely no substitute for our creativity.
There was someone who mentioned neural nets in the future. I think we can envision them with two possible outcomes : they are exact, in which case they stay computers which will not have any advantages over the human brain, or they should be trained and behave like other brains, in which case they will also have personalities and maybe personality deficiencies and so on, so they will probably also not have any advantages over the human brain.
Using the cut, paste, select and join commands...
Whoa there...
A transaction can and will act as a lock, but a lock will absolutely not act as a transaction.
Transactions are used to protect the consistency of the database in case of a system failure. When the system would be up again, uncommitted transactions are rolled back and the system is consistent again. This usually does not take too long.
When using only locks, in case of a system failure, all your indices could be inconsistent and this takes a serious rebuild of them all, plus finding out which added records have garbage in them.
Yeah, but you know, the Free Software/OSS community can also use bait-and-switch...
And the early nineties. I used to administer a small Novell 3.11 server, based upon a 33 MHz 386 processor, 8 Mb of RAM and 200 Mb of duplexed storage (raid-1). We served some 15 workstations from it and ran FoxPro across the network.
Pretty stable...
A few years ago I worked with Oracle and it was only easy to drop a column which was at the end of a row.
If we wanted to drop a column somewhere in the middle, we also had to write a script to query the database minus the deleted column, then delete the old table and rename the new table.
Novell used to offer Btrieve if one wanted to implement a transactional system. It worked together with their transactional system, which was an add-on.
I think that offering a real RDBMS on their servers, instead of indexed-sequential storage, should provide access to more developers.
Where they electronic, electric or mechanical failures ?
Mechanical failures are really a sign of bad quality from the manufacturer.
Electric failures (drive does not spin) could be a sign of bad quality from the manufacturer, but could also be related to heat.
Electronic failures are almost always a sign of bad handling by the company which builds them into the system. ESD precautions should be taken. However, ESD precautions should also be taken on the site where the systems are deployed. You will have more drive failures on sites with carpets than on sites with floors.
A year or six ago, Philips had such plans to create an add-on for motherboards to off-load video and audio processing from the CPU, and I think that there were other companies with the same plans.
It did not go through, because this would break the revenue model for Intel, which is based upon Moore's law : every eighteen months faster CPU's to create more demand.
So what did Intel do ? They created the MMX multimedia extensions and convinced several software publishers to support them. This way, they could say that their processors could handle all this processing and the Philips plans were almost doomed from the start.
Now they have MMX and fast processors, but some people still do not have enough processing power. The fact that 3D acceleration boards have success must mean that MMX is not really what it is supposed to be.
It seems to me that there is a cyclic movement between doing more things in software and then adding more specialised functions in hardware. I have seen this since the first articles on microcomputers appeared.
For the above suggestions (Speech recognition, database and network accelerator), I suggest you get "Computer architecture : A Quantitative Approach". Your suggestions have a far lower value than you might expect, because they would be costly to implement, and not used very much. I think that in that book there is even an example of a database accelerator on a mainframe system.
As for the network accelerator, "Computer architecture" also states that you can buy bandwidth, but you have to design for latency. One uses a network card to transfer data between computers. If you were to implement the things you said here in a network card, it would be a computer in itself. You would need to implement the communication between your computer and the network card as a network connection itself.
e.g. take a DNS on board your network card. Domain and host names are not a property of datagrams, but are things used the application to find hosts. This means that the computer should contact the network card as a DNS host, for which it needs a network stack.
If you want to simplify this, then you will have to create a new protocol between your computer and the network card, which is a waste of time I think.
Also, putting protocol stack functions on the network card effectively limits the card to the protocol chosen unless you implement it in software of course.
I know that here in Belgium amortisation of such high-tech stuff is three years. However, smaller companies tend to not replace PC's when they are still capable of doing all the work that they are needed for.
Mine is almost 6 years old. PPro 200 MHz. Since I upgraded my 2.4.16 kernel with preemptive and low-latency patches, I find that do not have the need for a faster processor at the moment.
Here in Belgium we have SABAM, which I think should be kind of neutral because it groups artists, authors and publishers.
This means that it should maybe be possible to co-operate with such an agency and an online download system, which makes music available for the price that the artist wants (and should get). Downloads and payments on that system could be transferred to the agencies' database for payment to the artist, while at the same time the artist could get weekly and monthly reports about the amount of money he should get.
What could be needed to offer artists freedom of corporations ?
The biggest problem I see here is to know what the overhead costs are in setting up and maintaining this system, cost which of course should be spread over all the artists and works. Thus, the more artists and works there are on the system, the lower the cost per item, which means that people would probably pay a much fairer price.
Well, DRM also does not imply trust or security.
Today, I found a link to a paper 'Thirty years later : Lessons from the Multics Security Evaluation'.
In it, a whole study is done about Multics and its security features. Multics was designed from the ground up with security in mind, and yet, on several points, didn't pass the test.
You can almost be sure that even with DRM and Palladium, it will not take a long time to breach all these things based upon them. Why ? Just because of the sloppy programming going on at MS.
Search the document and read it.
It also shows that for good security, you need some hardware features, but the necessary features have always been available in modern-day microprocessors. No need for something extra. It is just an extra factor which even could complicate things more from a security perspective.
My conclusion on Palladium ? Always the same about things Microsoft : we are not here to investigate if we are paranoid, but to investigate if we are paranoid enough.
Thus, people who fool themselves into thinking that their computing experience will be more secure due to Palladium, will be in for a nasty surprise.
People who think that Palladium will not be misused by Microsoft should show me something that has not been misused by them.
If it sinks due to bad engineering, then it is a Titanic. If it gets torpedoed by the geeks/ordinary users/... then it is the Lusitania.