Initially I wiew this as a good idea. Bad health advice can be downright dangerous, having a strong certification mechanism is a good thing.
However it does set a precedent, and we should think carefully how this, and other TLDs should be managed. A TLD is too valuable to be handed out frivolously. We need to be confident they will not be abused.
The present TLD system is pretty much silly. Other than for the US TLDs we have national TLDs. For most things on the net I couldn't care less about what nation whatever info I am after resides in. Functional naming instead of geographical is IMHO a good thing (TM). However when we do that we have to ask who will manage these new TLDs. With geographical naming it was easy. Whoever governs that particular area decides, but noone governs a functional area.
What we need is some robust, independent mechanism to grant and revoke rights to manage TLDs. With all the talk about owning real estate on the net maybe it's time the net got it's own government.
An entity not under the control of any single nation or corporation, perhaps somthing under the UN umbrella?
If I remember the HARM correctly it's designed to fly to an IP (Initial Point), where it deploys a chute and looks down for threats. (I assume it can also be fired directly).
This gives the weapons one of it's most important functions. Not to destroy a target but to force enemy radars to shut down. This buys you the time you need to move your strike over the air defences and take them out using regular munitions.
Of cource if the enemy doesn't notice a HARM has been fired than their radar dish will shortly be a large pile of junk.
Rambus doesn't support higher frequencies due to serialisation, but because they use transmission lines to transfer the bits.
While clock synch is an issue, this isn't what constrains the speed of traditional memory interfaces.
sdrams modules (both traditional and ddr) use a simple node driver on it's output and don't do any tailoring of the waveform or impedance matching. This is fine as long as you keep most of your signal energy at longer wavelenghts than the length of the wire (or more precicely pcb trace) so that you can fully charge the entire trace to the voltage you want. Speedwise this technique got pretty hard limits on ow fast we can go, which cannot be resolved by just improving processing or circuit technique.
However as we increase the datarate pr. pin it become infeasable to shorten the trace enough to do the transmission this way.
Enter the transmission line. On this kind of interface we can have multiple bit 'on' the wire at the same time. This does however tight control of the wire's analogue properties and the waveform issued by the driver. When we do this we must limit the number and severity of bends and corners on the wires, and both sender and reciever must be designed to spesific impedances.
Since an TL interface is expensive both on die and pcb resources rambus opted to serialize the architecture to reduce costs and and increase capasity.
Eventually we will have to go to TL interfaces, not only to ram but just about all interfaces on our computers. However, right now it's apparent that TL tech isn't quite there yet. When they can push the speed up to say 2GHz we might want to reevaluate, though hopefully without rambus' disruptive influences.
Absolute area means little. It's population density which matters. (actually customer density, but it's close enough).
After all larger area with same density means more customers.
Here in Norway we the olg analogue NMT900 net was phased out in favor of GSM during the last ten years, and this is not an easy country for cellular. Lots of mountains and hills and a population which lives rather spread out into the countryside.
At present almost all populated areas got coverage. (This is a lot more than just cities (cities what cities, norways largest city (by far) got a population of just 0.5-1M (don't you just love nested paranthesis)))
My impression is that the US situation is mostly due to standards issues. When GSM was first introduced, the US seemed to adopt a 'not invented here' position and pressed on with it's CDMA system. As we have seen later on GSM is adopted anyway and the US got two competing standards.
Meanwhile over here we are now gearing up into upgrading the GSM system to encompass GPRS (packet radio), and later on to UMTS. UMTS is designed to be a joint audio/data network which can dynamically allocate bandwith both for both fixed and varying datarates becoming a mobile equivalent to ADSL (as in offering alway-on network connectivity. Don't expect to use UMTS for high-bandwidth streaming media)
Whoa... There is no way you are going to drive a trace that goes end to end on the die. Not if you hope to maintain that kind of clockspeeds. (You can be certan that this does not use an early-alpha style single clock node).
Interconnect capasitance will kill your speed long before you run into serious transmissionline effects (yet).
You do however mention a very real problem in the last paragraph. Which isn't so much related to propagation speed as RC delays and buffer delays. What you descrbe is called clock skew and is a very serious consideration for any designer of clocked systems (and has been so for a long while).
These days we have less parametres of skew to worry about since we generally use TSPC type logic (True single phase clock). Which means we only distribute one clock signal (compare that to NORA with 2(4) signals (4 if you don't rely on local clock inverters) or up to 4(8) for four phase logic). With TSPC we are only concerned with distribution skew. And as you mention: multipath problems are the worst here. With a very straight pipe we usually just send the clock in the opposite direction of the data, and thus no problem.
What you describe doesn't sound like any online bank I have seen.
I've personally used two systems. One was based on a java applet which generated a local key based on a one time password I got sent from the bank via snailmail. My selected password only decrypts my local key which resides on my machine, which is then used for further communication with the bank.
The other system uses a device protected by a pin code which generates one-time passwords.
As to reactions regarding bad press: True this is feared, but it is also hard to avoid as security problems can just as well be discovered from the outside as from the inside. The first implementation of net.banking in norway was indeed unsecure. A couple of local geeks blew the whistle and the system was taken down immidiatly.
In Norway it is permitted to make copies of copyrighted works for personal use (Exceptions exist for computer software and databases). Personal use means copies for yourself and/or friends and family. Practically this means you can make copies of your music and hand out to a few friends. You can _not_ put up an mp3 on an open web-server (unless you own it's contents).
As for enfocement. TONO (our equivalent to RIAA) haven't been to active in the media (possibly because they are currently more occupied by infighting and accusations of corruption of their execs). However mp3 servers operating too openly will likely recieve a ceace and desist fairly soon.
What does being from Sony have to do with having DD5.1 and DTS? There is no special relationship between Sony and DTS. In truth, *ANY* dvd player with digital audio out (copper sp/dif or optical toslink) will do both DD5.1 and DTS because "bits is bits" and the actual DD5.1/DTS processing is being handled by the receiver, not the dvd player.
Actually this is not entirely true. While bits is bits, thos bits have flags with them that say what they represent on a DVD. Each audiotrack has flags that say what language they are and what coding they use (and possibly others). Early DVD players (like my DV505) will not pass audio marked with any other flags than those it is set to handle, which in my case is DD and PCM. Also if it should ever become an issue: modern DVD players will not pass tracks marked as SDDS tracks (Which is also an optional format approved for DVD, which icendentally is owned by sony)
B.T.W. dts isn't too smart to use as recent testing indicates that dts at half bitrate (700ish) suffers from more compressionartifacts than DD at 380 (Both in 5.1 channels encoded of cource). And yes half bitrate dts is what's commonly used as full bitrate takes too much bandwidth away from the picture, leaving a marred image on screen. Of cource even half bitrate dts can be expected to have unfortunate sideeffects on the image as availible peak bandwidth is reduced (DVD forum rules dictate that a DD or PCM track must be included).
The reason many claim to prefer dts is likely related to that until recently dts had strict control of the use of their encoders (i.e. only they encoded material). Investigations have shown that there were significat differences between the original master and the encoded audio that could not be explained by compression artifacts. In other words dts tinkered with the original director-approved audio before they encoded it. This can be used to good effect to fool people into thinking that what they are listening to is a superior format. The common tricks are to boost overall levels (an increase of signal level of as little as 0.5dB will be percieved as an 'improvement' but not as a level difference), boost surround levels (more 'impressive' effects, but can get tireing in the long run, and can just as easily be achieved by adjusting the decoder settings), and increasing bass output.
While she won't need the kind of shielding an orbiter requires, going at M1.5 is potentially still a warming experience.
I'm not suggesting it is necceccary a problem. It depends at what altitude this speed is achieved (as she falls she will actually slow down after she achieves local terminal velocity), as well as how long she is exposed to high friction.
For comparison: A f-15 can achieve M2.5 but is limited in endurance above 2.0-2.3 somwhere by thermal buildup (Go too fast too long and it overheats). A concorde is significantly longer due to thermal expansion after it's supersonic dash. So while she won't be burnt to a crisp, I wouldn't rule out that there might be some noticeable maybe even uncomfortable thermal effect.
True: Rambus has less actual pins on the interface. However the interface is a transmissionline which reguires the traces on the PCB to be tightly controlled with regard to geometry and impedance.
For each rambus trace you will need at least one dedicated groundplane (possibly two) which takes up pcb real estate (not unlike how ATA66 requires a ground wire between each signal wire).
Also you must limit bends/corners on the traces since they create impedance problems (reflections), this puts severe restrictions on routing of these traces which can in turn cause less efficient usage of routing resources.
Also note that the interfaces themselves become more complex when you use transmissionlines. You will need some serious analogue signal processing in each end, which has a tendency to eat up chip real estate and is considerably more sensitive to environmental effects and manufacturing tolerances than digital circuitry.
That we will eventually need a transmissionline interface to ram is pretty obvious, but we should not go that route before that is the only option left to us. As long as we can do better with ordinary signalling it will likely be cheaper.
16 bit sampling is likely also above and beyond what's detectable error by human ears. (When properly dithered that is, but that goes for any other resolution as well)
At 16 bit you got 96 dB dynamic range which for a normal room means you must have peak levels at 126 (ish) dBspl, (which translates to immidiate loss of hearing). to raise the lsb above the noise floor.
The human hearing got a dynamic range at about 40dB (But has a slowly adjustable bias).
With the high pad count on what's essentially a dataswitch, your core will easily be pad-limited (Note however that modern bonding techniques can get considerably more pads in on a die than before)
That said; it isn't free to utilize extra 'filler' silicon on the die, as this will lower yield as defects that prevously didn't drigger a fault since it was on whitespace, now causes a defective unit.
You start running into _serious_ problems when you treat the band as strictly what the human ear can pick up. Apart from the fact that subsonics are picked up by the inner ear and supersonics can be sensed though not heard through bone conduction, the trouble is that you get cancellation effects and distortions depending on how you roll off the extremes of the band. This is a nightmarish problem for CD audio, as it must put a _really_ steep filter above 22K if not still lower- a brick wall filter that is about as bad as you can get for causing interactions with lower frequencies. Personally, I prefer to start rolling off a lot lower but a lot more smoothly, but that's just me.
In modern oversampling DA converters this is done in the digital domain where one can make a very well behaved filter. AD konversion on the studio side typically uses a higher resolution which is downsampled to 16 bit also digitally.
As for the sawtooth, I'm afraid that's the reality. Look, if you take the input signal a bit higher, you start getting a subharmonic through the sampling which can be almost as loud as the sampled frequency!
Now there is no such thing as 'subharmonic distortion' usually this term is incorrectly used for intermodulation distortion or aliasing. In this case you seem to refer to the latter. Obviously if one doesn't filter out frequency components over the f/2 limit one will suffer from distortion. At 22.045 KHz the antialiasingfilter, be it digital or analogue, should have reduced the input level to below the quantification noise limits (or sufficiently close).
Why you imply that aliasing distortion should affect frequency components below f/2 are however beyond me. While it's correct that to achieve perfect DA decoding require infinite computing power, we can approach perfection sufficiently that it's not reasonable to think we can hear the difference.
You seem to think that only a tone whose frequency goes an integer amount into fs can be correctly decoded. This is _not_ the case neither in theory nor in RL. I suggest a closer examination of the universal sampling theorem.
Correct decoding will look at more than the sample ta T=t to determine et correct momentary value, preceeding and following samples will also be used.
Of course, if these people are willing to settle for second-rate quality, the option of borrowing a VCR and making a tape copy STILL EXISTS! remember: an s-video out port has _no idea_ what happens at the other end. No system will _ever_ be devised where it is more difficult to send the video into a recording device than it is to send the video into a TV to watch it. Actually a system called macrovision which has existed for quite some years is capable of scrambling a vcr (primarily by messing with it's automatic gain control). This system is mandatory on a DVD player. Usually macrovision doesn't scramble a display device (some projectors are susceptible though, and you might be able to see some artifacts in the top of the frame on some tvs) In europe where many (most?) purchase players modified to play any zone very often also get macrovision disabled.
You will have to speed the clock yourself. Fortunatly on modern cpus you don't have to deliver the full core clock from an external net.
The reason cooling a cpu increases it's performance is threefold (and there is one minor factor that reduces performance when temperature is reduced).
1. One of the most significat problems in modern high complexity ic is power disappation. So much power need be disappated so quickly on a so small area that designers is often have to limit performance just to avoid having the ic releasing the blue smoke.
2. A synchronous ic need to time it's clock according to it's slowest net. These are usually a long metal interconnect, whose delay is aproximately derived from it's RC product. Reducing temperature reduces resistance and thus reduces tau=RC.
3. MOS transistors increase their gm at lower tempertures.
4. A minor detractor of reduced temp is that mobility in doped Si is reduced. (semiconductors actually have better conductance at higher temp)
Vitals on CD vs. LP: CD bandwidth: DC-22KHz LP bandwidth: varies with amplitude. Low frequenzies cannot be reproduced loudly. Upper limit can start out as high as 50KHz, but will be worn down to 16-18 KHz during the first few playback sessions.
CD dynamic range: 96 dB LP Dynamic range: starts around 85 dB, but is quickly worn down to 65-75dB.
(This makes LP equivalent of at most 14-15 bit resolution)
Regarding MP3: I agree it is easy to tell the diffrence between the CD original and MP3 encoded versions even on modest equipment.
Personally I wiew mp3 as a potential replacement for MDs for portable use. Portable players with at least 256 Mb storage need be availible first though. Given the current state of affairs that should take 3 years (moores law), by that time hopefully most of the dust have settled. (and a better standard than MP3 hopefully have been chosen)
Slashdot Mirror
Initially I wiew this as a good idea. Bad health advice can be downright dangerous, having a strong certification mechanism is a good thing.
However it does set a precedent, and we should think carefully how this, and other TLDs should be managed. A TLD is too valuable to be handed out frivolously. We need to be confident they will not be abused.
The present TLD system is pretty much silly. Other than for the US TLDs we have national TLDs. For most things on the net I couldn't care less about what nation whatever info I am after resides in. Functional naming instead of geographical is IMHO a good thing (TM). However when we do that we have to ask who will manage these new TLDs. With geographical naming it was easy. Whoever governs that particular area decides, but noone governs a functional area.
What we need is some robust, independent mechanism to grant and revoke rights to manage TLDs. With all the talk about owning real estate on the net maybe it's time the net got it's own government.
An entity not under the control of any single nation or corporation, perhaps somthing under the UN umbrella?
If I remember the HARM correctly it's designed to fly to an IP (Initial Point), where it deploys a chute and looks down for threats. (I assume it can also be fired directly).
This gives the weapons one of it's most important functions. Not to destroy a target but to force enemy radars to shut down. This buys you the time you need to move your strike over the air defences and take them out using regular munitions.
Of cource if the enemy doesn't notice a HARM has been fired than their radar dish will shortly be a large pile of junk.
Rambus doesn't support higher frequencies due to serialisation, but because they use transmission lines to transfer the bits.
While clock synch is an issue, this isn't what constrains the speed of traditional memory interfaces.
sdrams modules (both traditional and ddr) use a simple node driver on it's output and don't do any tailoring of the waveform or impedance matching. This is fine as long as you keep most of your signal energy at longer wavelenghts than the length of the wire (or more precicely pcb trace) so that you can fully charge the entire trace to the voltage you want. Speedwise this technique got pretty hard limits on ow fast we can go, which cannot be resolved by just improving processing or circuit technique.
However as we increase the datarate pr. pin it become infeasable to shorten the trace enough to do the transmission this way.
Enter the transmission line. On this kind of interface we can have multiple bit 'on' the wire at the same time. This does however tight control of the wire's analogue properties and the waveform issued by the driver. When we do this we must limit the number and severity of bends and corners on the wires, and both sender and reciever must be designed to spesific impedances.
Since an TL interface is expensive both on die and pcb resources rambus opted to serialize the architecture to reduce costs and and increase capasity.
Eventually we will have to go to TL interfaces, not only to ram but just about all interfaces on our computers. However, right now it's apparent that TL tech isn't quite there yet. When they can push the speed up to say 2GHz we might want to reevaluate, though hopefully without rambus' disruptive influences.
Absolute area means little. It's population density which matters. (actually customer density, but it's close enough).
After all larger area with same density means more customers.
Here in Norway we the olg analogue NMT900 net was phased out in favor of GSM during the last ten years, and this is not an easy country for cellular. Lots of mountains and hills and a population which lives rather spread out into the countryside.
At present almost all populated areas got coverage. (This is a lot more than just cities (cities what cities, norways largest city (by far) got a population of just 0.5-1M (don't you just love nested paranthesis)))
My impression is that the US situation is mostly due to standards issues. When GSM was first introduced, the US seemed to adopt a 'not invented here' position and pressed on with it's CDMA system. As we have seen later on GSM is adopted anyway and the US got two competing standards.
Meanwhile over here we are now gearing up into upgrading the GSM system to encompass GPRS (packet radio), and later on to UMTS. UMTS is designed to be a joint audio/data network which can dynamically allocate bandwith both for both fixed and varying datarates becoming a mobile equivalent to ADSL (as in offering alway-on network connectivity. Don't expect to use UMTS for high-bandwidth streaming media)
Also, interconnect capasitance get more significance over gate capacitance as chips become more complex and gates smaller.
Whoa... There is no way you are going to drive a trace that goes end to end on the die. Not if you hope to maintain that kind of clockspeeds. (You can be certan that this does not use an early-alpha style single clock node).
Interconnect capasitance will kill your speed long before you run into serious transmissionline effects (yet).
You do however mention a very real problem in the last paragraph. Which isn't so much related to propagation speed as RC delays and buffer delays. What you descrbe is called clock skew and is a very serious consideration for any designer of clocked systems (and has been so for a long while).
These days we have less parametres of skew to worry about since we generally use TSPC type logic (True single phase clock). Which means we only distribute one clock signal (compare that to NORA with 2(4) signals (4 if you don't rely on local clock inverters) or up to 4(8) for four phase logic). With TSPC we are only concerned with distribution skew. And as you mention: multipath problems are the worst here. With a very straight pipe we usually just send the clock in the opposite direction of the data, and thus no problem.
What you describe doesn't sound like any online bank I have seen.
I've personally used two systems. One was based on a java applet which generated a local key based on a one time password I got sent from the bank via snailmail. My selected password only decrypts my local key which resides on my machine, which is then used for further communication with the bank.
The other system uses a device protected by a pin code which generates one-time passwords.
As to reactions regarding bad press: True this is feared, but it is also hard to avoid as security problems can just as well be discovered from the outside as from the inside. The first implementation of net.banking in norway was indeed unsecure. A couple of local geeks blew the whistle and the system was taken down immidiatly.
In Norway it is permitted to make copies of copyrighted works for personal use (Exceptions exist for computer software and databases). Personal use means copies for yourself and/or friends and family. Practically this means you can make copies of your music and hand out to a few friends. You can _not_ put up an mp3 on an open web-server (unless you own it's contents).
As for enfocement. TONO (our equivalent to RIAA) haven't been to active in the media (possibly because they are currently more occupied by infighting and accusations of corruption of their execs). However mp3 servers operating too openly will likely recieve a ceace and desist fairly soon.
What does being from Sony have to do with having DD5.1 and DTS? There is no special relationship between Sony and DTS. In truth, *ANY* dvd player with digital audio out (copper sp/dif or optical toslink) will do both DD5.1 and DTS because "bits is bits" and the actual DD5.1/DTS processing is being handled by the receiver, not the dvd player.
Actually this is not entirely true. While bits is bits, thos bits have flags with them that say what they represent on a DVD. Each audiotrack has flags that say what language they are and what coding they use (and possibly others). Early DVD players (like my DV505) will not pass audio marked with any other flags than those it is set to handle, which in my case is DD and PCM. Also if it should ever become an issue: modern DVD players will not pass tracks marked as SDDS tracks (Which is also an optional format approved for DVD, which icendentally is owned by sony)
B.T.W. dts isn't too smart to use as recent testing indicates that dts at half bitrate (700ish) suffers from more compressionartifacts than DD at 380 (Both in 5.1 channels encoded of cource). And yes half bitrate dts is what's commonly used as full bitrate takes too much bandwidth away from the picture, leaving a marred image on screen. Of cource even half bitrate dts can be expected to have unfortunate sideeffects on the image as availible peak bandwidth is reduced (DVD forum rules dictate that a DD or PCM track must be included).
The reason many claim to prefer dts is likely related to that until recently dts had strict control of the use of their encoders (i.e. only they encoded material). Investigations have shown that there were significat differences between the original master and the encoded audio that could not be explained by compression artifacts. In other words dts tinkered with the original director-approved audio before they encoded it. This can be used to good effect to fool people into thinking that what they are listening to is a superior format. The common tricks are to boost overall levels (an increase of signal level of as little as 0.5dB will be percieved as an 'improvement' but not as a level difference), boost surround levels (more 'impressive' effects, but can get tireing in the long run, and can just as easily be achieved by adjusting the decoder settings), and increasing bass output.
While she won't need the kind of shielding an orbiter requires, going at M1.5 is potentially still a warming experience.
I'm not suggesting it is necceccary a problem. It depends at what altitude this speed is achieved (as she falls she will actually slow down after she achieves local terminal velocity), as well as how long she is exposed to high friction.
For comparison: A f-15 can achieve M2.5 but is limited in endurance above 2.0-2.3 somwhere by thermal buildup (Go too fast too long and it overheats). A concorde is significantly longer due to thermal expansion after it's supersonic dash. So while she won't be burnt to a crisp, I wouldn't rule out that there might be some noticeable maybe even uncomfortable thermal effect.
True: Rambus has less actual pins on the interface. However the interface is a transmissionline which reguires the traces on the PCB to be tightly controlled with regard to geometry and impedance.
For each rambus trace you will need at least one dedicated groundplane (possibly two) which takes up pcb real estate (not unlike how ATA66 requires a ground wire between each signal wire).
Also you must limit bends/corners on the traces since they create impedance problems (reflections), this puts severe restrictions on routing of these traces which can in turn cause less efficient usage of routing resources.
Also note that the interfaces themselves become more complex when you use transmissionlines. You will need some serious analogue signal processing in each end, which has a tendency to eat up chip real estate and is considerably more sensitive to environmental effects and manufacturing tolerances than digital circuitry.
That we will eventually need a transmissionline interface to ram is pretty obvious, but we should not go that route before that is the only option left to us. As long as we can do better with ordinary signalling it will likely be cheaper.
16 bit sampling is likely also above and beyond what's detectable error by human ears. (When properly dithered that is, but that goes for any other resolution as well)
At 16 bit you got 96 dB dynamic range which for a normal room means you must have peak levels at 126 (ish) dBspl, (which translates to immidiate loss of hearing). to raise the lsb above the noise floor.
The human hearing got a dynamic range at about 40dB (But has a slowly adjustable bias).
With the high pad count on what's essentially a dataswitch, your core will easily be pad-limited (Note however that modern bonding techniques can get considerably more pads in on a die than before)
That said; it isn't free to utilize extra 'filler' silicon on the die, as this will lower yield as defects that prevously didn't drigger a fault since it was on whitespace, now causes a defective unit.
In modern oversampling DA converters this is done in the digital domain where one can make a very well behaved filter. AD konversion on the studio side typically uses a higher resolution which is downsampled to 16 bit also digitally.
As for the sawtooth, I'm afraid that's the reality. Look, if you take the input signal a bit higher, you start getting a subharmonic through the sampling which can be almost as loud as the sampled frequency!
Now there is no such thing as 'subharmonic distortion' usually this term is incorrectly used for intermodulation distortion or aliasing. In this case you seem to refer to the latter. Obviously if one doesn't filter out frequency components over the f/2 limit one will suffer from distortion. At 22.045 KHz the antialiasingfilter, be it digital or analogue, should have reduced the input level to below the quantification noise limits (or sufficiently close).
Why you imply that aliasing distortion should affect frequency components below f/2 are however beyond me. While it's correct that to achieve perfect DA decoding require infinite computing power, we can approach perfection sufficiently that it's not reasonable to think we can hear the difference.
You seem to think that only a tone whose frequency goes an integer amount into fs can be correctly decoded. This is _not_ the case neither in theory nor in RL. I suggest a closer examination of the universal sampling theorem.
Correct decoding will look at more than the sample ta T=t to determine et correct momentary value, preceeding and following samples will also be used.
Of course, if these people are willing to settle for second-rate quality, the option of borrowing a VCR and making a tape copy STILL EXISTS! remember: an s-video out port has _no idea_ what happens at the other end. No system will _ever_ be devised where it is more difficult to send the video into a recording device than it is to send the video into a TV to watch it. Actually a system called macrovision which has existed for quite some years is capable of scrambling a vcr (primarily by messing with it's automatic gain control). This system is mandatory on a DVD player. Usually macrovision doesn't scramble a display device (some projectors are susceptible though, and you might be able to see some artifacts in the top of the frame on some tvs) In europe where many (most?) purchase players modified to play any zone very often also get macrovision disabled.
You will have to speed the clock yourself. Fortunatly on modern cpus you don't have to deliver the full core clock from an external net.
The reason cooling a cpu increases it's performance is threefold (and there is one minor factor that reduces performance when temperature is reduced).
1. One of the most significat problems in modern high complexity ic is power disappation. So much power need be disappated so quickly on a so small area that designers is often have to limit performance just to avoid having the ic releasing the blue smoke.
2. A synchronous ic need to time it's clock according to it's slowest net. These are usually a long metal interconnect, whose delay is aproximately derived from it's RC product. Reducing temperature reduces resistance and thus reduces tau=RC.
3. MOS transistors increase their gm at lower tempertures.
4. A minor detractor of reduced temp is that mobility in doped Si is reduced. (semiconductors actually have better conductance at higher temp)
Vitals on CD vs. LP:
CD bandwidth: DC-22KHz
LP bandwidth: varies with amplitude. Low frequenzies cannot be reproduced loudly. Upper limit can start out as high as 50KHz, but will be worn down to 16-18 KHz during the first few playback sessions.
CD dynamic range: 96 dB
LP Dynamic range: starts around 85 dB, but is quickly worn down to 65-75dB.
(This makes LP equivalent of at most 14-15 bit resolution)
Regarding MP3: I agree it is easy to tell the diffrence between the CD original and MP3 encoded versions even on modest equipment.
Personally I wiew mp3 as a potential replacement for MDs for portable use. Portable players with at least 256 Mb storage need be availible first though. Given the current state of affairs that should take 3 years (moores law), by that time hopefully most of the dust have settled. (and a better standard than MP3 hopefully have been chosen)