Short-circuit current is only responsible for 10-20% of switching power. The rest is dissipated in the transistor through charging and discharing all the nodal capacitances (due to transistor gates, transistor diffusions and wiring capacitance). Since typical circuit styles are non-adiabatic, this charge/discharge power component would not go away even if we could completely eliminate short-circuit currents.
Making transistors smaller certainly reduces their gate capacitance but it also reduces their current drive by the same proportion. These two effects cancel each other out! So how can transistors get faster from generation to generation?
Transistors get faster by increasing electron mobility and/or increasing gate capacitance per unit area and/or reducing diffusion junction/sidewall capacitance per unit area/perimiter and/or reducing (local) interconnect capacitance since smaller transistors are closer together.
It's easy to reduce the tunneling current through the gate. All you have to do is increase the thickness of the insulator. Unfortunately, this has the detrimental effect of reducing the effective capacitance of the gate, which in turn lowers the amount of current conducted by atransistor of a given size (lowering the current also lowers the speed). To make up for the lowered gate capacitance, researchers have been trying to increase the dielectric constant of the insulator. I'm guessing that they're proposing a method to increase the dielectric constant of the gate insulator. The devil is in the details of improving the dielectric constant without screwing up later processing steps or reducing the mechanical integrity of the wafer, etc.
Summary:
To lower gate leakage simply increase dielectric thickness.
To make up for lost speed due to higher dielectric thickness, increase dielectric constant.
The computer industry, however, tends to move slowly when it comes to major overhauls of computer architecture.
Huh? The computer industry, and more specifically, the processor industry, is the fastest adopter of new technology in any endeavor in human history. Anyone have an example of a faster-moving industry?
Darkmail is primarily used in distributed denial of service (DDoS) attacks and directory harvest attacks (DHA) in which a specific domain is hit with a flood of emails through an alphabetical list of names.
But over the last year darkmail is being used to brute-force spam through filters and is clogging up bandwidth.
Basically, it seems that darkmail is bulk mail sent to a domain with the advance knowledge that much of it will not reach a destination.
Zero gain bandwidth isn't really the technical term for it.
There are actually two common metrics for the max speed of the transistor: fT and fMAX. fT is the frequency at which the current gain of the transistor is equal to 1 (not zero---the confusion may arise from the fact that a gain of 1 equals a gain of 0 dB). fMAX is the frequency at which the power gain of the transistor is equal to 1. The fMAX is a much more useful and meaningful metric than fT but is extremely difficult to measure as it is very sensitive to the test conditions. Because of the difficulty of measuring fMAX, often fT is reported instead. It's not clear from the article which of these two frequencies they are reporting in this research.
Typically, there is a loss of speed when connecting multiple transistors, so the fastest circuits will operate at some fraction of fMAX. My guess is that Intel's chips probably operation at about 0.05 to 0.1 of fMAX. Anyone who knows the fMAX of of a typical 130nm or 90nm process care to confirm?
Don't feel sorry for Paul Graham. He chose his path and he enjoys it. You chose a different path and are happy with your choice. Good for both you. I don't feel sorry for either of you.
It takes a lot of work to learn how to program in Java correctly -- even more effort than would be required to learn Python I'd venture (based on my extensive experience...).
Seems like a good motivation to avoid Java. Often, part of being smart is also being somewhat lazy. Based on your description, I would much rather learn Python than Java
A good Java programmer has to be smart.
We are comparing relative intelligence here. Obviously, anyone who can learn ANY modern programming language is reasonably smart. Nevertheless, there are large ranges of intelligence even within the group of people smart enough to learn to program.
I can agree that I've encountered a whole lot of Java developers out there who write bad code. But I don't think the ratio is any different than in any other popular language.
I think Paul Graham would disagree that this ratio is the same for all languages. His experience and understanding of programming languages is deep (which lends credence to his thoughts, but obviously doesn't mean he is automatically correct). Is your experience and understanding deep enough to really justify your claim? Can you give the approximate ratio for 3 different programming languages?
On a related note, I was having dinner at a restaurant and my waiter asked me for a recommendation for a good email program. So I guess it turns outs that I have a server that needs a client.
Is this the end of moores law, at least in the form of CPU speeds doubling every 18 months?
There are essentially two CPUs, I doubt each of them will get 2x faster the next 1.5 years:)
There have been quite a few posts pointing out that Moore's law actually refers to exponential growth in transistor density rather than speed.
The posters are technically correct, but the term Moore's law has come to encompass any processor-related metric that changes at an exponential pace, including processor performance, clock rate, and power consumption. Of course, these metrics are directly related to transistor size and density, so it makes sense that they have changed exponentially.
I guess what confuses people is that there is a lot of research into trying to make two processor systems exactly twice as fast as a single processor system. Keep in mind this is for a single application. When a single application (or process) is spread across multiple CPUs, the CPUs often must communicate information to complete the computation. The cost of this cross-CPU communication is high. A single CPU system, however, has no communication bottleneck. Thus the performance of a 2-CPU system is often much less that twice that of the single CPU system--but always better than just one processor as noted by the parent posting.
But if we are running multiple, independent applications/processes, then there's little problem. There is no need for communication between unrelated processes. So there's no communication bottleneck. If you run an environment with computation-intensive applications that are indepedent, you will get twice the performance* with a 2-CPU system versus a single CPU system.
*I have simplified the situation. There is indeed overhead, but this is minor compared to parallelizing a single application/process.
Two processors are better than one, period.
...and in some cases, two processors are twice as good as single processor.
Sorry for adding to a sentence that ends with the word period. But I just can't help it, period.
First off, First Monday touts itself as a peer-reviewed journal. I have to say, that this is probably the shittiest article I have ever read in a peer-reviewed journal (related to serious technical material---most soft science journal are essentially worthless). First Monday just got added to my list of journals to not waste my time reading
These same 5 attributes exist in the closed source world. I'm sure many of you have used closed source software with poor UI's, documentation, feature-centric, etc. These problems are present in all software. Perhaps they are more prevalent in open source software, but that argument would have to be well-researched to be convincing, unlike the slop in this article.
Just do a "s/open source/closed source/g" on the article, and it still makes sense. Even the part about religious blindness is applicable to closed source developers!
In implementation terms, it takes some time to charge up the address bus, so you increase bandwidth and execution speed by charging up address n, but doing a quick read of n+1, n+2, n+3, and more on the latest CPUs. You only have to wiggle the two low-order address lines for the extra reads, so you don't pay the pre-charge penalty that you would for access randomly in memory.
This is incorrect. It has nothing to do with charging the address lines. Loading multiple sequential locations is slow on the first access and fast on the subsequent bytes because a whole memory row (made of multiple words) is read at once. This full memory row (typically around 1kbit) is transferred from the slower capacitive DRAM storage to faster transistor-based flip-flops. The subsequent sequential words are already available in the flip-flops so it's faster to route them off-chip since the slow DRAM access is avoided.
Re:Biggest PC annoyance...
on
PC Annoyances
·
· Score: 4, Funny
The most difficult thing in the world is to know
how to do a thing and to watch someone else doing
it wrong, without commenting.
-- T.H. White
This comment doesn't make sense. What you're saying violates the most basic law of economics for businesses: do not take sunk costs into account.
Anyway, your reasoning is clearly fallacious:
With all of these companies having geared up for LCD production and building huge plants while downsizing CRTs, it's unlikely that the shift to LED tech is going to happen anytime soon - no matter how cheap it is.
You mean that even if it cost $1 to build an LED plant, they would still stick with LCD plants just because they were already built? The reasons there are no LED display panels are technological and economic, but they have nothing to do with what you are talking about.
Self-interest is rarely the main driving force in our life. To accept the idea
that you are driven by self-interest is demeaning.... It's like admitting
that you are essentially a dog, whose major preoccupation is five o'clock, when
the little food pellets go into your bowl.
-- John Ralston Saul, "A Wondrous Uncertainty" in Queen's Quarterly, Spring 2002
I am quite sure there are some people out there who used Alpha-based workstations back when Digital made them.
In addition to the Alpha's, I have Sun Blade 100 at my desk at school, which is a desktop with a 64-bit processor (Ultrasparc II).
On the Apple G5 page it actually says: "The Power Mac G5 is the worldâ(TM)s fastest personal computer and the first with a 64-bit processor..." Note the phrase personal computer. I have to agree that the 64-bit Alpha's and Sun Ultrasparc machines are not personal computers.
I haven't seen this posted before. In a news.com article, IBM's alleged violations are listed:
Specifically, the transferred code includes the Journaled File System (JFS), extensions to make Linux work on a multiprocessor server employing the non-uniform memory access (NUMA) technique, Sontag said. In addition, he said read-copy update (RCU) for relieving some memory bottlenecks on multiprocessor servers, was transferred.
Slashdot Mirror
My apologies. You are indeed correct.
Low-k dielectrics are for the insulator between wires. High-k dielectrics are for the insulator between the gate and channel.
Some clarifications:
Short-circuit current is only responsible for 10-20% of switching power. The rest is dissipated in the transistor through charging and discharing all the nodal capacitances (due to transistor gates, transistor diffusions and wiring capacitance). Since typical circuit styles are non-adiabatic, this charge/discharge power component would not go away even if we could completely eliminate short-circuit currents.
Making transistors smaller certainly reduces their gate capacitance but it also reduces their current drive by the same proportion. These two effects cancel each other out! So how can transistors get faster from generation to generation?
Transistors get faster by increasing electron mobility and/or increasing gate capacitance per unit area and/or reducing diffusion junction/sidewall capacitance per unit area/perimiter and/or reducing (local) interconnect capacitance since smaller transistors are closer together.
It's easy to reduce the tunneling current through the gate. All you have to do is increase the thickness of the insulator. Unfortunately, this has the detrimental effect of reducing the effective capacitance of the gate, which in turn lowers the amount of current conducted by atransistor of a given size (lowering the current also lowers the speed). To make up for the lowered gate capacitance, researchers have been trying to increase the dielectric constant of the insulator. I'm guessing that they're proposing a method to increase the dielectric constant of the gate insulator. The devil is in the details of improving the dielectric constant without screwing up later processing steps or reducing the mechanical integrity of the wafer, etc.
Summary:
The computer industry, however, tends to move slowly when it comes to major overhauls of computer architecture.
Huh? The computer industry, and more specifically, the processor industry, is the fastest adopter of new technology in any endeavor in human history. Anyone have an example of a faster-moving industry?
Beware, the article is quite technical:
If you extrapolate our 21 quasars out to the rest of the sky, you get a whole lot of quasars.
From the article:
Darkmail is primarily used in distributed denial of service (DDoS) attacks and directory harvest attacks (DHA) in which a specific domain is hit with a flood of emails through an alphabetical list of names.
But over the last year darkmail is being used to brute-force spam through filters and is clogging up bandwidth.
Basically, it seems that darkmail is bulk mail sent to a domain with the advance knowledge that much of it will not reach a destination.
Zero gain bandwidth isn't really the technical term for it.
There are actually two common metrics for the max speed of the transistor: fT and fMAX. fT is the frequency at which the current gain of the transistor is equal to 1 (not zero---the confusion may arise from the fact that a gain of 1 equals a gain of 0 dB). fMAX is the frequency at which the power gain of the transistor is equal to 1. The fMAX is a much more useful and meaningful metric than fT but is extremely difficult to measure as it is very sensitive to the test conditions. Because of the difficulty of measuring fMAX, often fT is reported instead. It's not clear from the article which of these two frequencies they are reporting in this research.
Typically, there is a loss of speed when connecting multiple transistors, so the fastest circuits will operate at some fraction of fMAX. My guess is that Intel's chips probably operation at about 0.05 to 0.1 of fMAX. Anyone who knows the fMAX of of a typical 130nm or 90nm process care to confirm?
Don't feel sorry for Paul Graham. He chose his path and he enjoys it. You chose a different path and are happy with your choice. Good for both you. I don't feel sorry for either of you.
God, now I remember why I don't post here any more.
Now I remember why I don't respond to people who don't post anymore.
It takes a lot of work to learn how to program in Java correctly -- even more effort than would be required to learn Python I'd venture (based on my extensive experience...).
Seems like a good motivation to avoid Java. Often, part of being smart is also being somewhat lazy. Based on your description, I would much rather learn Python than Java
A good Java programmer has to be smart.
We are comparing relative intelligence here. Obviously, anyone who can learn ANY modern programming language is reasonably smart. Nevertheless, there are large ranges of intelligence even within the group of people smart enough to learn to program.
I can agree that I've encountered a whole lot of Java developers out there who write bad code. But I don't think the ratio is any different than in any other popular language.
I think Paul Graham would disagree that this ratio is the same for all languages. His experience and understanding of programming languages is deep (which lends credence to his thoughts, but obviously doesn't mean he is automatically correct). Is your experience and understanding deep enough to really justify your claim? Can you give the approximate ratio for 3 different programming languages?
I have a client that needs a server.
On a related note, I was having dinner at a restaurant and my waiter asked me for a recommendation for a good email program. So I guess it turns outs that I have a server that needs a client.
Is this the end of moores law, at least in the form of CPU speeds doubling every 18 months? There are essentially two CPUs, I doubt each of them will get 2x faster the next 1.5 years :)
There have been quite a few posts pointing out that Moore's law actually refers to exponential growth in transistor density rather than speed.
The posters are technically correct, but the term Moore's law has come to encompass any processor-related metric that changes at an exponential pace, including processor performance, clock rate, and power consumption. Of course, these metrics are directly related to transistor size and density, so it makes sense that they have changed exponentially.
For those with access to IEEE articles, Gordon Moore (Intel founder, who Moore's law is named after) wrote an interesting paper called No exponential is forever: but "Forever" can be delayed!.
Very true. To go off on a tangent...
I guess what confuses people is that there is a lot of research into trying to make two processor systems exactly twice as fast as a single processor system. Keep in mind this is for a single application. When a single application (or process) is spread across multiple CPUs, the CPUs often must communicate information to complete the computation. The cost of this cross-CPU communication is high. A single CPU system, however, has no communication bottleneck. Thus the performance of a 2-CPU system is often much less that twice that of the single CPU system--but always better than just one processor as noted by the parent posting.
But if we are running multiple, independent applications/processes, then there's little problem. There is no need for communication between unrelated processes. So there's no communication bottleneck. If you run an environment with computation-intensive applications that are indepedent, you will get twice the performance* with a 2-CPU system versus a single CPU system.
*I have simplified the situation. There is indeed overhead, but this is minor compared to parallelizing a single application/process.
Two processors are better than one, period....and in some cases, two processors are twice as good as single processor.
Sorry for adding to a sentence that ends with the word period. But I just can't help it, period.
A day doesn't go by when I don't talk to a Fortune 1000 customer ... that are not looking at dipping their feet into the Linux desktop.
If only I didn't have a nickel for every time someone didn't tell me the exact same thing, then I would definitely never be not rich, no?
First off, First Monday touts itself as a peer-reviewed journal. I have to say, that this is probably the shittiest article I have ever read in a peer-reviewed journal (related to serious technical material---most soft science journal are essentially worthless). First Monday just got added to my list of journals to not waste my time reading
These same 5 attributes exist in the closed source world. I'm sure many of you have used closed source software with poor UI's, documentation, feature-centric, etc. These problems are present in all software. Perhaps they are more prevalent in open source software, but that argument would have to be well-researched to be convincing, unlike the slop in this article.
Just do a "s/open source/closed source/g" on the article, and it still makes sense. Even the part about religious blindness is applicable to closed source developers!
This would seem to confirm Virus creators are sharing more code.
So, do they prefer GPL or BSD license?
In implementation terms, it takes some time to charge up the address bus, so you increase bandwidth and execution speed by charging up address n, but doing a quick read of n+1, n+2, n+3, and more on the latest CPUs. You only have to wiggle the two low-order address lines for the extra reads, so you don't pay the pre-charge penalty that you would for access randomly in memory.
This is incorrect. It has nothing to do with charging the address lines. Loading multiple sequential locations is slow on the first access and fast on the subsequent bytes because a whole memory row (made of multiple words) is read at once. This full memory row (typically around 1kbit) is transferred from the slower capacitive DRAM storage to faster transistor-based flip-flops. The subsequent sequential words are already available in the flip-flops so it's faster to route them off-chip since the slow DRAM access is avoided.
The most difficult thing in the world is to know
how to do a thing and to watch someone else doing
it wrong, without commenting.
-- T.H. White
This comment doesn't make sense. What you're saying violates the most basic law of economics for businesses: do not take sunk costs into account.
Anyway, your reasoning is clearly fallacious:
With all of these companies having geared up for LCD production and building huge plants while downsizing CRTs, it's unlikely that the shift to LED tech is going to happen anytime soon - no matter how cheap it is.
You mean that even if it cost $1 to build an LED plant, they would still stick with LCD plants just because they were already built? The reasons there are no LED display panels are technological and economic, but they have nothing to do with what you are talking about.
A pox on the moderators who modded up the parent!
Self-interest is rarely the main driving force in our life. To accept the idea that you are driven by self-interest is demeaning. ... It's like admitting
that you are essentially a dog, whose major preoccupation is five o'clock, when
the little food pellets go into your bowl.
-- John Ralston Saul, "A Wondrous Uncertainty" in Queen's Quarterly, Spring 2002
From the article:
"Never argue against something on behalf of moths," he warns. "People will just laugh at you. Talk about ecosystems instead."
Point well taken ;)
FYI, TIA = Total Information Awareness
"world's first 64-bit desktop processor"
I am quite sure there are some people out there who used Alpha-based workstations back when Digital made them.
In addition to the Alpha's, I have Sun Blade 100 at my desk at school, which is a desktop with a 64-bit processor (Ultrasparc II).
On the Apple G5 page it actually says: "The Power Mac G5 is the worldâ(TM)s fastest personal computer and the first with a 64-bit processor..." Note the phrase personal computer. I have to agree that the 64-bit Alpha's and Sun Ultrasparc machines are not personal computers.
I haven't seen this posted before. In a news.com article, IBM's alleged violations are listed:
Specifically, the transferred code includes the Journaled File System (JFS), extensions to make Linux work on a multiprocessor server employing the non-uniform memory access (NUMA) technique, Sontag said. In addition, he said read-copy update (RCU) for relieving some memory bottlenecks on multiprocessor servers, was transferred.