For most server needs, take a look at the Supermicro cases. I own the now-classic SC-750 and have no complaints about it.
Yeah, I know they're just OEMs, the SC-750 is actually made by some other company (which I can't recall now), but anyway if it was picked by Supermicro then it must be very high quality.
Also, Supermicro always bundles high-quality power supplies, an increasingly important point with the powerhogs that are Dual Athlons and Xeons. In my case, the Sparkle FSP300-60GT -- hasn't failed yet (and the load's high on it, believe me, plus it's 24/7), and should be enough power for most needs. Unless you're building the ultimate peltier/watercooling rig.
While we're at it, how about death penalty for Flash ``developers''? I know it's totally unrelated, but these Flash intros are tens of times more annoying than web page defacements. Thank God Konqueror doesn't come with the plugin installed.
I'm concerned that widespread adoption of crypto is being held back by the lack of education on the subject.
I've had in mind for some time the idea of developing secure IRC-like chat software. It would mostly take advantage of the authentication and non-repudiation properties of public key cryptosystems, with privacy coming right behind. Most likely such a system would fail, since ordinary people don't take into account even the most basic principles of key management. How can you trust someone based solely on crypto authenticaton, while they stick their (already weak) password in a Post-it note on their monitor?
Is a crypto-aware society utopic? One whose people would protect their secret keys with their life? Technology enables us to do great things, but social aspects can hold it back.
First, I offer my support for your actions. Such acts as the development of PGP are not understood but many generations later. Surely you will go down in history alongside the greatest heroes of mankind.
Most of us are aware of the ethical issues behind backdoor'd crypto, and why they won't solve the problem (and possibly hurt many legitimate uses of strong crypto). Being a technical audience, I'm sure the Slashdot population would like to know how it could be implemented, and whether such backdoors would weaken the algorithms (from a theoretical standpoint).
As an update to my previous post, here are some stats on my newly finished 1 gigabit Pi run:
Entropy = 7.999998 bits per byte.
Optimum compression would reduce the size
of this 134217728 byte file by 0 percent.
Chi square distribution for 134217728 samples is 306.26, and randomly
would exceed this value 2.50 percent of the times.
Arithmetic mean value of data bytes is 127.5018 (127.5 = random).
Monte Carlo value for Pi is 3.141822921 (error 0.01 percent).
Serial correlation coefficient is -0.000185 (totally uncorrelated = 0.0).
Look at the original post for stats on the first 512 megabits of Pi.
Also, I had made slight modifications to the ent program (used to generate the stats above), in order to treat the input as a stream of 16-bit values, not 8-bit as done in the above stats. Here are the 512 Mb and 1 Gb stats output by this modified version:
Entropy = 15.998588 bits per short.
Optimum compression would reduce the size
of this 33554432 short file by 0 percent.
Chi square distribution for 33554432 samples is 65664.47, and randomly
would exceed this value 50.00 percent of the times.
Arithmetic mean value of data shorts is 32765.0013 (32767.5 = random).
Monte Carlo value for Pi is 3.142281362 (error 0.02 percent).
Serial correlation coefficient is 0.000000 (totally uncorrelated = 0.0).
Entropy = 15.999293 bits per short.
Optimum compression would reduce the size
of this 67108864 short file by 0 percent.
Chi square distribution for 67108864 samples is 65745.70, and randomly
would exceed this value 50.00 percent of the times.
Arithmetic mean value of data shorts is 32767.4050 (32767.5 = random).
Monte Carlo value for Pi is 3.141822921 (error 0.01 percent).
Serial correlation coefficient is 0.000000 (totally uncorrelated = 0.0).
According to Wolfram Research's Mathematica, Log[x] is base e, while Log[10, x] is base 10. Of course, your TI-89 knows better, right? Perhaps you can take the time to ask a mathematician -- a real mathematician, not a high school teacher.
Mathematicians usually write logarithms in the base e as simply log. While it is usually taught that log assumes base 10, the base-e logarithm is much more important in math than any other logarithm. If you had been in a calculus course, you'd know that.
Develop a file-sharing network where data is encrypted using PGP's style of encryption.
Perfectly possible. You don't even need to worry about how it'd happen -- just use a cryptographic library. There's something along these lines using GPG (IIRC, it's called GPGme.) If properly implemented in C++, you could seamlessly switch between secure and unsecure mode, without having to rewrite any functions! (Of course, once we start stripping off any application-specific aspects, we'd eventually arrive at some sort of Secure Sockets Layer, with public-key emphasis and tight integration with the application.)
Cryptography is a great field. It enables never-thought-before stuff in the fields of privacy, authentication, data integrity and non-repudiation. SSH and PGP are but small examples of what can be done.
That's why we need to develop cryptographic awareness in the general public. If everyone was informed about the benefits of cryptography, there'd be pressure against export restrictions and the like. Even more important is the question of key management -- it might be worth implementing many of the marvelous ideas enabled by cryptography, if only people managed to understand how important a private key is, and started seriously protecting them.
Sorry, there's no such thing as proof in statistics. However, I'm sure the owner of a cryptographic key, which happened to be generated from a non-random stream of Pi, wouldn't care about losing his private key in exchange for a discovery of this magnitude.
There's nothing about transcendental numbers that makes their digits random; pi and e happen to be special. For instance, the first number proven to be transcendental was (some variation on) 0.101001000100001000001...; that sequence isn't random by any definition.
I should have put it more clearly -- there's nothing special about transcendental numbers in general, indeed. It's just that I've analyzed two of them, Pi and e, and both have uniform statistical distribution and so forth. However, each new addition to this hypothetical library would have to go through the same extensive testing I've applied here, being transcendental or irrational or whatever.
I'd recommend to anyone interested in projects like this to look at George Marsaglia's page; his tester may help you avoid releasing crap.
I have done some testing with DIEHARD as well -- all randomness analyzers seem to back my ideas (:
I had this feeling as well. That's why my initial library project had two requirements:
1. small seeds (perhaps some 256 bits would do) of a truly random source.
2. a way of computing square roots and logarithms to arbitrary precision.
The idea was to XOR at least two irrational and/or transcendental numbers, such as square roots (of non-squared numbers, of course!), pi, e, logarithms, etc. It'd be necessary to use some of the random bits to choose which of the numbers will be calculated (pick from one of the four choices above, or others that might be implemented). Then, should we choose a square root or logarithm, we'd need more random bits to pick whether it'd be the square root/logarithm of 32857 or 18764874, say. The rest of the random bits would be used to pick a starting bit in the calculation stream, inside each of the calculated numbers. At last, all streams would be XORed, and that would be the output.
Although it seems like a lot of calculation, remember that computers are getting more powerful everyday, and calculating hundreds of thousands of digits takes a second or so in gigahertz computers; and a truly random source of data (such as timing the decay of radioactive atoms) is slow at generating large streams, but can perfectly cope with the small seeds we need for this library.
I can't find an expansion for Pi, but for the golden ratio (1.6180339...) they go like this:
Phi = 1+1/(1+1/(1+1/(1+1/(1+1/(1+...)))))
What's interesting, Phi can be wrote as.5*(sqrt(5) + 1). Number theory is fascinating. But I'm straying out of my way now. Try to write the fraction on paper, and you'll understand what continued fractions are.
I did some very interesting work this year with this, in the course of planning a high-quality pseudo-random library.
I calculated the first 512 megabits of Pi, and then started splitting up the file in smaller pieces, to study whether they had an "information-theoretic randomness quality". That is, they're not random (you can calculate them), but they exhibit desirable randomness properties, such as uniform statistical distribution.
Here's the output of John Walker's ent program for 512 megabits of Pi:
Entropy = 7.999997 bits per byte.
Optimum compression would reduce the size
of this 67108864 byte file by 0 percent.
Chi square distribution for 67108864 samples is 245.38, and randomly
would exceed this value 50.00 percent of the times.
Arithmetic mean value of data bytes is 127.4938 (127.5 = random).
Monte Carlo value for Pi is 3.142281720 (error 0.02 percent).
Serial correlation coefficient is -0.000145 (totally uncorrelated = 0.0).
For the entropy test, a completely random sample would have an entropy of 8.0 bits per byte, and the ideal Chi Square distribution would be 256.0 (considering there are 256 degrees of freedom in an 8-bit data structure, or 2**8 possibilities.) As you can see, that's about as random as you can get. And the larger the samples you feed it, the more it converges to the ideal values.
I've also done some testing with other transcendental numbers, such as e (2.718281828...), and they all seem to show great randomness properties, in the information-theoretic sense at least. However, I have a feeling to "trust" Pi more than e, given that you can write e in form of continued fractions with repeating patterns, and nobody has yet found a pattern in the continued fractions of Pi.
As for my pseudo-random library project, my programming skills are quite bad, but if you have some knowledge of scientific computing (multiplication algorithms using FFTs, for example), you can contact me and I might revive the idea.
You expect a viable platform with Brookdale? The damn thing's much slower than the i850. And as you said in the beginning, RDRAM is acceptable now. As I see it, only DDR might possibly compete with RDRAM.
Brookdale was already in Intel's roadmaps for a long while. They're just waiting for their contract with Rambus to end, since it's preventing them from releasing DDR products.
Plus, they designed the P4 from the ground up to support RDRAM. It'd be just stupid to quit supporting them now, that it's becoming almost affordable to buy Rambus memory.
Personally, I loathe Rambus as much as the next guy, but if you think rationally, this decision should have been taken by Intel a long time ago -- now it doesn't make sense anymore.
While it is hard to picture how to make it useful in this context, with the software available today, it is possible to ascertain the creator and time of creation of any piece of data. How? Public-key encryption and signing techniques. Just keep your private key is secure, and find a reliable (as in: never cracked by script kiddies) timestamp server. Then, sign your message using your private key (no encryption), hash the signed message, send it to the timestamp server for signing, then append the server's signature to your message. There: you can be certain of the person who signed this text (just check it with the person's public key), plus the time the message was sent to the timestamp server (just check it the server's public key.) A byproduct of signing is that the message cannot be tampered with. If you change its contents, both your signature and the timestamp server's signature will be invalid, and while you can fake your signature again, you can't have the timestamp server sign it again without showing it was done in a later time.
That brings an interesting idea: you could have all your webserver's logs signed and timestamped periodically. If you can trust the administrator of the machine, then those logs would be safe from tampering.
While hardly on topic, would anyone care to tell me why Carl Sagan, in Chapter 13 of his book "The Demon-Haunted World", labels cryptography as pseudo-science? Hopefully the Portuguese translation of the book is incorrect. I really appreciate the work of this guy, but I can't picture him labeling a field of mathematics as pseudo-science.
In general, any cryptographic algorithm patent is a good patent. It takes a lot of research to come up with these algorithms.
I really couldn't come up with anything else, but in general, you can spot a valid patent by excluding all others that don't meet these requirements:
-it must be non-obvious;
-it must be valid by itself and not only for being something done on a computer or the Web;
-there is no prior art.
Requirements #1 and #3 require some expertise from the patent issuers, and the USPTO is clearly lacking such expertise (although for some patents even common sense would do, but they seem to lack it as well.) Requirement #2 is a mere consequence of the other two.
We need some sort of peer review of patents. Before being issued, it must go public for people to comment on it, clue the examiners on whether there is prior art, or even if a previous patent exists, but it didn't contain the word "Web." And a paradigm shift would be nice as well: instead of granting all patents except a few, grant only a patent once the author proves it deserves it, kinda like when you defend a Ph.D. thesis. If I wasn't clear, think of it as whether you are guilty until you prove yourself innocent, or innocent until you are proven guilty.
Here in Brazil, speed limitation is commonplace. The local ISP, which provides service for a few other towns (summing up, some 5 million inhabitants), has a 34 Mbps Internet link (as I've heard, American ISPs have multiple-T3 lines.)
They've just started providing ADSL service. Under optimal conditions, they measured up to 9 Mbps downstream, but only offer service plans up to 2 Mbps (that means they artifically cap bandwidth according to each user's service plan). All bandwidth figures quoted are downstream only, upstream ratio is half of that. Prices are ca. US$ 40 for 256 kbps, and US$ 170 for 512 kbps. These are home-user service plans, which means you're leased an IP (using DHCP) for 1 hour, then it gets changed. If you want fixed IPs and hosting privileges, you'll pay US$ 600 for 256 kbps, US$ 1,200 for 512 kbps, US$ 2,200 for 1 Mbps and US$ 4,100 for 2 Mbps.
There are times when I just wish I lived in the US. Cheaper hardware, faster connections...
When people speak of the 1-click patent, they may not completely grasp how generic and broad it is. It's not only about a method of buying something with a single click -- it's about obvious use of cookies and certain other techniques, for which prior art most certainly exists.
Whenever such patents are issued, the public blames the very idea of patent, instead of the real target, obviously the patent office, a.k.a. the guys short of common sense.
A so-called patent reform doesn't need to change the laws regarding patents. They're fine. The current concept of patents, when correctly applied, should bring all the benefits usually associated to it (that applies to software as well; for instance, look at the RSA patent.) The public must realize the ones to blame for the state of patents today are: those who approve applications that don't fit in the definition of patent, and judges who enforce them -- as I pointed out, people in desperate need of common sense.
As a mind experiment, imagine a world stripped of all stupid patents (not restricted to software); would you still complain about the current patent law?
What else can be done with this serial cable? Around here, they're selling it for US$ 100, but if the cable's only use is to sync phone numbers, I won't get it. Can I do EEPROM dumps/reflashs with it, whether using Linux or Windows software?
The news report is a bit biased towards Intel. What is this with "AMD is keeping the crown in the 'Value' category"? AMD never lost it. If you look carefully at the benchmarks, you can grab a Duron 600 and it'll still beat whatever Celeron you throw at it, 100 MHz bus or not, in most of the benchmarks. If you're shopping for a value CPU with an external graphics card, the Duron is the only rational choice. Plus, if you grab a decent motherboard, most of the chips can be overclocked to 30-50% of their stock frequency. You just can't go wrong.
Slashdot Mirror
Yeah, I know they're just OEMs, the SC-750 is actually made by some other company (which I can't recall now), but anyway if it was picked by Supermicro then it must be very high quality.
Also, Supermicro always bundles high-quality power supplies, an increasingly important point with the powerhogs that are Dual Athlons and Xeons. In my case, the Sparkle FSP300-60GT -- hasn't failed yet (and the load's high on it, believe me, plus it's 24/7), and should be enough power for most needs. Unless you're building the ultimate peltier/watercooling rig.
While we're at it, how about death penalty for Flash ``developers''? I know it's totally unrelated, but these Flash intros are tens of times more annoying than web page defacements. Thank God Konqueror doesn't come with the plugin installed.
Phil,
I'm concerned that widespread adoption of crypto is being held back by the lack of education on the subject.
I've had in mind for some time the idea of developing secure IRC-like chat software. It would mostly take advantage of the authentication and non-repudiation properties of public key cryptosystems, with privacy coming right behind. Most likely such a system would fail, since ordinary people don't take into account even the most basic principles of key management. How can you trust someone based solely on crypto authenticaton, while they stick their (already weak) password in a Post-it note on their monitor?
Is a crypto-aware society utopic? One whose people would protect their secret keys with their life? Technology enables us to do great things, but social aspects can hold it back.
Phil,
First, I offer my support for your actions. Such acts as the development of PGP are not understood but many generations later. Surely you will go down in history alongside the greatest heroes of mankind.
Most of us are aware of the ethical issues behind backdoor'd crypto, and why they won't solve the problem (and possibly hurt many legitimate uses of strong crypto). Being a technical audience, I'm sure the Slashdot population would like to know how it could be implemented, and whether such backdoors would weaken the algorithms (from a theoretical standpoint).
Thanks.
Look at the original post for stats on the first 512 megabits of Pi.
Also, I had made slight modifications to the ent program (used to generate the stats above), in order to treat the input as a stream of 16-bit values, not 8-bit as done in the above stats. Here are the 512 Mb and 1 Gb stats output by this modified version:
According to Wolfram Research's Mathematica, Log[x] is base e, while Log[10, x] is base 10. Of course, your TI-89 knows better, right? Perhaps you can take the time to ask a mathematician -- a real mathematician, not a high school teacher.
Mathematicians usually write logarithms in the base e as simply log. While it is usually taught that log assumes base 10, the base-e logarithm is much more important in math than any other logarithm. If you had been in a calculus course, you'd know that.
Perfectly possible. You don't even need to worry about how it'd happen -- just use a cryptographic library. There's something along these lines using GPG (IIRC, it's called GPGme.) If properly implemented in C++, you could seamlessly switch between secure and unsecure mode, without having to rewrite any functions! (Of course, once we start stripping off any application-specific aspects, we'd eventually arrive at some sort of Secure Sockets Layer, with public-key emphasis and tight integration with the application.)
Cryptography is a great field. It enables never-thought-before stuff in the fields of privacy, authentication, data integrity and non-repudiation. SSH and PGP are but small examples of what can be done.
That's why we need to develop cryptographic awareness in the general public. If everyone was informed about the benefits of cryptography, there'd be pressure against export restrictions and the like. Even more important is the question of key management -- it might be worth implementing many of the marvelous ideas enabled by cryptography, if only people managed to understand how important a private key is, and started seriously protecting them.
Also, there are other uses for random numbers outside of generating cryptographic keys; for example, Monte Carlo simulations.
Sorry, there's no such thing as proof in statistics. However, I'm sure the owner of a cryptographic key, which happened to be generated from a non-random stream of Pi, wouldn't care about losing his private key in exchange for a discovery of this magnitude.
I should have put it more clearly -- there's nothing special about transcendental numbers in general, indeed. It's just that I've analyzed two of them, Pi and e, and both have uniform statistical distribution and so forth. However, each new addition to this hypothetical library would have to go through the same extensive testing I've applied here, being transcendental or irrational or whatever.
I'd recommend to anyone interested in projects like this to look at George Marsaglia's page; his tester may help you avoid releasing crap.
I have done some testing with DIEHARD as well -- all randomness analyzers seem to back my ideas (:
What John Walker's program refers to as optimum compression is Huffman coding, IIRC.
I had this feeling as well. That's why my initial library project had two requirements:
1. small seeds (perhaps some 256 bits would do) of a truly random source.
2. a way of computing square roots and logarithms to arbitrary precision.
The idea was to XOR at least two irrational and/or transcendental numbers, such as square roots (of non-squared numbers, of course!), pi, e, logarithms, etc. It'd be necessary to use some of the random bits to choose which of the numbers will be calculated (pick from one of the four choices above, or others that might be implemented). Then, should we choose a square root or logarithm, we'd need more random bits to pick whether it'd be the square root/logarithm of 32857 or 18764874, say. The rest of the random bits would be used to pick a starting bit in the calculation stream, inside each of the calculated numbers. At last, all streams would be XORed, and that would be the output.
Although it seems like a lot of calculation, remember that computers are getting more powerful everyday, and calculating hundreds of thousands of digits takes a second or so in gigahertz computers; and a truly random source of data (such as timing the decay of radioactive atoms) is slow at generating large streams, but can perfectly cope with the small seeds we need for this library.
Sorry, that's not a continued fraction.
.5*(sqrt(5) + 1). Number theory is fascinating. But I'm straying out of my way now. Try to write the fraction on paper, and you'll understand what continued fractions are.
I can't find an expansion for Pi, but for the golden ratio (1.6180339...) they go like this:
Phi = 1+1/(1+1/(1+1/(1+1/(1+1/(1+...)))))
What's interesting, Phi can be wrote as
Here's the output of John Walker's ent program for 512 megabits of Pi:
For the entropy test, a completely random sample would have an entropy of 8.0 bits per byte, and the ideal Chi Square distribution would be 256.0 (considering there are 256 degrees of freedom in an 8-bit data structure, or 2**8 possibilities.) As you can see, that's about as random as you can get. And the larger the samples you feed it, the more it converges to the ideal values.I've also done some testing with other transcendental numbers, such as e (2.718281828...), and they all seem to show great randomness properties, in the information-theoretic sense at least. However, I have a feeling to "trust" Pi more than e, given that you can write e in form of continued fractions with repeating patterns, and nobody has yet found a pattern in the continued fractions of Pi.
As for my pseudo-random library project, my programming skills are quite bad, but if you have some knowledge of scientific computing (multiplication algorithms using FFTs, for example), you can contact me and I might revive the idea.
You expect a viable platform with Brookdale? The damn thing's much slower than the i850. And as you said in the beginning, RDRAM is acceptable now. As I see it, only DDR might possibly compete with RDRAM.
Brookdale was already in Intel's roadmaps for a long while. They're just waiting for their contract with Rambus to end, since it's preventing them from releasing DDR products.
Plus, they designed the P4 from the ground up to support RDRAM. It'd be just stupid to quit supporting them now, that it's becoming almost affordable to buy Rambus memory.
Personally, I loathe Rambus as much as the next guy, but if you think rationally, this decision should have been taken by Intel a long time ago -- now it doesn't make sense anymore.
One or two tiny changes? You mean covering the new stuff in kernel 2.4, and updating other sections in the book for 2.4 compliance, is a tiny change?
I'll buy this book, and even if I owned the first edition I'd still buy it.
While it is hard to picture how to make it useful in this context, with the software available today, it is possible to ascertain the creator and time of creation of any piece of data. How? Public-key encryption and signing techniques. Just keep your private key is secure, and find a reliable (as in: never cracked by script kiddies) timestamp server. Then, sign your message using your private key (no encryption), hash the signed message, send it to the timestamp server for signing, then append the server's signature to your message. There: you can be certain of the person who signed this text (just check it with the person's public key), plus the time the message was sent to the timestamp server (just check it the server's public key.) A byproduct of signing is that the message cannot be tampered with. If you change its contents, both your signature and the timestamp server's signature will be invalid, and while you can fake your signature again, you can't have the timestamp server sign it again without showing it was done in a later time.
That brings an interesting idea: you could have all your webserver's logs signed and timestamped periodically. If you can trust the administrator of the machine, then those logs would be safe from tampering.
While hardly on topic, would anyone care to tell me why Carl Sagan, in Chapter 13 of his book "The Demon-Haunted World", labels cryptography as pseudo-science? Hopefully the Portuguese translation of the book is incorrect. I really appreciate the work of this guy, but I can't picture him labeling a field of mathematics as pseudo-science.
In general, any cryptographic algorithm patent is a good patent. It takes a lot of research to come up with these algorithms.
I really couldn't come up with anything else, but in general, you can spot a valid patent by excluding all others that don't meet these requirements:
-it must be non-obvious;
-it must be valid by itself and not only for being something done on a computer or the Web;
-there is no prior art.
Requirements #1 and #3 require some expertise from the patent issuers, and the USPTO is clearly lacking such expertise (although for some patents even common sense would do, but they seem to lack it as well.) Requirement #2 is a mere consequence of the other two.
We need some sort of peer review of patents. Before being issued, it must go public for people to comment on it, clue the examiners on whether there is prior art, or even if a previous patent exists, but it didn't contain the word "Web." And a paradigm shift would be nice as well: instead of granting all patents except a few, grant only a patent once the author proves it deserves it, kinda like when you defend a Ph.D. thesis. If I wasn't clear, think of it as whether you are guilty until you prove yourself innocent, or innocent until you are proven guilty.
Here in Brazil, speed limitation is commonplace. The local ISP, which provides service for a few other towns (summing up, some 5 million inhabitants), has a 34 Mbps Internet link (as I've heard, American ISPs have multiple-T3 lines.)
They've just started providing ADSL service. Under optimal conditions, they measured up to 9 Mbps downstream, but only offer service plans up to 2 Mbps (that means they artifically cap bandwidth according to each user's service plan). All bandwidth figures quoted are downstream only, upstream ratio is half of that. Prices are ca. US$ 40 for 256 kbps, and US$ 170 for 512 kbps. These are home-user service plans, which means you're leased an IP (using DHCP) for 1 hour, then it gets changed. If you want fixed IPs and hosting privileges, you'll pay US$ 600 for 256 kbps, US$ 1,200 for 512 kbps, US$ 2,200 for 1 Mbps and US$ 4,100 for 2 Mbps.
There are times when I just wish I lived in the US. Cheaper hardware, faster connections...
When people speak of the 1-click patent, they may not completely grasp how generic and broad it is. It's not only about a method of buying something with a single click -- it's about obvious use of cookies and certain other techniques, for which prior art most certainly exists.
Whenever such patents are issued, the public blames the very idea of patent, instead of the real target, obviously the patent office, a.k.a. the guys short of common sense.
A so-called patent reform doesn't need to change the laws regarding patents. They're fine. The current concept of patents, when correctly applied, should bring all the benefits usually associated to it (that applies to software as well; for instance, look at the RSA patent.) The public must realize the ones to blame for the state of patents today are: those who approve applications that don't fit in the definition of patent, and judges who enforce them -- as I pointed out, people in desperate need of common sense.
As a mind experiment, imagine a world stripped of all stupid patents (not restricted to software); would you still complain about the current patent law?
What else can be done with this serial cable? Around here, they're selling it for US$ 100, but if the cable's only use is to sync phone numbers, I won't get it. Can I do EEPROM dumps/reflashs with it, whether using Linux or Windows software?
The news report is a bit biased towards Intel. What is this with "AMD is keeping the crown in the 'Value' category"? AMD never lost it. If you look carefully at the benchmarks, you can grab a Duron 600 and it'll still beat whatever Celeron you throw at it, 100 MHz bus or not, in most of the benchmarks. If you're shopping for a value CPU with an external graphics card, the Duron is the only rational choice. Plus, if you grab a decent motherboard, most of the chips can be overclocked to 30-50% of their stock frequency. You just can't go wrong.