And the SS10 didn't have the ZX framebuffer of the SS20 - 24 bits accelerated graphics. 2nd one available as an option. That was an awesome machine, still one of the greatest computer design ever.
I started using UNIX with SunOS 4.1.4 on SPARCstation 1+. Still have my SS2 somewhere in the attic. Still have my Ultra 1 Creator. Can't throw them away.
I like the speed of my W3680 - but it's just not the same.
Cue this old joke...
- How many hardware engineers does it take to change a light bulb?
- None, we'll fix it in software.
Doing stuff in software to make hardware easier has been tried before (and before this kid was born, perhaps why he thinks this is new). It failed. Transputer, i960, i432, Itanium, MTA, Cell, a slew of others I don't remember...
As for the grid, nice, but not exactly new. Tilera, Adapteva, KalRay,...
Explain to them how to fix the problem once and for all:
1) block outgoing port 25 to everything but their own mail servers; 2a) add an optional feature in each customer account to reopen outgoing port 25; 2b) add an optional feature in each customer account to pick the reverse DNS entry; 3) tell every other ISP/mail servers operators what they have just done, so they get un-blacklisted since they won't be sending much spam any more.
This should block most of the outgoing spam without any side-effects, since power users will still be able to operate their own mail servers, complete with reverse FQDN. Non-power users won't notice a thing.
Also, they will save money on bandwidth to the outside world.
That's what my (strictly residential) ISP has been doing for almost a decade. Works perfectly well for everyone involved.
http://www.adapteva.com/epipha... 64 cores, mesh network that extends off the chip, in production.</p><p>Try harder MIT:-p</p></quote>
They already tried harder : http://www.tilera.com/. And as another post mentioned, Intel Knights Corner is cache coherent on 61 cores (62 architectured).
The summary doesn't get the point of the article: what's novel is not the presence of cache coherency, it's just the new way of implementing snoop-based cache coherency over their network. Cache coherency for a large number of cores can be very expensive time-wise, so any idea to improve it is more than wecome.
Why is everybody thinking this is big news? ftp://download.nvidia.com/CUDAOpen64/ The previous compiler, based upon Open64, has been available in source form since CUDA 1.0. They (partially) switched to LLVM in 4.1, and they also release the source code. They didn't have to, because unlike Open64 LLVM is not GPL, so it's nice of them, but it's not exactly earth-shattering news...
The Library of Congress used to have a goal of including complete hard copies, at least for items of US origin and 'good grade' (that is, they aimed to have copies of things such as hardback books that were intended to last, more than, say, ephemera such as the pulp magazines). However, that goal has become an obvious impossibility due to sheer volume. After about 1960, the library began being more selective.
And the situation is infinitely worse for other medias. Not only aren't people trying to preserve them, in many case they have been actively destroyed, in particular television broadcasts.
If I buy a Chablis or a Burgundy I want a particular type of wine. So what that these wines originated in certain regions in France?
They didn't "originate". If it's a burgundy, then it hast to come from the region of Burgundy. It's that simple.
Also, for the record: if you buy a Chablis, you also buy a Burgundy. Chablis is a sub-region of Burgundy.
I don't give a damn where it was made. I would say most people who drink them don't know or care either.
Some of us haven't ruined their taste buds with bad beers and ketchup sauce, so we do care. Where the wine was produced makes a lot of difference to the taste. If you can't tell the difference, please go back to drinking Budweiser.
I'm told by a French friend who is a wine buff that the Aussie wines he can buy are superior to French wines (seriously), so this makes the whole thing sound like a ploy to recapture an ailing market.
There is no such thing as "superior", either way. There is such as thing as "different". Then it's a matter of taste. Australia, California, Chile, Algeria all make very good wines. They just aren't Burgundy, or Champagne.
Would you expect a "Scotch Whisky" to come from Polland? Obviously no. It doesn't preclude Japanese to make great Single Malt Whiskies. They just don't make Scotch Whiskies.
Think of it as a trademark, shared by all the producers from one geographic region. You can't buy a Macintosh from Hewlett-Packard, can you? So why should you be able to buy a Burgundy from someone that isn't located in the region of Burgundy, and therefore doesn't share in the trademark?
And how many MB can you address with a 32-bit pointer under the IEEE recommendations?
1) I don't know...
2) It doesn't [censored] matter!
3) It's exactly 4 GiB, or 4096 MiB, how hard is that?
4) I don't use 32 bits pointer anymore anyway.
See, it's very easy: just add a little 'i' in there, and it works exactly like before, just unambiguously.
Over 300 posts and counting, and all because people can't type 'i' to make sure there's no possible mistake... The world is doomed, I tell you, doomed!:-)
(sigh) No. The maximum available memory was measured in Megabytes, then Gigabytes. It was always base 2.
Well, not at first, 'cause a megabyte (of whatever size) of memory didn't exist yet.
Second, it was wrong but easy. It's still wrong and for memory, it's still easy. That's what it's still in use for memory, and is being ditched from everywhere else.
The main problem was that the binary prefix came too late. Old habits die hard. But as AA proves, it's possible to ditch a bad habit if you really want to. It's a matter of willpower. Yes we can, if I may so bold as to say so.
(for the record, I have been involve in computing since before Bill Gates informed us that 640 Kilobytes should be more than anyone will ever need, so I actually was there as all this unfolded)
And the fact that you're the new kid on the block is important because...?
Because our computers, almost since their earliest inception, work in base-2 arithmetic.
And this matters because...?
The SI prefix are used to denotes quantity. Except for the total size of semi-conductor memory (and sub-elements thereof), those quantities are usually completely unrelated to power of 2.
Disk sizes are not a power of 2. Files stored on them are even more arbitrary in size. Same for memory requirements of various codes. Heck, once upon a time, HPC programs would overallocate arrays to avoid power of 2 allocation (multiple of the page size wrecked havoc on direct-mapped caches)!
Frequency are not power of 2. Your 3 GHz processors runs at 3*1000^3 Hz (well, probably not very precisely:-), not 3.22*1000^3 HZ. Same goes for the (directly related) bandwidth. Latency doesn't even come close.
Heck, these days, even memory buses are moving away from strict power of 2: GTX 275 have a 448 bits-wide memory bus, Core i7's can be described as 192 bits.
And for the few cases where it matters (usually not end-user visible) that's what the freaking/frelling/rutting/smegging/[pick you favorite SciFi show euphemism] binary units are for !
(I'm not going to try to make sense from the rest of your post, because I can't see how the number of bits in a bytes is related to the discussion in any way).
Uh.. the inch is technically an SI unit. It is defined as exactly 2.54 cm.
No, it's not. SI uses the metre for length measurement, and nothing else. You can alter it with the various prefixes, and there's is only one thousand meters in a kilometre, not twenty-four more.
The "inch" from the United States customary units is defined as 2.54 centimetres, but it doesn't make it part of the SI..
That's just plain wrong. If my laptop has 3 Gigabytes of RAM in it, there had sure as hell better be 2^30 individually addressable locations. If there are only 10^9 addresses, they ripped me of by ((2^30) - (10^9))
(sigh).
Your laptop doesn't have 3 Gigabytes. It has 3 Gibibytes. That's 3221225472 bytes, or a bit more than 3.22 Gigabytes. Think of it that way: you got 7.37% more memory than you paid for. You ripped them off! Happy now?
The fact that both the salesdrones and the buyers are ignorant doesn't change the definition of "Giga" and "Gibi".
So far, only the hard drive manufacturers have catched up. When will the "3.22 gigabytes of memory!" advertisements show up?
The lasting ambiguity for hard drives has perhaps been less a matter of computer science than one of marketing. (The pervasiveness of inch measurements is a heavy hint at uninterest in SI.)
Actually, people who use the metric system outweigh people using imperial by a wide margin. In my entourage, nobody except for geeks knows how long an inch is. As you can guess, I don't live in the US.
As for the marketing: yes, marketdroids will do anything to sell. We all know that. But when "anything" becomes "make use of an established international standard", how can we protest?
And C.S. is at fault there. Why did it hijack prefixes to make them mean something else? It was stupid from the beginning. For Kilo I can understand, the error is small ; but from Mega onward, it was just plain laziness. That's why I try to use binary prefixes these days. Atonement:-)
Obligatory old timer anecdote: remember those "1.2MB" and "1.44MB" floppy disk? Worst of them all: the megabytes here were neither 1024^2 nor 10^6, but 1024*10^3, as in 1200 KiB and 1440 KiB.
The SI prefixes have been around for nearly 5 decades, and have a specific meaning used by everybody. Every scientist uses them in one way or another, and for every last one of of them, they have the same meaning.
Why can't we, the C.S. people, accept that?
Giga is 10^9. It has been 10^9 since it was created. It was never, ever meant to be anything but 10^9.
If you want to talk about 1024^3, then it's Gibi. Gibi is 2^30 since it was created. It was never, ever meant to be anything but 2^30.
Get over it.
(and yes, I try to always use GiB whenever it's appropriate).
I originally thought this was breaking 18 U.S.C. Chapter 119, 2510 to 2522 (?), but no.
*IF* they only alter the answer of their own DNS servers to their clients, when the client has made a request to said DNS servers, then they're probably in the clear. There is two communications: one from the client (C) to the Bell server (B), then one from B to the authoritative server (S). S then answer NXDOMAIN to B, which then returns a completely different information to C. So they're not intercepting anything.
OTOH, *IF* they hijack all the port 53 requests to the outside world (which I doubt), then it's very likely 2511(1)(a) and (d) applies. They still could argue under 2511(2)(a)(i) that it's "necessary"...
Then again, IANAL.
OTOH, even if it's legal, it's still absolutely wrong.
First there's a typo ; the basic ML555 is $2200, not $1200.
Someone managed to synthesize a S1 (single-core T1, with multithreads support) for a Virtex 4 LX60. You might be able to squeeze a S1 into the Virtex 5 LX50T of the ML555, but I'm not certain, it mght be a bit short on available slices. Beside, it seems that they didn't go any further than that, and didn't actually try to upload the bitsream, let alone boot an OS. All that is assuming either the synthesized HDL included all required bits beyond the CPU to do so, such as a memory controller to access the on-board SO-DIMM and what-not, or that there's enough room on the FPGA to accomodate said bits. And that you have the IP of those bits, of course.
Then the T2 will probably be larger, even scaled down to an hypothetical single-core S2. I seriously doubt than you'll be able to squeeze a fully bootable S2 system in less than a V4 LX160. A PCI(-e,-X) board with one of those and some memory for your OS & stuff will probably be way over $3000. That's not counting the software tools, as free version of the tools are limited to the smaller devices, and the already mentioned IP that might be necessary...
Plus, you'll need the skills to bolt all the HDL, software and tools parts together.
It's likely doable, but at a very high cost in both money & time.
PCI-e boards loaded with FPGA exist, but you obviously haven't tried to price one. Although you can get a basic prototyping board with a small Virtex-5 and 256 MiB or RAM for about $1200, cards with large/high speed grade FPGA and a decent amount of memory will set you back anywhere between $3000 and $25000. None of them will run complex SPARC v9 code at a speed anywhere near that of the real thing, because that's not what FPGAs are good at (even though they'll be much better at emulating a chip with loads of slow cores than a chip with few fast cores).
Buying the real thing would make a lot more sense.
O rly? I've seen advertisements for "business DSL" service with service levels for home offices and small businesses.
My bad ; what I really meant whas that the ADSL providers who target individuals expect them not to care for fax.
And it is this subset of providers which actually offers unlimited free communications to some foreign countries, including the USA. I have no doubt that there's other kinds of offers (maybe including unlimited calls) made to businesses, but they're not comparable to the AOL offer in the article and so are not relevant.
All this was only to answer to the poster that mentioned that Skype, at.02c a minute between Europe and the USA, was a bargain and not a racket as described (implicitely) in the article.
And if you need an extension on each floor of your house for "emergency" purposes, then what?
You use DECT phones (and pray power doesn't fail), or you stick with real POTS lines. Anyway,
I don't see the difference with the VoIP-in-computer approach: both have the same problems
wrt reliability ; but the integrated VoIP approach it a lot easier for the end user.
It ain't perfect, but communications don't come much cheaper than free...
For people who rely on fax, do the telcos offer an add-on fax service?
The ADSL providers target private individuals who don't need a fax, not businesses ; beside,
the country I know best and talk about doesn't have telcos - it has one telco, the former
state monopoly. Everyone else is doing telephony over a network ; I'm not really sure
if it's VoIP for everyone or if some do VoATM or Vo<whatever> ; but it ain't POTS,
that's for sure.
And e-mail is a lot better than fax anyway <geek smirking>
Slashdot Mirror
The MythBusters did that one 7 years ago: https://en.wikipedia.org/wiki/MythBusters_(2010_season)#Left_Hand_Turn
Why would I need a terminal *emulator* ? :-)
https://en.wikipedia.org/wiki/...
https://en.wikipedia.org/wiki/... (XTerm still emulates the vector graphics of the 4014...)
https://en.wikipedia.org/wiki/...
I still remember the time when "multiple terminals" meant "grab another nearby vt100 while the first one's busy". Then there was screen(1), and we saw that it was good
How "modified" was the Shield to run Ubuntu? Can I buy a Shield and get Ubuntu on it today? Or is this benchmarking an exercise in futility?
And the SS10 didn't have the ZX framebuffer of the SS20 - 24 bits accelerated graphics. 2nd one available as an option. That was an awesome machine, still one of the greatest computer design ever.
:-(
I started using UNIX with SunOS 4.1.4 on SPARCstation 1+. Still have my SS2 somewhere in the attic. Still have my Ultra 1 Creator. Can't throw them away.
I like the speed of my W3680 - but it's just not the same.
Guess I'm old
Cue this old joke...
...
- How many hardware engineers does it take to change a light bulb?
- None, we'll fix it in software.
Doing stuff in software to make hardware easier has been tried before (and before this kid was born, perhaps why he thinks this is new). It failed. Transputer, i960, i432, Itanium, MTA, Cell, a slew of others I don't remember...
As for the grid, nice, but not exactly new. Tilera, Adapteva, KalRay,
It's obvious. Satoshi Nakamoto is a pseudonym of Nicolas Bourbaki.
Explain to them how to fix the problem once and for all:
1) block outgoing port 25 to everything but their own mail servers;
2a) add an optional feature in each customer account to reopen outgoing port 25;
2b) add an optional feature in each customer account to pick the reverse DNS entry;
3) tell every other ISP/mail servers operators what they have just done,
so they get un-blacklisted since they won't be sending much spam any more.
This should block most of the outgoing spam without any side-effects,
since power users will still be able to operate their own mail servers,
complete with reverse FQDN. Non-power users won't notice a thing.
Also, they will save money on bandwidth to the outside world.
That's what my (strictly residential) ISP has been doing for almost a decade.
Works perfectly well for everyone involved.
http://www.adapteva.com/epipha... :-p</p></quote>
64 cores, mesh network that extends off the chip, in production.</p><p>Try harder MIT
They already tried harder : http://www.tilera.com/. And as another post mentioned, Intel Knights Corner is cache coherent on 61 cores (62 architectured).
The summary doesn't get the point of the article: what's novel is not the presence of cache coherency, it's just the new way of implementing snoop-based cache coherency over their network. Cache coherency for a large number of cores can be very expensive time-wise, so any idea to improve it is more than wecome.
Why is everybody thinking this is big news?
ftp://download.nvidia.com/CUDAOpen64/
The previous compiler, based upon Open64, has been available in source form since CUDA 1.0. They (partially) switched to LLVM in 4.1, and they also release the source code. They didn't have to, because unlike Open64 LLVM is not GPL, so it's nice of them, but it's not exactly earth-shattering news...
And because a picture straight from the horse's mouth is worth a thousand words, here's what NVidia has to say about it:
http://http.developer.nvidia.com/GPUGems3/gpugems3_ch36.html
Go to 36.5, figure 36-11 & 36-13.
The Library of Congress used to have a goal of including complete hard copies, at least for items of US origin and 'good grade' (that is, they aimed to have copies of things such as hardback books that were intended to last, more than, say, ephemera such as the pulp magazines). However, that goal has become an obvious impossibility due to sheer volume. After about 1960, the library began being more selective.
And the situation is infinitely worse for other medias. Not only aren't people trying to preserve them, in many case they have been actively destroyed, in particular television broadcasts.
Two examples of the casualties:
The relevant wikipedia category is http://en.wikipedia.org/wiki/Category:Lost_television_programs. It's hard to believe so much television history has been lost forever.
If I buy a Chablis or a Burgundy I want a particular type of wine. So what that these wines originated in certain regions in France?
They didn't "originate". If it's a burgundy, then it hast to come from the region of Burgundy. It's that simple. Also, for the record: if you buy a Chablis, you also buy a Burgundy. Chablis is a sub-region of Burgundy.
I don't give a damn where it was made. I would say most people who drink them don't know or care either.
Some of us haven't ruined their taste buds with bad beers and ketchup sauce, so we do care. Where the wine was produced makes a lot of difference to the taste. If you can't tell the difference, please go back to drinking Budweiser.
I'm told by a French friend who is a wine buff that the Aussie wines he can buy are superior to French wines (seriously), so this makes the whole thing sound like a ploy to recapture an ailing market.
There is no such thing as "superior", either way. There is such as thing as "different". Then it's a matter of taste. Australia, California, Chile, Algeria all make very good wines. They just aren't Burgundy, or Champagne. Would you expect a "Scotch Whisky" to come from Polland? Obviously no. It doesn't preclude Japanese to make great Single Malt Whiskies. They just don't make Scotch Whiskies. Think of it as a trademark, shared by all the producers from one geographic region. You can't buy a Macintosh from Hewlett-Packard, can you? So why should you be able to buy a Burgundy from someone that isn't located in the region of Burgundy, and therefore doesn't share in the trademark?
And how many MB can you address with a 32-bit pointer under the IEEE recommendations?
1) I don't know...
2) It doesn't [censored] matter!
3) It's exactly 4 GiB, or 4096 MiB, how hard is that?
4) I don't use 32 bits pointer anymore anyway.
See, it's very easy: just add a little 'i' in there, and it works exactly like before, just unambiguously.
Over 300 posts and counting, and all because people can't type 'i' to make sure there's no possible mistake... The world is doomed, I tell you, doomed! :-)
(sigh) No. The maximum available memory was measured in Megabytes, then Gigabytes. It was always base 2.
Well, not at first, 'cause a megabyte (of whatever size) of memory didn't exist yet.
Second, it was wrong but easy. It's still wrong and for memory, it's still easy. That's what it's still in use for memory, and is being ditched from everywhere else.
The main problem was that the binary prefix came too late. Old habits die hard. But as AA proves, it's possible to ditch a bad habit if you really want to. It's a matter of willpower. Yes we can, if I may so bold as to say so.
(for the record, I have been involve in computing since before Bill Gates informed us that 640 Kilobytes should be more than anyone will ever need, so I actually was there as all this unfolded)
And the fact that you're the new kid on the block is important because...?
Because our computers, almost since their earliest inception, work in base-2 arithmetic.
And this matters because...? The SI prefix are used to denotes quantity. Except for the total size of semi-conductor memory (and sub-elements thereof), those quantities are usually completely unrelated to power of 2.
Disk sizes are not a power of 2. Files stored on them are even more arbitrary in size. Same for memory requirements of various codes. Heck, once upon a time, HPC programs would overallocate arrays to avoid power of 2 allocation (multiple of the page size wrecked havoc on direct-mapped caches)!
Frequency are not power of 2. Your 3 GHz processors runs at 3*1000^3 Hz (well, probably not very precisely :-), not 3.22*1000^3 HZ. Same goes for the (directly related) bandwidth. Latency doesn't even come close.
Heck, these days, even memory buses are moving away from strict power of 2: GTX 275 have a 448 bits-wide memory bus, Core i7's can be described as 192 bits.
And for the few cases where it matters (usually not end-user visible) that's what the freaking/frelling/rutting/smegging/[pick you favorite SciFi show euphemism] binary units are for !
(I'm not going to try to make sense from the rest of your post, because I can't see how the number of bits in a bytes is related to the discussion in any way).
Uh.. the inch is technically an SI unit. It is defined as exactly 2.54 cm.
No, it's not. SI uses the metre for length measurement, and nothing else. You can alter it with the various prefixes, and there's is only one thousand meters in a kilometre, not twenty-four more.
The "inch" from the United States customary units is defined as 2.54 centimetres, but it doesn't make it part of the SI..
That's just plain wrong. If my laptop has 3 Gigabytes of RAM in it, there had sure as hell better be 2^30 individually addressable locations. If there are only 10^9 addresses, they ripped me of by ((2^30) - (10^9))
(sigh).
Your laptop doesn't have 3 Gigabytes. It has 3 Gibibytes. That's 3221225472 bytes, or a bit more than 3.22 Gigabytes. Think of it that way: you got 7.37% more memory than you paid for. You ripped them off! Happy now?
The fact that both the salesdrones and the buyers are ignorant doesn't change the definition of "Giga" and "Gibi".
So far, only the hard drive manufacturers have catched up. When will the "3.22 gigabytes of memory!" advertisements show up?
The lasting ambiguity for hard drives has perhaps been less a matter of computer science than one of marketing. (The pervasiveness of inch measurements is a heavy hint at uninterest in SI.)
Actually, people who use the metric system outweigh people using imperial by a wide margin. In my entourage, nobody except for geeks knows how long an inch is. As you can guess, I don't live in the US. As for the marketing: yes, marketdroids will do anything to sell. We all know that. But when "anything" becomes "make use of an established international standard", how can we protest?
And C.S. is at fault there. Why did it hijack prefixes to make them mean something else? It was stupid from the beginning. For Kilo I can understand, the error is small ; but from Mega onward, it was just plain laziness. That's why I try to use binary prefixes these days. Atonement :-)
Obligatory old timer anecdote: remember those "1.2MB" and "1.44MB" floppy disk? Worst of them all: the megabytes here were neither 1024^2 nor 10^6, but 1024*10^3, as in 1200 KiB and 1440 KiB.
The SI prefixes have been around for nearly 5 decades, and have a specific meaning used by everybody. Every scientist uses them in one way or another, and for every last one of of them, they have the same meaning.
Why can't we, the C.S. people, accept that?
Giga is 10^9. It has been 10^9 since it was created. It was never, ever meant to be anything but 10^9.
If you want to talk about 1024^3, then it's Gibi. Gibi is 2^30 since it was created. It was never, ever meant to be anything but 2^30.
Get over it.
(and yes, I try to always use GiB whenever it's appropriate).
> I originally thought this was breaking 18 U.S.C. Chapter 119, 2510 to 2522 (?), but no.
Sorry, completely irrelevant: I didn't notice it's Bell Canada, so U.S. laws don't matter.
I have no doubt there's some US ISP doing the same thing :-(
IANAL.
I originally thought this was breaking 18 U.S.C. Chapter 119, 2510 to 2522 (?), but no.
*IF* they only alter the answer of their own DNS servers to their clients, when the client has made a request to said DNS servers, then they're probably in the clear. There is two communications: one from the client (C) to the Bell server (B), then one from B to the authoritative server (S). S then answer NXDOMAIN to B, which then returns a completely different information to C. So they're not intercepting anything.
OTOH, *IF* they hijack all the port 53 requests to the outside world (which I doubt), then it's very likely 2511(1)(a) and (d) applies. They still could argue under 2511(2)(a)(i) that it's "necessary"...
Then again, IANAL.
OTOH, even if it's legal, it's still absolutely wrong.
Unfortunately no I'm not saying that :-)
First there's a typo ; the basic ML555 is $2200, not $1200.
Someone managed to synthesize a S1 (single-core T1, with multithreads support) for a Virtex 4 LX60. You might be able to squeeze a S1 into the Virtex 5 LX50T of the ML555, but I'm not certain, it mght be a bit short on available slices. Beside, it seems that they didn't go any further than that, and didn't actually try to upload the bitsream, let alone boot an OS. All that is assuming either the synthesized HDL included all required bits beyond the CPU to do so, such as a memory controller to access the on-board SO-DIMM and what-not, or that there's enough room on the FPGA to accomodate said bits. And that you have the IP of those bits, of course.
Then the T2 will probably be larger, even scaled down to an hypothetical single-core S2. I seriously doubt than you'll be able to squeeze a fully bootable S2 system in less than a V4 LX160. A PCI(-e,-X) board with one of those and some memory for your OS & stuff will probably be way over $3000. That's not counting the software tools, as free version of the tools are limited to the smaller devices, and the already mentioned IP that might be necessary...
Plus, you'll need the skills to bolt all the HDL, software and tools parts together.
It's likely doable, but at a very high cost in both money & time.
PCI-e boards loaded with FPGA exist, but you obviously haven't tried to price one. Although you can get a basic prototyping board with a small Virtex-5 and 256 MiB or RAM for about $1200, cards with large/high speed grade FPGA and a decent amount of memory will set you back anywhere between $3000 and $25000. None of them will run complex SPARC v9 code at a speed anywhere near that of the real thing, because that's not what FPGAs are good at (even though they'll be much better at emulating a chip with loads of slow cores than a chip with few fast cores).
Buying the real thing would make a lot more sense.
My bad ; what I really meant whas that the ADSL providers who target individuals expect them not to care for fax. And it is this subset of providers which actually offers unlimited free communications to some foreign countries, including the USA. I have no doubt that there's other kinds of offers (maybe including unlimited calls) made to businesses, but they're not comparable to the AOL offer in the article and so are not relevant.
All this was only to answer to the poster that mentioned that Skype, at .02c a minute between Europe and the USA, was a bargain and not a racket as described (implicitely) in the article.
You use DECT phones (and pray power doesn't fail), or you stick with real POTS lines. Anyway, I don't see the difference with the VoIP-in-computer approach: both have the same problems wrt reliability ; but the integrated VoIP approach it a lot easier for the end user.
It ain't perfect, but communications don't come much cheaper than free...
For people who rely on fax, do the telcos offer an add-on fax service?
The ADSL providers target private individuals who don't need a fax, not businesses ; beside, the country I know best and talk about doesn't have telcos - it has one telco, the former state monopoly. Everyone else is doing telephony over a network ; I'm not really sure if it's VoIP for everyone or if some do VoATM or Vo<whatever> ; but it ain't POTS, that's for sure.
And e-mail is a lot better than fax anyway <geek smirking>