But if their figures are vaguely comparable, the world's fastest computer is a volunteer effort to look for Space Aliens
They're not vaguely comparable. Linpack simply wouldn't work at all over a seti@home style system. You would get the best performance by finding the fastest single machine or LAN running seti@home and have them alone run linpack, totally igorning the thousands of other computers. If you tried to get MPI and linpack running on every computer in that 58 TFLOPS number, the actual performance would be something like 0 FLOPS. Someone's power would go out before you could complete even the smallest benchmark.
....maybe i'm obtuse, but i keep hearing about this thing as "..and we're only seeing X% of its real potential right now!"....
When people are saying "54% of peak" they aren't saying, "54% of the real potential," but something more like, "54% of some made up number than could never be achieved."
It's like saying that if a Kawasaki Ninja ZX-12R was dropped from the upper atmosphere, its aerodynamics would give it a terminal velocity of 1000 mph. But it can only go 198.8 mph on the road, that's only 19.8% of peak!
The cluster has a theoretical peak of 17.6TFlops/s if I did my math right (8GFlops/s per processor), but they are only turning in an actual score of 9.56TFlops/s, for an efficiency of only 54%.
The reason the efficiency is low isn't so much because the the 9.56 TFLOPS is a low number, but rather that the theoretical peak of 17.6 is unrealistically high. The only way you could get 17.6 is if you did nothing but paired multiply-add sequences entirely out of cache. No real code does this and so the 17.6 number is really nothing more than marketing bullshit. When an It2 or Xeon clsuter or NEC's Earth Simulator get better efficiencies it's because their made up "peak" numbers are more realistic than the one the marketing people used for the G5.
You could calculate a new marketing BS peak number where multiply-add only counted as a single flop, or you took into account some realistic cache miss rate. The new lower theoretical peak would give you a much higher efficency.
Pretty good info. However, the nominal voltage of Li-ion batteries is 3.4V or 3.5V, depending on the type. The maximum voltage they reach during charging is 4.1V or 4.2V, but they quickly drop down to the nominal voltage.
Lithium batteries are required to have a built in fuse, to keep them from exploding if they are shorted. Bare lithium cells have a built in thermisor based fused. Cells for OEM use in sealed battery packs don't have a built in fuse, but use an electronic one in the battery pack.
Exactly! I was beginning to feel old. Though the templates didn't go above the keys at first, originally the ten function keys were in two columns offset on the left side of keyboard. The template was a rectangle with a large hole in the middle that neatly fit around the function keys. Pretty much all major apps used one. In the days before mice, GUIs and menus, all the app's functions would be assigned to a function key plus modifier key. I think Alt-F7 was reveal codes and shift F6 was format in word perfect, but don't hold me to that, it's been like 10 years since I've used it.
I believe the keyboards also accepted a signal that made them switch to
"non-back-compatability" mode where all keys produce a unique code, and I think all
Windows and Linux systems set this.
You send the keyboard command 0xF0, then select the keyboard scan code set.
Sets 2 and 3 and generate a unique code for each key press, then F0 and the same code for the key release. Set 1 (original XT) sends a unique code for some keys for a press, and that code plus 0x80 for a release. Lots of special keys have a 0xe0 prefix, and some, like the cursor keys as you pointed out, send shift+number key. The pause key generates control down, num-lock down, control up, num-lock up. There wasn't a pause key on the XT, you pushed control-numlock to pause output. There wasn't a print screen either, it was shift numberpad * key.
The original PC keyboard, before the 101 key enhanced keyboard came out for the PC-AT, didn't have separate cursor keys. There was just one set of keys, used for both numbers and cursor control. If num lock was on, you got numbers. If num lock was off, then 8,4,6, and 2 were up, left, right and down. 7 is home, 9 is pageup, 1 is end and 3 is pagedown, 0 was insert and . delete. I don't remember there being a separate "enter" key by the numeric pad, just a giant + key that was as long as 9, 6, and 3.
When the AT came out, all kinds of cool new keys were added. F11 and F12, the inverted-T of full time cursor keys, and extra control and alt on the right side.
My antique northgate omnikey keyboard has the cursor keys in a 3x3 square with pageup/down/home/end, like they belong. Inverted T sucks. It also has the letters for control keys in red, shift in green, and alt in blue. Bonus karma for the first person to correctly answer why that is.
If the items weren't recovered, the store would lose 100% of the item's cost, not just the 10% comission. So, if normal shoplifting results in a 10% markup, then loss prevention people getting a 10% commission would reduce the shoplifting markup to 1%.
Except the store has to pay the salaries of the door checkers. In reality, the biggest effect is that the presence of a door check should deter a large amount of shoft-lifting in the first place.
The power pins on a PATA drive are huge compared to the size of the data pins. That's because power connectors need to be larger to carry higher currents. Data lines have almost no current, so they can be much smaller. SATA uses the same size pins for data and power, so it needs to use multiple pins for increased current carrying capability.
Just like an ATX motherboard power connector has seven black ground lines, four 5 volt lines, three 3.3V lines, and one 12V line. And then when the P4 came out, they needed to add the secondary connector with two more 12V lines for the extra power. Those ATX pins aren't big enough one pin to carry all the current, so they use mutliple pins.
This is probably the largest reason I don't use IDE in production outside of workstations. SCSI drives
normally have 128-256 heads (unless something has drastically changed, in which case I'll no doubt be
corrected), where IDE in any flavor has 16.
You're totally wrong about the number of heads. The number the BIOS reports is just some fake value for legacy compatibility with 1980s era PC BIOS design. Real harddrives have more like 1 to 9 heads. A huge full size drive (for you youngsters, that's the size of two normal CD-ROM drives on top of each other) might have 16 heads. No hard drive, except maybe some two ton refrigerator sized monster from the 60s, has 256 heads.
BTW, more heads don't buy you anything, except
more heat and noise. Drives can only read from one head at a time.
If you check current prices, the Xeon isn't much more expensive than the AthlonMP. Pricewatch has the 2.2Ghz xeon at $245 and the athlonMP 2200+ at $204. Each of these machines is interconnected with a Quadrics board that probably costs more than $2000, so an extra $80 for CPUs isn't much.
Why not use AMD anyway? There are xeon motherboards with chipsets like the Intel E7500 and ServerWorks GC-HE that have greater memory bandwidth and PCI bandwidth than the AMD 760MPX. For many problems in scientific computing, memory bandwidth is what is important, not CPU speed.
There are chips designed to connect two PCI busses together, called PCI-PCI Bridges. For instance, I have an Intel dual port ethernet card with one:
Bus 0, device 12, function 0:
PCI bridge: Digital Equipment Corporation DECchip 21152 (rev 3).
Master Capable. Latency=64. Min Gnt=4.
But you can't use this to connect a rack of computers. For one thing the max cable length for connecting two busses would be just a few inches. For putting PCI cards in 1.75" high 1U rackmount cases, there are PCI risers with a short ribbon cable that connects to the PCI slot. Even these short cables often cause timing problems. For instance, with the riser, cards may only work in the first one or two slots that will otherwise work in all the slots.
But even if you could cable all the computers together on one giant PCI bus, it would still be a bad idea. A good 24 port gigabit ethernet switch (~$2000) has a 480MB/sec switching fabric, to support full speed full duplex on each port. 32 bit 33Mhz PCI is only about 132 MB/sec, not nearly as fast. You'd need a 64 bit 66 Mhz PCI bus to keep up. And there are more expensive gbit switches with more ports that have 100 Gbit/sec fabric. And this is just gbit ethernet, the slowest and cheapest of the high speed interconnects used in modern Beowulf clusters.
There are faster ways to connect computers than gigabit ethernet. The EE times article is very untechnical, but this one has some more information. LLNL has used a very fast and very expensive interface called quadrics. This is probably the fastest way to connect computers in a Beowulf. People like Cray/SGI and IBM have faster things still, but they cost real big bucks. Other ways to connect a Beowulf are the above mentioned gigabit ethernet (~$100-$250 a node for up to 24 nodes), myrinet (~$1400-$2000/node up to 128 nodes), and SCIhardware and software (~$1400-$2100/node). Myrinet uses a switch like gigabet ethernet and the largest switch they have is 128 ports. SCI is switchless, each card has multiple cables (1-3), and is connected in into a ring, 2D or 3D torus.
For a given battery chemistry, the time it takes to fully charge the battery is relatively constant. Li-ion cells for instance normally take two hours to fully charge. The maximum charge rate for li-ion cells is normally 1C, where C is the capacity of the battery in amp-hours. So if you have a small 600 mAh cell it change take a charge at a max rate of 600 mA, while a larger 1800 mAh cell could be charged at 1.8 amps, but both will take about the same time to charge. NiMH and especially nicad batteries can charged at higher rates, like 3C or more. But still, a one minutes charge time would mean a peak charge rate of more than 60C, a no battery can take that.
I've never seen a "low profile" TV capture card. In fact, I've never even seen anything other than an ethernet card for sale in the low profile form factor. I've used a bunch of ~$40 V2000 bookPC cases to make some x-terminals for work, and I'd love to use one to make a home theater DivX/MP3 player. Except they use those stupid low profile slots so I can't stick any PCI cards in the thing. I want to get NTSC and toslink output somehow.
If you want video input, you could get an external DV capture box or even a DV video camera. These convert your analog video into DV, which gets sent to your computer via firewire. Now you just have to find a motherboard with builtin firewire or a low profile firewire card. Another option would be the canopus advc50, it's about the size of a pci card but is designed to fit in a 5 1/4" drive bay and uses a harddisk power connector. Get a motherboard with an internal firewire header and you could have a totally internal solution. These external DV boxes cost something like $200-$300, but are much better quality than those $60 PCI capture cards.
You can also try to find a case that uses a PCI riser to give one or two pci slots parallel to the motherboard. Most PCI capture cards that I've seen aren't very large, and should be able to fit. I can't find any cases like this that would make a nice home theater style component. There are those overpriced shuttle cubes cases, but a cube is not the right shape.
There are other RF connectors like BNC, but with some feature different. For instance, there is a TNC connector that's the same size but has threads instead of a bayonet mount. Seems logical that B and T stand for Bayonet and Threaded, doesn't it? There is also an N connector that looks like a TNC but is much larger. That's probably where the "Baby N Connector" version of what BNC stands for came from.
Accoring to the rules, player's robots will be competing with 0 or more other robots from other contestants. What if a group of people each enter a robot, then have the robots work together.
You could open a socket and have the other robots try to connect, then communicate that way. That might be hard, if for example the robots are running on different machines or the organizers check for open ports.
Since all the robots have almost complete information, you don't need to communicate. Have your robot do a little dance at the beginning, left right left right up down or something, to identify it as a team member. Your robot knows what the team members are doing because it can just compute what their decisions will be. The only information you lack is the weight and destination of a package that a teammate picked up.
You could have the robot with the lowest X & Y coordinates be the leader. The other robots stay around him so he doesn't get bumped. Or carry packages to him to deliver. Or hang next to the home bases, and when another robot moves onto them, bump them so they can't pick the package. Since it takes one turn to pick up a package, I think it would be trivial to make a robot that hanges near a base and can prevent any single other robot from ever picking the package.
It's a Hitchhiker's Guide reference. In the series by Douglas Adams, there is a computer that called "Deep Thought" that finds the answer to life, the universe, and everything else. (Which is 42). A famous chess playing computer was named Deep Thought. IBM's Deep Blue is a take on that name and IBM's "corporate color" of blue.
More like a lack of documentation effort. I remember reading those exact same "download a binary and rename it" docs when I started in 1994! By now they are like 10 years out of date.
To install paradise 2000, which is on linux, you just unpack the tarball, copy a config file into your home dir, and run the program. Pick your server from the window that pops up. It has an INSTALL file that describes everything. It's a lot less trouble than installing quake and qspy and mesa libraries etc. I was going to create a RPM package, but this article caught me (and all other netrek developers) by surprise.
It's like the software and FTP list, which lists servers that dissapeared years ago, but not the new ones. If you look on the ftp.netrek.org site, clients like trekhopd and BRM are listed right along with COW, netrekxp and paradise-2000. There is no mention that some clients are from the late 80s and some from this century.
Re:I learned network programming from Netrek
on
Netrek
·
· Score: 2
I added one feature still missing (I think) from all others: it would read all the messages to you so
that you didn't have to look down at the message window. Important when dodging torps! The
Amiga speech device made it easy.
I've added speech to Paradise 2000, my pardise client for Linux. It's somewhat better than the amiga version. I have special case code for things like bombing messages, dooshes, kills, etc.
The RCD macros can be configured too, for example you say "voice.generic: %?%S=SB%{help base&}" and you'll just get the message "help base" spoken too you instead of a line of numbers with the base's status.
I think someone has compiled Ted Turner for XDarwin, check the netrek ftp site here for the Ted Turner binaries, there is one for Darwin.
I also think I've seen someone playing with Paradise 2.99 for Darwin, but I don't see the binary anywhere. As far as I know, no one has
ported COW or BRMH to MacOS.
New clients for Netrek
on
Netrek
·
· Score: 3, Interesting
It's not quite true that there have been no client changes in years. My client has had new versions come out every few months since 1999, with lots of new features in each version. A new version of a windows client just came out too, but it's not nearly as nice as my client.
<plug mode>
You can read more and download the software from my web page for Paradise 2000, the ultimate Linux netrek client.
It has a nice sound system and can use IBM's ViaVoice for linux to do speech synthesis of messages and macros. Getting the IBM ViaVoice TTS package for linux is hard now, maybe/. should do a story on that. </plug mode>
One problem Netrek has right now is lack of servers. The one popular server is often full. It also has had bad lag for most people recently, since it is a redhat and openoffice mirror, both which have released major new versions.
I remember the nights of playing netrek on those crappy X-terms with their 12" monitors. It was at the ACC, closed down in 1999 I think.
There are still some people playing netrek. If you use Linux, check out my client Paradise-2000. If you download IBM's viavoice
package for linux, you can have it speak messages and macros outloud to you.
The memory usage of KDE and gnome is just huge. We have a few xterminals running off a single linux server, and the number of processes that get started for kde/gnome is a real problem.
With the versions and default sessions I get with redhat 7.2, I measued the memory usage of KDE and gnome. KDE weighs in at a hefty 95 MB, while gnome uses "only" 41 MB. For comparison, the fvwm2 setup I use includes an email checker that tells me how many mesages I have, a clock, a loadgraph, cpu usage graphs (per CPU), button bar, and virtual desktop pager. The workings of the window manager itself are more configurable than either gnome or KDE. And all this is only 4MB! That's about 24 times less than KDE.
You have download for free a 30 second resolution
digital elevation for the whole planet from the USGS. It's called GTOPO30 and it's been avialable since 1996.
As to why more isn't avialable for free, it costs
money to create data like this. One way or another, someone needs to pay for it. If taxes aren't enough to provide hi-res images of the entire world, then that leaves private enterprise.
Weatherballoons are usually let up in area's where commercial flight is
scarce. If we want to use this technology for relaying radiowaves etc. this
will change.. so we have to reroute all commercial airtraffic to avoid
accidents?
No, the ballons are usually sent up at airports! And they've been doing this since the thirties, so I think if it was problem, we would know about it by now.
Slashdot Mirror
It's like saying that if a Kawasaki Ninja ZX-12R was dropped from the upper atmosphere, its aerodynamics would give it a terminal velocity of 1000 mph. But it can only go 198.8 mph on the road, that's only 19.8% of peak!
You could calculate a new marketing BS peak number where multiply-add only counted as a single flop, or you took into account some realistic cache miss rate. The new lower theoretical peak would give you a much higher efficency.
Pretty good info. However, the nominal voltage of Li-ion batteries is 3.4V or 3.5V, depending on the type. The maximum voltage they reach during charging is 4.1V or 4.2V, but they quickly drop down to the nominal voltage.
Lithium batteries are required to have a built in fuse, to keep them from exploding if they are shorted. Bare lithium cells have a built in thermisor based fused. Cells for OEM use in sealed battery packs don't have a built in fuse, but use an electronic one in the battery pack.
Exactly! I was beginning to feel old. Though the templates didn't go above the keys at first, originally the ten function keys were in two columns offset on the left side of keyboard. The template was a rectangle with a large hole in the middle that neatly fit around the function keys. Pretty much all major apps used one. In the days before mice, GUIs and menus, all the app's functions would be assigned to a function key plus modifier key. I think Alt-F7 was reveal codes and shift F6 was format in word perfect, but don't hold me to that, it's been like 10 years since I've used it.
Sets 2 and 3 and generate a unique code for each key press, then F0 and the same code for the key release. Set 1 (original XT) sends a unique code for some keys for a press, and that code plus 0x80 for a release. Lots of special keys have a 0xe0 prefix, and some, like the cursor keys as you pointed out, send shift+number key. The pause key generates control down, num-lock down, control up, num-lock up. There wasn't a pause key on the XT, you pushed control-numlock to pause output. There wasn't a print screen either, it was shift numberpad * key.
The original PC keyboard, before the 101 key enhanced keyboard came out for the PC-AT, didn't have separate cursor keys. There was just one set of keys, used for both numbers and cursor control. If num lock was on, you got numbers. If num lock was off, then 8,4,6, and 2 were up, left, right and down. 7 is home, 9 is pageup, 1 is end and 3 is pagedown, 0 was insert and . delete. I don't remember there being a separate "enter" key by the numeric pad, just a giant + key that was as long as 9, 6, and 3.
When the AT came out, all kinds of cool new keys were added. F11 and F12, the inverted-T of full time cursor keys, and extra control and alt on the right side.
My antique northgate omnikey keyboard has the cursor keys in a 3x3 square with pageup/down/home/end, like they belong. Inverted T sucks. It also has the letters for control keys in red, shift in green, and alt in blue. Bonus karma for the first person to correctly answer why that is.
If the items weren't recovered, the store would lose 100% of the item's cost, not just the 10% comission. So, if normal shoplifting results in a 10% markup, then loss prevention people getting a 10% commission would reduce the shoplifting markup to 1%.
Except the store has to pay the salaries of the door checkers. In reality, the biggest effect is that the presence of a door check should deter a large amount of shoft-lifting in the first place.
Just like an ATX motherboard power connector has seven black ground lines, four 5 volt lines, three 3.3V lines, and one 12V line. And then when the P4 came out, they needed to add the secondary connector with two more 12V lines for the extra power. Those ATX pins aren't big enough one pin to carry all the current, so they use mutliple pins.
BTW, more heads don't buy you anything, except more heat and noise. Drives can only read from one head at a time.
If you check current prices, the Xeon isn't much more expensive than the AthlonMP. Pricewatch has the 2.2Ghz xeon at $245 and the athlonMP 2200+ at $204. Each of these machines is interconnected with a Quadrics board that probably costs more than $2000, so an extra $80 for CPUs isn't much.
Why not use AMD anyway? There are xeon motherboards with chipsets like the Intel E7500 and ServerWorks GC-HE that have greater memory bandwidth and PCI bandwidth than the AMD 760MPX. For many problems in scientific computing, memory bandwidth is what is important, not CPU speed.
Bus 0, device 12, function 0: PCI bridge: Digital Equipment Corporation DECchip 21152 (rev 3). Master Capable. Latency=64. Min Gnt=4.
But you can't use this to connect a rack of computers. For one thing the max cable length for connecting two busses would be just a few inches. For putting PCI cards in 1.75" high 1U rackmount cases, there are PCI risers with a short ribbon cable that connects to the PCI slot. Even these short cables often cause timing problems. For instance, with the riser, cards may only work in the first one or two slots that will otherwise work in all the slots.
But even if you could cable all the computers together on one giant PCI bus, it would still be a bad idea. A good 24 port gigabit ethernet switch (~$2000) has a 480MB/sec switching fabric, to support full speed full duplex on each port. 32 bit 33Mhz PCI is only about 132 MB/sec, not nearly as fast. You'd need a 64 bit 66 Mhz PCI bus to keep up. And there are more expensive gbit switches with more ports that have 100 Gbit/sec fabric. And this is just gbit ethernet, the slowest and cheapest of the high speed interconnects used in modern Beowulf clusters.
There are faster ways to connect computers than gigabit ethernet. The EE times article is very untechnical, but this one has some more information. LLNL has used a very fast and very expensive interface called quadrics. This is probably the fastest way to connect computers in a Beowulf. People like Cray/SGI and IBM have faster things still, but they cost real big bucks. Other ways to connect a Beowulf are the above mentioned gigabit ethernet (~$100-$250 a node for up to 24 nodes), myrinet (~$1400-$2000 /node up to 128 nodes), and SCIhardware and software (~$1400-$2100 /node). Myrinet uses a switch like gigabet ethernet and the largest switch they have is 128 ports. SCI is switchless, each card has multiple cables (1-3), and is connected in into a ring, 2D or 3D torus.
For a given battery chemistry, the time it takes to fully charge the battery is relatively constant. Li-ion cells for instance normally take two hours to fully charge. The maximum charge rate for li-ion cells is normally 1C, where C is the capacity of the battery in amp-hours. So if you have a small 600 mAh cell it change take a charge at a max rate of 600 mA, while a larger 1800 mAh cell could be charged at 1.8 amps, but both will take about the same time to charge. NiMH and especially nicad batteries can charged at higher rates, like 3C or more. But still, a one minutes charge time would mean a peak charge rate of more than 60C, a no battery can take that.
I've never seen a "low profile" TV capture card. In fact, I've never even seen anything other than an ethernet card for sale in the low profile form factor. I've used a bunch of ~$40 V2000 bookPC cases to make some x-terminals for work, and I'd love to use one to make a home theater DivX/MP3 player. Except they use those stupid low profile slots so I can't stick any PCI cards in the thing. I want to get NTSC and toslink output somehow.
If you want video input, you could get an external DV capture box or even a DV video camera. These convert your analog video into DV, which gets sent to your computer via firewire. Now you just have to find a motherboard with builtin firewire or a low profile firewire card. Another option would be the canopus advc50, it's about the size of a pci card but is designed to fit in a 5 1/4" drive bay and uses a harddisk power connector. Get a motherboard with an internal firewire header and you could have a totally internal solution. These external DV boxes cost something like $200-$300, but are much better quality than those $60 PCI capture cards.
You can also try to find a case that uses a PCI riser to give one or two pci slots parallel to the motherboard. Most PCI capture cards that I've seen aren't very large, and should be able to fit. I can't find any cases like this that would make a nice home theater style component. There are those overpriced shuttle cubes cases, but a cube is not the right shape.
There are other RF connectors like BNC, but with some feature different. For instance, there is a TNC connector that's the same size but has threads instead of a bayonet mount. Seems logical that B and T stand for Bayonet and Threaded, doesn't it? There is also an N connector that looks like a TNC but is much larger. That's probably where the "Baby N Connector" version of what BNC stands for came from.
You could open a socket and have the other robots try to connect, then communicate that way. That might be hard, if for example the robots are running on different machines or the organizers check for open ports.
Since all the robots have almost complete information, you don't need to communicate. Have your robot do a little dance at the beginning, left right left right up down or something, to identify it as a team member. Your robot knows what the team members are doing because it can just compute what their decisions will be. The only information you lack is the weight and destination of a package that a teammate picked up.
You could have the robot with the lowest X & Y coordinates be the leader. The other robots stay around him so he doesn't get bumped. Or carry packages to him to deliver. Or hang next to the home bases, and when another robot moves onto them, bump them so they can't pick the package. Since it takes one turn to pick up a package, I think it would be trivial to make a robot that hanges near a base and can prevent any single other robot from ever picking the package.
It's a Hitchhiker's Guide reference. In the series by Douglas Adams, there is a computer that called "Deep Thought" that finds the answer to life, the universe, and everything else. (Which is 42). A famous chess playing computer was named Deep Thought. IBM's Deep Blue is a take on that name and IBM's "corporate color" of blue.
More like a lack of documentation effort. I remember reading those exact same "download a binary and rename it" docs when I started in 1994! By now they are like 10 years out of date.
To install paradise 2000, which is on linux, you just unpack the tarball, copy a config file into your home dir, and run the program. Pick your server from the window that pops up. It has an INSTALL file that describes everything. It's a lot less trouble than installing quake and qspy and mesa libraries etc. I was going to create a RPM package, but this article caught me (and all other netrek developers) by surprise.
It's like the software and FTP list, which lists servers that dissapeared years ago, but not the new ones. If you look on the ftp.netrek.org site, clients like trekhopd and BRM are listed right along with COW, netrekxp and paradise-2000. There is no mention that some clients are from the late 80s and some from this century.
I've added speech to Paradise 2000, my pardise client for Linux. It's somewhat better than the amiga version. I have special case code for things like bombing messages, dooshes, kills, etc. The RCD macros can be configured too, for example you say "voice.generic: %?%S=SB%{help base&}" and you'll just get the message "help base" spoken too you instead of a line of numbers with the base's status.
I also think I've seen someone playing with Paradise 2.99 for Darwin, but I don't see the binary anywhere. As far as I know, no one has ported COW or BRMH to MacOS.
<plug mode>
You can read more and download the software from my web page for Paradise 2000, the ultimate Linux netrek client.
It has a nice sound system and can use IBM's ViaVoice for linux to do speech synthesis of messages and macros. Getting the IBM ViaVoice TTS package for linux is hard now, maybe /. should do a story on that.
</plug mode>
One problem Netrek has right now is lack of servers. The one popular server is often full. It also has had bad lag for most people recently, since it is a redhat and openoffice mirror, both which have released major new versions.
There are still some people playing netrek. If you use Linux, check out my client Paradise-2000. If you download IBM's viavoice package for linux, you can have it speak messages and macros outloud to you.
With the versions and default sessions I get with redhat 7.2, I measued the memory usage of KDE and gnome. KDE weighs in at a hefty 95 MB, while gnome uses "only" 41 MB. For comparison, the fvwm2 setup I use includes an email checker that tells me how many mesages I have, a clock, a loadgraph, cpu usage graphs (per CPU), button bar, and virtual desktop pager. The workings of the window manager itself are more configurable than either gnome or KDE. And all this is only 4MB! That's about 24 times less than KDE.
As to why more isn't avialable for free, it costs money to create data like this. One way or another, someone needs to pay for it. If taxes aren't enough to provide hi-res images of the entire world, then that leaves private enterprise.