It's to indoctrinate them when they're young to the concept of Big Brother tracking their every move. Then when they they get to be voting age, they'll be more receptive to legislation requiring everyone to have an RFID tag implanted.
Seriously, I'm not joking at all. If you've been carrying an RFID tag as long as you can remember, requiring it by law won't seem like a big deal at all. Laws that take away freedoms are preceded by education campaigns to convince the public they want to give the freedom up.
The highest gear they will have on their normal bike is a 53x11. That's 53 in the front and 11 on the back. This is pretty normal for a road bike actually. Everyone wants the same stuff Lance uses, even though an 11 tooth cog is useless to most riders. Search google groups for rec.bicycles.tech and "11 tooth" for a few Holy Wars on the subject.
Cassettes, the cluster of cogs on the back, are pretty much a single unit and the manufacturers only make a few different sizes. Shimano Dura-Ace 10 speed (10 speeds in the rear, with two or three chainrings in the front for 20 or 30 speeds total) comes in six versions, with small cogs of 11 or 12 teeth and larges from 21 to 27.
On a good sized hill, you can easily go faster than 30 mph just by coasting. In fact, on a fast hill you can go often go faster by not pedaling and getting more aerodynamic than by trying to pedal faster. Much of the argument of why "nobody need an 11 tooth cog" is based on that.
If someone wants to go faster, they might also try getting the RPMs they can pedal at higher. Toe clips or fancy clipless pedals make a huge difference. Having the saddle too low is a very common miss-adjustment. You'll see lots of people riding around with the knees sticking out to the side as the pedal because they're too low. 75 to 90 RPM is a good average to ride at, but people who don't know any better will use 50-60 RPM, which is close to walking cadence. According to my bicycle computer, my average for the last few months of commuting is 86 RPM and my max is 197 RPM.
First of all, the human eye isn't the most sentitive to blue light, it much more sentitive to green. You can see the human eye's response curve here
and a breakdown of color vs wavelength.
Secondly, the lumen or candela rating already takes this into account. At the peak of photopic vision, 555 nm (green), there are 683 lumens per watt. If you had one watt of blue light, it would only be about 100 lumens, because the human eye is less sensitive to that wavelength.
In other words, one watt of green light appears brighter than one watt of blue light, because humans are more sensitive to that color. One lumen of green light is just as bright as one lumen on blue light, because the lumen measurement takes this into account. That's the whole point of lumens, they are watts times luminous efficacy for human vision.
They used both a DC-DC supply and a LM317 converter. The DC-DC supply was a pre-made thing that takes in something like 12V and supplies what you need for an ATX motherboard, 12V, 5V, 3.3V, -3.3V, etc. Lots of low-power and small form factor computers use them, as they dissipate less heat and are smaller than normal AC-DC switching supplies.
Then they used a LM317 linear regulator to provide 7V from 12V for the wireless bridge. The bridge's power supply was rated for 1.4 amps. A 5V drop over 1.4 amps is 7 Watts wasted as heat. Not impressive in something that's supposed to be low power. With their 20 DegC/watt heatsink, that is a 140 deg C temp rise from ambient to the TO-220 case. Easily out of spec. The wireless bridge surely doesn't draw the full 1.4A in operation, that's what saves them from their regulator melting.
They probably should have seen if their bridge would have run from the 5V line or if the bridge's power supply could handle 12V. They probably didn't even need the 7V regulator.
If they did need one, it would have been much better to use a switching regulator that would be around 95% efficient instead of their 58% efficient linear regulator. TI powertrends makes integrated switching regulator modules that would work perfectly. I used one for to power a digital camera off a 12V deep cycle battery so it could take photos on a timer for several days straight.
I also have to wonder why they used locking switches instead of just getting a case with a locking front panel. I've got a rack of such 2U cases at work, they're not hard to come by.
The switch from the 2nd hard drive's power was unecessary too, you can turn the drive off with software.
SPECfp is general compiled code. Both the Top500 benchmark and a lot of scientific
applications rely solely on matrix multiplications.
Ok, let's look at top500 and see what kind of LINPACK performance you get.
LANL has a 2816 processor Opteron 2GHz cluster, with Myrinet, with a Rmax of 8051.
PNNL has a 1936 processors Itanium 2 1.5Ghz cluster, with Quadrics, with an Rmax of 8633.
Per processor, that's 2.85 GFLOPS for the opteron vs 4.45 for the Itanium2. Scale the Opteron 246 to a 248 and you should get 3.135 GFLOPs. That puts the Itanium2 at 42% faster, per CPU.
Now look at the cost. Those Opteron nodes cost about 1/4 as much as the Itanium2 nodes. For half the money, you could buy an Opteron cluster twice the size, and get more overall performance. That's what matters in the end, what total performance will you get for your budget.
I wouldn't say Infiniband is the popular choice right now. Other than VT's G5 cluster, there really isn't anyone else using it. For Beowulf interconnects, the most popular seems to be Myrinet for smaller systems and Quadrics for the largest systems. LLNL used Quadrics' QsNetII for their cluster, and you can get the price list for it. A 1024 node interconnect was just over $4 million, about $4k per node.
The thing is, if you look at the price difference of an Opteron system vs Itanium2 system, and even when you factor in a pricey interconnect like QsNetII in addition to the smaller per-node costs like cases, power supplies, etc., the Opteron still comes out ahead by a large margin.
How is it apples and oranges? Why would I build a cluster with four way opterons when I could get a faster cluster using two way? If there was a $670 Itanium2 processor for clusers, I'd use that. But there isn't, and you have to look at what exists. The fact is, Itanium just isn't very impressive for a cluster. If you think Itanium is so great, show me some numbers. Calling people names just shows you don't have any facts to back up your argument.
Are you saying that clusters built with Opteron 248 chips don't work, because they're not "Big Iron" chips? That's absurd.
All this business about what is "big iron" and what is workstation, what is designed for floating point performance, what has an optimized instruction set, and so on is irrelevant. It all comes down to GFLOPS, dollars, watts, and floor space. All the rest is just meaningless marketing buzzwords of interest to no one but fan boys with an agenda that isn't supported by real numbers.
I don't know if I would call NUMALink a true shared memory system. It is NUMA after all! I was thinking of 32-64 way machines with a true shared memory system, or large vector machines based on SX-6 processors for example.
But look at NUMALink4, its got 6.4 GB/sec per link bandwith and 240ns latency.
QsNetII is just under 1 GB/sec bandwidth, the limit of PCI-X, with a latency of 3us.
So, NUMALink4 has 6.4 times the badwidth and 12.5 times less latency than QsNetII. That a much larger performance difference than Opteron vs Itanium!
I wish those graphs were easier to read! From the looks of it, at ~32 processors the 1.5 GHz It2 has a 12 second run time and the 1.8 GHz Opteron is 24 seconds. The It2 cluster has Quadrics high end Elan4 interconnect, while the interconnect of the Opterion cluster isn't listed. It might have GigE for all we know.
The It2 probably cost around 5 times as much as the opterons, so a real comparison would be 32 It2 processors vs 160 Opterons. With the scaling shown for that model, the Opterons of equilivent cost would be 2-3 times faster than the Itaniums.
Ok, checked them again. The best 1.5 GHz Itanium2 SPECfp2000 score is 2148 while the opteron 248 is 1691. That's 27% faster. I'd hardly call that smoked.
The Opteron 248 is $670 on pricewatch, while the 1.5 GHz It2 is $5200! The motherboards are like $1400 vs $400.
You have to keep in mind that this isn't a single machine, it's a cluster. You could take the money spent on an Itanium2 cluster, and buy an opteron cluster with five times as many processors. I am well aware that one does not get perfect scaling. But if you are running something on a cluster in the first place, I have a hard time imagining something that is faster with one fifth as many 27% faster processors. Yes, there are codes that would be faster on 1000 Itanium2 vs 5000 Opterons, but you would never runs these on cluster, because they would be faster still shared memory system.
You can get hydrogen from water, sorry you missed that one in 1st grade science.
Water just doesn't break apart into oxygen and hydrogen on it's own, it takes energy! More energy that you get back when you burn the hydrogen by recombining it with oxygen.
I think you've got that backwards, Quadrics is the performance leading, not the price/performance leader. Myrinet, SCI, and Infiniband all beat it in price/performance. Quadrics is faster, and scales to more nodes than the others.
According to Quadrics latest price list, the cards are $1200 each, $913 per port for a 64 node switch, and $185-$265 for a cable. That's $2300/node.
Myrinet cards are $595, the switch is $400 per port for 64 nodes, and the cables are ~$50. That's $1050/node.
Quadric's price for a 1024 node interconnect is $4,176,094. That's hardly chump change. The bandwith is about 10x higher than gigabit ethernet, and the latency about 100x lower.
Do you have any kind of benchmark where the Itanium smokes the Opteron? The Itanium does have a greater memory bandwidth, but not by a lot. If you look at the spec benchmarks, it can be faster on some of them, but not by a lot. However, the Itamium is a lot more expensive!
Compared to a Xeon or AthlonMP cluster, the Itanium faired poorly in price/performance. The only reason to use Itaniums was if you needed 64 bits for more than 4GB of memory, or needed high single CPU performance for a pooly parallized application. (Of course if your application parallizes poorly, a cluster is probably a bad choice to begin with). Then Opterion came out and changed all that. It's 64 bits, it's fast, and it's a fraction of the price of the Itanium2.
I just purchased a new Beowulf cluster. The decision was between Xeons vs Opterons. The Opterons had better price/performance, but the Xeons would fit in better with our existing Pentium3 Beowulf, other ia32 servers, and existing software. In the end, we went with Opterons. Itanium2 was never even in contention. Just one look at the price and performce of a Itanium2 system was all it took to cross it of the list.
The 386SX could still run 32bit code. It had a 16 bit data bus, and fewer address lines as well I think. But it could run Windows/386 (remember that?) and use "386 Enhanced mode", while a 286, which was both cheaper and faster than a 386SX, couldn't.
No reason you can't call fortran routines from LAPACK or whatnot from a C program.
Fortran was the first high level programming language. It was created to write scientific programs in. Hence, the some of the oldest programs around are scientific programs written in fortran. That's why scientific code has this huge legacy of fortran, that you don't see elsewhere.
It seems like your trying to say that the C compiler must produce 16 bit code for a 386, but the java JIT compiler will produce 32 bit code.
The problem with that is that the 386 is 32 bit! You're comparing a C compiler making 16 bit code for a 286 to a java compiler for a 32 bit platform. Unless you know of a java runtime that works on 286 or worse processor, that's not a fair comparison.
It's still silly anyways. Compilers can produce code tuned for different CPUs. There is no need to compile for the lowest common denominator. When I compile scientific programs at work, I sure as hell don't compile for a 286. I don't even have a compiler than can produce 16 bit code! I always tune the compile for the specific CPUs.
It's funny, but I just rackmounted a beowulf cluster at work that was built in those things. It had 19 mid-tower cases, one two of those wire racks. One of the racks had two 1000VA and two 1400VA UPSes on the bottom.
The wire racks were nice, and didn't cost much. I was able to run all the power and ethernet cables though the shelves. The spacing on the wires is just enough to fit the small end of power cord though.
Our server room is full, and we wanted to get some more equpiment in it. So I figured, rackmount our existing cluster. I bought a bunch of 2U rackmount cases with no power supply for $125, a 42U enclosed cabinet for about $1k, and another $700 for sliding rails to mount the cases on. The servers use the old slot 1 pentium3 CPUs, which won't fit in a 1U case. I re-used the power supplies out of the mid-tower cases to save money.
The cabinet takes about 2/3 the floor space of one of the wire racks, and the cluster was using two wire racks before. It's also a lot easier to slide the servers out on the sliding rails than it was to get those stuff off the shelves. Sure looks nicer too.
What you're talking about isn't fair use. Making copies for personal use to change formats is fair use. Make copies to distribute for money is not fair use. It's a very obvious distinction.
I had to work for two months with no cubicle after our office moved into a new building, so everyone else could finish using all the furniture they wanted. Then I got to build my own out of the leftover scraps no one else wanted.
Seemed pretty nice after having to work laying down on the floor or putting my keyboard on my lap and the monitor on a small bookshelf. Forget Aeron chairs, I remember the day I got a chain with ARMS. They weren't padded and I couldn't adjust the height, but at least I had something to rest my elbows on.
I once was employed by an insurance product publishing company in indianapolis. The project manager
made all his decisions with rock-paper-scissors. I'm not making this up. Whenever a bug needed to be
fixed, he would call all the developers into a room and play RPS until there was a loser.
Doesn't sound like such a bad idea to me. I've been in lots of situations where multiple developers will think their way is the one true way, and argue ad nauseum about it. In truth, none of the ideas are clearly the best, and just picking something and being done with it will save a lot of time and aggrevation.
The imaged based "read the letters" CAPTCHAs that don't have any kind of semantics question are generated by a computer. You can have a program spit out thousands of unique images a second, from a set of billions, there's no way a human could answer all of them. That's why these systems are used, because it's easy to have a computer make up new unique questions. But these kinds of questions are not that hard for a computer to answer.
So you say, "use a semantics question, those are harder for a computer to answer." And sure they are, but you have a new problem. It's hard for computers to generate them too! You had a human generating the new questions, which means that it's now possible for a human to find the answers to all the questions. By having a human come up with all the questions, you've lost the CA part of CAPTCHA.
It's kind of a catch-22. Questions that are easy for a computer to generate, are easy for a computer to answer. Questions that are hard for a computer to answer, are hard for a computer to generate.
Someone could just spend five minutes figuring out all the answers to your fixed set of 20-30 questions . Answering your questions is probably easier than making up in the first place.
Slashdot Mirror
Seriously, I'm not joking at all. If you've been carrying an RFID tag as long as you can remember, requiring it by law won't seem like a big deal at all. Laws that take away freedoms are preceded by education campaigns to convince the public they want to give the freedom up.
The highest gear they will have on their normal bike is a 53x11. That's 53 in the front and 11 on the back. This is pretty normal for a road bike actually. Everyone wants the same stuff Lance uses, even though an 11 tooth cog is useless to most riders. Search google groups for rec.bicycles.tech and "11 tooth" for a few Holy Wars on the subject.
Cassettes, the cluster of cogs on the back, are pretty much a single unit and the manufacturers only make a few different sizes. Shimano Dura-Ace 10 speed (10 speeds in the rear, with two or three chainrings in the front for 20 or 30 speeds total) comes in six versions, with small cogs of 11 or 12 teeth and larges from 21 to 27.
On a good sized hill, you can easily go faster than 30 mph just by coasting. In fact, on a fast hill you can go often go faster by not pedaling and getting more aerodynamic than by trying to pedal faster. Much of the argument of why "nobody need an 11 tooth cog" is based on that.
If someone wants to go faster, they might also try getting the RPMs they can pedal at higher. Toe clips or fancy clipless pedals make a huge difference. Having the saddle too low is a very common miss-adjustment. You'll see lots of people riding around with the knees sticking out to the side as the pedal because they're too low. 75 to 90 RPM is a good average to ride at, but people who don't know any better will use 50-60 RPM, which is close to walking cadence. According to my bicycle computer, my average for the last few months of commuting is 86 RPM and my max is 197 RPM.
I'm almost certain The Right Stuff did have an intermission, at least when I saw it.
First of all, the human eye isn't the most sentitive to blue light, it much more sentitive to green. You can see the human eye's response curve here and a breakdown of color vs wavelength.
Secondly, the lumen or candela rating already takes this into account. At the peak of photopic vision, 555 nm (green), there are 683 lumens per watt. If you had one watt of blue light, it would only be about 100 lumens, because the human eye is less sensitive to that wavelength.
In other words, one watt of green light appears brighter than one watt of blue light, because humans are more sensitive to that color. One lumen of green light is just as bright as one lumen on blue light, because the lumen measurement takes this into account. That's the whole point of lumens, they are watts times luminous efficacy for human vision.
They used both a DC-DC supply and a LM317 converter. The DC-DC supply was a pre-made thing that takes in something like 12V and supplies what you need for an ATX motherboard, 12V, 5V, 3.3V, -3.3V, etc. Lots of low-power and small form factor computers use them, as they dissipate less heat and are smaller than normal AC-DC switching supplies.
Then they used a LM317 linear regulator to provide 7V from 12V for the wireless bridge. The bridge's power supply was rated for 1.4 amps. A 5V drop over 1.4 amps is 7 Watts wasted as heat. Not impressive in something that's supposed to be low power. With their 20 DegC/watt heatsink, that is a 140 deg C temp rise from ambient to the TO-220 case. Easily out of spec. The wireless bridge surely doesn't draw the full 1.4A in operation, that's what saves them from their regulator melting.
They probably should have seen if their bridge would have run from the 5V line or if the bridge's power supply could handle 12V. They probably didn't even need the 7V regulator.
If they did need one, it would have been much better to use a switching regulator that would be around 95% efficient instead of their 58% efficient linear regulator. TI powertrends makes integrated switching regulator modules that would work perfectly. I used one for to power a digital camera off a 12V deep cycle battery so it could take photos on a timer for several days straight.
I also have to wonder why they used locking switches instead of just getting a case with a locking front panel. I've got a rack of such 2U cases at work, they're not hard to come by.
The switch from the 2nd hard drive's power was unecessary too, you can turn the drive off with software.
Ok, let's look at top500 and see what kind of LINPACK performance you get.
LANL has a 2816 processor Opteron 2GHz cluster, with Myrinet, with a Rmax of 8051.
PNNL has a 1936 processors Itanium 2 1.5Ghz cluster, with Quadrics, with an Rmax of 8633.
Per processor, that's 2.85 GFLOPS for the opteron vs 4.45 for the Itanium2. Scale the Opteron 246 to a 248 and you should get 3.135 GFLOPs. That puts the Itanium2 at 42% faster, per CPU.
Now look at the cost. Those Opteron nodes cost about 1/4 as much as the Itanium2 nodes. For half the money, you could buy an Opteron cluster twice the size, and get more overall performance. That's what matters in the end, what total performance will you get for your budget.
I wouldn't say Infiniband is the popular choice right now. Other than VT's G5 cluster, there really isn't anyone else using it. For Beowulf interconnects, the most popular seems to be Myrinet for smaller systems and Quadrics for the largest systems. LLNL used Quadrics' QsNetII for their cluster, and you can get the price list for it. A 1024 node interconnect was just over $4 million, about $4k per node.
The thing is, if you look at the price difference of an Opteron system vs Itanium2 system, and even when you factor in a pricey interconnect like QsNetII in addition to the smaller per-node costs like cases, power supplies, etc., the Opteron still comes out ahead by a large margin.
How is it apples and oranges? Why would I build a cluster with four way opterons when I could get a faster cluster using two way? If there was a $670 Itanium2 processor for clusers, I'd use that. But there isn't, and you have to look at what exists. The fact is, Itanium just isn't very impressive for a cluster. If you think Itanium is so great, show me some numbers. Calling people names just shows you don't have any facts to back up your argument.
Are you saying that clusters built with Opteron 248 chips don't work, because they're not "Big Iron" chips? That's absurd.
All this business about what is "big iron" and what is workstation, what is designed for floating point performance, what has an optimized instruction set, and so on is irrelevant. It all comes down to GFLOPS, dollars, watts, and floor space. All the rest is just meaningless marketing buzzwords of interest to no one but fan boys with an agenda that isn't supported by real numbers.
I don't know if I would call NUMALink a true shared memory system. It is NUMA after all! I was thinking of 32-64 way machines with a true shared memory system, or large vector machines based on SX-6 processors for example.
But look at NUMALink4, its got 6.4 GB/sec per link bandwith and 240ns latency.
QsNetII is just under 1 GB/sec bandwidth, the limit of PCI-X, with a latency of 3us.
So, NUMALink4 has 6.4 times the badwidth and 12.5 times less latency than QsNetII. That a much larger performance difference than Opteron vs Itanium!
I wish those graphs were easier to read! From the looks of it, at ~32 processors the 1.5 GHz It2 has a 12 second run time and the 1.8 GHz Opteron is 24 seconds. The It2 cluster has Quadrics high end Elan4 interconnect, while the interconnect of the Opterion cluster isn't listed. It might have GigE for all we know.
The It2 probably cost around 5 times as much as the opterons, so a real comparison would be 32 It2 processors vs 160 Opterons. With the scaling shown for that model, the Opterons of equilivent cost would be 2-3 times faster than the Itaniums.
Ok, checked them again. The best 1.5 GHz Itanium2 SPECfp2000 score is 2148 while the opteron 248 is 1691. That's 27% faster. I'd hardly call that smoked.
The Opteron 248 is $670 on pricewatch, while the 1.5 GHz It2 is $5200! The motherboards are like $1400 vs $400.
You have to keep in mind that this isn't a single machine, it's a cluster. You could take the money spent on an Itanium2 cluster, and buy an opteron cluster with five times as many processors. I am well aware that one does not get perfect scaling. But if you are running something on a cluster in the first place, I have a hard time imagining something that is faster with one fifth as many 27% faster processors. Yes, there are codes that would be faster on 1000 Itanium2 vs 5000 Opterons, but you would never runs these on cluster, because they would be faster still shared memory system.
You can get hydrogen from water, sorry you missed that one in 1st grade science.
Water just doesn't break apart into oxygen and hydrogen on it's own, it takes energy! More energy that you get back when you burn the hydrogen by recombining it with oxygen.
I think you've got that backwards, Quadrics is the performance leading, not the price/performance leader. Myrinet, SCI, and Infiniband all beat it in price/performance. Quadrics is faster, and scales to more nodes than the others.
According to Quadrics latest price list, the cards are $1200 each, $913 per port for a 64 node switch, and $185-$265 for a cable. That's $2300/node.
Myrinet cards are $595, the switch is $400 per port for 64 nodes, and the cables are ~$50. That's $1050/node.
Quadric's price for a 1024 node interconnect is $4,176,094. That's hardly chump change. The bandwith is about 10x higher than gigabit ethernet, and the latency about 100x lower.
Do you have any kind of benchmark where the Itanium smokes the Opteron? The Itanium does have a greater memory bandwidth, but not by a lot. If you look at the spec benchmarks, it can be faster on some of them, but not by a lot. However, the Itamium is a lot more expensive!
Compared to a Xeon or AthlonMP cluster, the Itanium faired poorly in price/performance. The only reason to use Itaniums was if you needed 64 bits for more than 4GB of memory, or needed high single CPU performance for a pooly parallized application. (Of course if your application parallizes poorly, a cluster is probably a bad choice to begin with). Then Opterion came out and changed all that. It's 64 bits, it's fast, and it's a fraction of the price of the Itanium2.
I just purchased a new Beowulf cluster. The decision was between Xeons vs Opterons. The Opterons had better price/performance, but the Xeons would fit in better with our existing Pentium3 Beowulf, other ia32 servers, and existing software. In the end, we went with Opterons. Itanium2 was never even in contention. Just one look at the price and performce of a Itanium2 system was all it took to cross it of the list.
They didn't stop in the middle, they stopped one episode from the end.
It was run a second time, but got canned around the 2/3s mark.
Reminds me I need to go the rent the DVDs and finish watching it, except now I've forgotten what was happening.
If living with it the rest of his life is such a punishment, then why doesn't he kill himself?
He has a choice, the kid he murdered doesn't.
The 386SX could still run 32bit code. It had a 16 bit data bus, and fewer address lines as well I think. But it could run Windows/386 (remember that?) and use "386 Enhanced mode", while a 286, which was both cheaper and faster than a 386SX, couldn't.
No reason you can't call fortran routines from LAPACK or whatnot from a C program.
Fortran was the first high level programming language. It was created to write scientific programs in. Hence, the some of the oldest programs around are scientific programs written in fortran. That's why scientific code has this huge legacy of fortran, that you don't see elsewhere.
It seems like your trying to say that the C compiler must produce 16 bit code for a 386, but the java JIT compiler will produce 32 bit code.
The problem with that is that the 386 is 32 bit! You're comparing a C compiler making 16 bit code for a 286 to a java compiler for a 32 bit platform. Unless you know of a java runtime that works on 286 or worse processor, that's not a fair comparison.
It's still silly anyways. Compilers can produce code tuned for different CPUs. There is no need to compile for the lowest common denominator. When I compile scientific programs at work, I sure as hell don't compile for a 286. I don't even have a compiler than can produce 16 bit code! I always tune the compile for the specific CPUs.
It's funny, but I just rackmounted a beowulf cluster at work that was built in those things. It had 19 mid-tower cases, one two of those wire racks. One of the racks had two 1000VA and two 1400VA UPSes on the bottom.
The wire racks were nice, and didn't cost much. I was able to run all the power and ethernet cables though the shelves. The spacing on the wires is just enough to fit the small end of power cord though.
Our server room is full, and we wanted to get some more equpiment in it. So I figured, rackmount our existing cluster. I bought a bunch of 2U rackmount cases with no power supply for $125, a 42U enclosed cabinet for about $1k, and another $700 for sliding rails to mount the cases on. The servers use the old slot 1 pentium3 CPUs, which won't fit in a 1U case. I re-used the power supplies out of the mid-tower cases to save money.
The cabinet takes about 2/3 the floor space of one of the wire racks, and the cluster was using two wire racks before. It's also a lot easier to slide the servers out on the sliding rails than it was to get those stuff off the shelves. Sure looks nicer too.
What you're talking about isn't fair use. Making copies for personal use to change formats is fair use. Make copies to distribute for money is not fair use. It's a very obvious distinction.
Seemed pretty nice after having to work laying down on the floor or putting my keyboard on my lap and the monitor on a small bookshelf. Forget Aeron chairs, I remember the day I got a chain with ARMS. They weren't padded and I couldn't adjust the height, but at least I had something to rest my elbows on.
So you say, "use a semantics question, those are harder for a computer to answer." And sure they are, but you have a new problem. It's hard for computers to generate them too! You had a human generating the new questions, which means that it's now possible for a human to find the answers to all the questions. By having a human come up with all the questions, you've lost the CA part of CAPTCHA.
It's kind of a catch-22. Questions that are easy for a computer to generate, are easy for a computer to answer. Questions that are hard for a computer to answer, are hard for a computer to generate.
Someone could just spend five minutes figuring out all the answers to your fixed set of 20-30 questions . Answering your questions is probably easier than making up in the first place.