I was thinking the same thing. Forget seeing outside the glasses. Increase the FOV and add a head tracker. Use independant feeds for the 2 displays and call Sony. I'm still waiting for the mass-market VR gear and games that are now practical.
The way to stop this nonsense is cleary to make wildass fun of it in a beer commercial. Remember all those idiots that used to say "what's up" in passing in the hallway? That all stopped after the WWHhhaazzzuuuuup beer commercials. If you make them feel like a jackass on a BEER commercial, they'll stop in short order. Or should I say: If you apply a poor quality rating to the suppliers deliverables in a widely accepted public forum, the resultant reconfiguration of the parameters is likely to change toward a positive outcome.
"Please re-read my comment within the context of the discussion at hand."
I did re-read your last post. I think I understand your point perfectly - you're concerned about what happens to your code that gets appended to anothers work, as opposed to someone improving upon that other work (a pure derivative). I think you understand the GPL side of things too. I just felt like trying the debating game with this topic today even though it's usually pointless. It's been fun. Thanks for that.
"I'd futher argue that when someone releases their code under the GPL because they are compelled to do so by the license that it hardly qualifies as "selfless" behavior."
No one is compelling you to do anything. If you chose to build on a GPLed work, you have chosen to provide people your code upon request when you distribute it (quite selfless). GPL doesn't force anyone to make that choice.
" I think you meant to say GPL instead of BSD in that last paragraph."
No, I really meant BSD. BSD allows people to deny others freedom. A little example:
A writes code and releases it under a license (either BSD or GPL).
B uses code from A in his project and provides binary to people(and source if A had chosen GPL).
C uses the software from B.
If A had used BSD license, B can deprived C of access to full source code for the software C is using (which is derived from or includes part of A). If A had chosen the GPL, C would be ensured those rights granted to B by A. GPL is about providing everyone the freedom to modify and redistribute the software they use, not just the middle-man (B). BSD offers B the option to deny C that freedom. GPL means B has to offer C the same freedom that B had. That's what I mean when I said the BSD license offers the freedom to deny peoples freedom. This is in fact a restriction the GPL places on B when he redistributes code. B is not allowed to deny C the same freedoms that B enjoys if A uses GPL.
"Under BSD-licensing, I can work at home on the same subject matter that I do at my place of employment. And you have the gall to tell me that the BSD is actually *more* restrictive?"
That's not a licensing difference. That's your employer being rather generous. Or perhaps not, since they can incorporate the work you do at home into the stuff you do at work - you've given them permission because your home stuff is BSD licensed? A lot of employers would think the stuff you do at home belongs to them - especially if it's "the same subject matter". They might just have a fit if think you're just working for free and then discover you're giving away half of their product. Unless all the right people in the company have signed off on this, you're on really thin ice.
"What if I think providing the source code for my project is stupid and a waste of space?"
Then you shouldn't release your code under a BSD or GPL license. People who use the GPL license do not want you to incorporate their code into such a work - i.e. they'd see that as being a leech.
"It's just denying that developer the freedom to choose an alternate license on the code that they wrote as a byproduct of using the original code."
It's denying that developer the freedom to choose an alternate license for the original GPLed code. Actually, it doesn't deny him anything (copyright law does) it just doesn't grant it where BSD does. He may choose whateven license he likes for his own code. If he combines them, then it's a derived work and the original author (of the GPLed part) is just ensuring that all derivatives of his code are kept free (in the sense that he sees it).
I should not have said anything. GPL vs BSD licenses is an argument that should not be started. Most people familiar with them understand the arguments on both sides and feel one or the other is "more free". That said, I will take one more shot at it;-)
"The BSD license does not deny anyone anything. You said it yourself "BSD allows". GPL disallows."
If you think using other peoples code is any sort of privlege or freedom, then BSD allows people to deny others that freedom. OK, the license doesn't deny peoples freedom - it just grants people the freedom to deny others that freedom. RMS phrased this very nicely in an interview, but I can't find it right now. It really is the crux of the debate, and it seems to be a very polarizing issue.
"If anything, licensing under BSD instead of the GPL is the most selfless act a software developer can make."
So you seem to think that giving away code is a "selfless act". And I'd be inclined to agree. If that's what you think, then GPL promotes further selfless behavior, while BSD allows the selfish to well... be selfish. Now as the author, releasing under which license offers more selflessness?
"Those who deny freedom to others deserve it not for themselves." -- Abraham Lincoln
"In other words, isnt' the GPL a restriction on the use of the SOURCE and not the resulting compiled binaries?"
The binaries are a derivative of the source and are therefore bound by the GPL. If you distribute binaries, you are required by GPL to offer full source code to those binaries. That means if you use unaltered GPLed source files along with some of your own, the full source code (including yours) must be offered under the GPL to people whom you give the binary.
"but doesn't the "Compiler Boundry Rule" serve as a rule-of-thumb..."
Almost. It's the "Linker Boundary Rule". Source files get turned into object files, which are then linked into an executable. The executable is a derivative of all the source files. The LGPL allows you to use dynamic linking (at runtime) to the GPLed code from non-GPLed code. LGPL was apparently a compromise, which I think has been deprecated by the FSF now that a critical mass of Free Software is available. People still use LGPL, but that's just to allow commercial interests to link their code (i.e. to stroke their ego with more users). I expect LGPL to go away very slowly (you can release derived code under regular GPL).
You are correct that you can include a GPLed database server on a CD with your own application and not have to GPL the application because they are really 2 seperate works that happen to come on the same CD.
"Also, remember matter in roughly uniform distribution in a a sphere (that you're outside of) acts exactly like all the mass was in the center of the sphere, and that a shell of matter of roughly even distribution that you're inside of has no net gravitational attraction, so the math isn't that complicated here."
A roughly disk shaped galaxy is very different from a uniform spherical distribution of mater. If we try to slice a galaxy like an onion, we get rings of matter - not shells. The gravitational force acting on a body inside a ring of uniform construction is NOT zero as it is inside a spherical shell of uniform density. Nor does a disk act like all its mass is concentrated in the center. This is another one of the mistaken assumptions of some of the dark matter proponents.
If you were refering to the dark matter sphere, then yes, you can adjust the density of the dark sphere as a function of radius to create any galactic rotation curve you like (with some limitations).
"It's obvious that the galactic rotation rates are wrong given only visible matter."
As I said before, the shape of the rotation curve should not taper off if there was only regular matter in a galaxy. That assumption has no basis in mathematics, yet it's the basis for the DM argument. You're right, the math isn't that hard, but Kepler doesn't apply and niether do uniform shells of matter.
My whole point is that the evidence for dark matter is based on some mathematical generalizations that don't carry over to multi-body disk shaped galaxies.
"A disk makes sense for matter that tends to form clumps as a result of collisions, and a sphere makes perfect sense for matter that doesn't. It's not all that unintuitive or surprising, given the data now in hand."
Based on what model does a sphere make perfect sense? Assuming dark matter exists, my intuition tells me it is either a) attracted to regular matter or b) attracted to itself. Otherwise it would not clump. If either of these is correct, a sphere is not the correct shape. If the DM is not rotating, the cloud would collapse and not have a large radius. If it is rotating, it will bulge in the middle and not remain a spherical distribution. I will maintain that there is nothing natural or intuitive about a spherical blob of dark matter with an embedded galaxy. I'd agree that there may be undetected "normal" matter in galaxies, but I'm not so sure that's needed either. I just wish people would stop claiming the expected rotation curve is based on Keplers laws and not Newtons with many bodies. Anyone who thinks 2 body Kepler equations are valid for entire galaxies needs is not qualified to be working in that area. As far as I can tell, this flawed assumption was (and still is) the basis for "dark matter".
"Besides the argument that stars orbiting around the fringes of galaxies appear to be moving too fast to stay in orbit without extra mass, what other observable evidence of dark matter is there?"
I looked into that whole thing. Most of the people who make that claim refer to Keplers laws of motion for orbiting bodies. If you assume the stars orbit a heavy core and don't interact with each other you get a galactic rotation curve that tapers off with radius. Real measured curves are nearly flat, so they conclude some "dark matter" that has some really unintuitive properties (see below). My own calculations of a rotation curve for a uniform flat disk of stars using interactions between all stars shows velocity increasing roughly linearly all the way out, and increasing even faster toward the edge. I don't think we should be suprised that observation lies somewhere between these two models. My distribution of stars is not accurate.
Stupid properties of dark matter: The interaction with regular matter must be asymetric. Why? Because they model it as a sphere of dark matter enclosing a disk shaped galaxy to get the expected rotation curve. If dark matter interacted with itself and visible matter in the same way visible matter interacts with itself, they should have the same distribution. I think they just observed that a big sphere of stuff would make their flawed model match reality and said "oh there must be this goofy thing here". Remember, there are NO direct observations of dark matter (or energy).
The tragedy of Einstein is that he convinced physicists that strange nonintuitive things are a part of the universe. This encourages the promotion of nifty off-the-wall sounding theories that make headlines to get funding.
I've said it here before: The only dark matter is between the astrophisicists ears.
What if I want to provide an open access point? What if a business (like Panera Bread) wants to provide one? Don't they just set it up and let people use it without explicit permission given? I know you can look up all their location on the net, so it's sort of advertised, but if you use their network are you notified that its OK? The default being promoted seems to be that if it's open you may use it.
There is a reason to think that too. If you need someones OK to use their network, how exactly are you supposed to get it? i.e. you need to access their network to know weather it's OK.
I think by RL standards you have a _very_ clear-cut case of pre-meditated theft.
I think given the number of places that deliberately provide free access (some whole cities) and the fact that WEP exists and is encouraged for security, we have a clear case of someone getting upset when they were too lazy to get a better understanding of how things work. They should also realize that they were lucky this guy wasn't hacking them, and that someone who was being malicious would probably not have been caught.
BTW, your involuntary sharing of food example is just stupid. Food is finite. Unused bandwith does not get used at a later time. Sharing bandwidth when you want to max it out IS annoying, and one should take steps to keep people out.
Well, there's the whole YRO section which usually complains about governments. And then articles complaining about the censorship that would be frowned upon right?
If you're refering to the Michigan poletown plant, Michigan has since switched its position on the issue. The state no longer force private->private transfers. That's not to say they couldn't start again. As a result of this supreme court case, one michigan rep is trying to get the state constitution ammended to disallow it by law in the future.
I just read more of their documentation. They sort an array of (key, pointer) pairs, where the pointer is to the rest of a record. Kind of like takeing structures and pulling the key value out into the tuple. This makes their sort more useful than I originally thought, but it is still not the same as qsort even when just using it for floats. The graphs don't specify which GPUsort they used (they do have one that isn't even a tuple version).
Implementing a similar "tuple sort" on a CPU with the same restrictions would be much faster than generic qsort. They claim the Intel C++ qsort is optimized using hyper threading and SSE instructions (they don't specify so I assume it's just a compiler setting), but I don't see that offering the same advantages of restricting it to a "tuple sort".
Theirs is more useful than I first thought, but I still think the CPU should be able to match it by imposing similar restrictions.
BTW, I'm also a strong believer in heapsort vs quicksort. Most comparisons I've seen show quicksort about 1.5x to 2x faster, but they also omit some optimizations of heapsort that I use. Heapsort is guaranteed to be O(n*log(n)) where quicksort is probably that fast, but degrades to O(n^2) if you're unlucky.
"Methinks your statement carries with it a load of unstated assumptions... care to spell them out for us?"
What I meant is that the GPU version is NOT generic like the qsort() CPU version. By a "less generic" version on the CPU I meant one that is optimized for a specific data type like "float". The standard quicksort does not sort a bunch of numbers, it sorts a list of pointers to things (like structs for example) when you provide it a pointer to a compare function that compares the things.
One can easily increase the performance of qsort by removing the generic function call to a compare funciton, and replacing it with a straight compare. That alone should make it run quite a bit faster (probably more than the 2x it's trailing in this comparison).
If the GPU sort is sorting actual data rather than pointers to data, it is even less generic than I assumed and the CPU version could be sped up again.
So my big assumption is that the GPU sort they implemented does not work on generic structures.
If it does sort pointers to structures (making an ass of me), and is only limited in the actual compare operation (i.e. no strings in the structures) then the CPU version can still be sped up by hard coding the offset (or passing it in) to a numeric element in a structure and removing the generic function call. The significance of this in terms of performance should not be underestimated, and would not be unfair since the GPU version has a similar restriction.
Optimizing for a particular case on a piece of hardware and then comparing it to unoptimized code on another piece of hardware is not a fair comparison.
To me, all they've done is show that the GPU really isn't as general purpose as the fanboys keep telling us. They didn't implement a generic sort because it was too hard. Even doing what they did is fairly tricky and required writing new code rather than running the same code on both platforms.
" MS is to big to become "irrelevent."......at this stage in the game I see MS sticking around for quite a long time."
Sure, Novell was still around when they decided to go the Linux route. I wouldn't have called them "relevant" at that point though - at least not in the way one thinks of MS today. BTW, my employer still uses Groupwise and some other Novell stuff. It takes a long time for something to go away completely, but that's doesn't mean it's still "relevant". SCO is still around too:-) For me personally, MS is already irrelevant - they have nothing to offer that I can't get a GPLed (read free in this context) equivalent for.
Relevance is probably an S-curve function of market share and MS is just not to the steep part yet.
"The implementation can handle both 16 and 32-bit floats."
So it's "hard coded" for a couple types. The standard qsort has you pass a function pointer to a comparison routine so it can sort anything. Standard qsort also sorts a list of pointers to the items - I bet this GPU sort works directly on the data. Implementing a less generic version on the CPU is likely to result in it being faster than the GPU sort.
The blurb at the end about increasing GPU speed with each generation is crap too. Both CPU and GPU performance are now limited by power dissipation issues.
It remains more effective to write stuff like this on the CPU rather than code for 2 different devices. Imposing the same limitations on generic C code will usually result in the same or better performance on the CPU. These "general purpose GPU" programs always seem to illustrate why the GPU is not general purpose.
I just run a pair of home-built speakers. Each has a 15" woofer, 6" (or is it 8?) midrange and two horns. All reasonably matched for power and frequency response. Amp is some old POS 80W per channel thing a friend gave me. It just kicks ass. Subwoofers tend to resonate at a particular frequency and just produce a thump rather than reproducing the correct sound. Big woofers respond down to ~20Hz correctly. The only thing that might be an improvement is having rears, but I don't have a theater setup - just music.
1) the rest of the blade is moving backward through the air (relative to the blade not the vehicle). See other responses to similar statements above.
2) Increasing the blade speed is only possible until its air speed = mach 1 and then things shake apart. Lots of helicopters can go over 100MPH, it's getting to 350MPH (mach 0.5) that's a problem. You have forward motion of 350mph and with a "stationary" blade tip on one side you get 700MPH blade tip on the other side. Then you're done.
Even fixed wing aircraft with a propeller keep their tip speed at or below about 0.9 mach. Above that speed really bad things start to happen.
" I stay away from such keyboards as Dvorak and "Natural"."
If by "Natural" you mean the funny Microsoft "broke in half" keyboard, I'd recommend you try it. I got one, and IMHO it's the only good thing Microsoft ever made. You don't need to learn anything, as the layout is still QWERTY. If you know how to type properly it's a no-brainer. If you've got some bad habbits like typing "Y" with the wrong hand, you'll be forced to fix that. It annoyed my wife, but didn't slow her down and she's a writer. I prefer the improved wrist position. YMMV but I'd recommend at least trying the thing for a week.
"I don't know which model it was, Cleo Bergssen or Nelia Cozza."
Right. Now if someone could produce the pictures from OUI we could just look and you'd know which one it is. Having written the first functional Cinematronics emulator, and having talked to both Tim and Scott, I've always wanted to see the original picture. At least it's a good excuse to check out some porn;-)
" I'm not sure what the point is that you're trying to make. But yes, I do end up passing 11-1 on the way to the weather"
I tried to say that even if the weather subchannel contains no commercials it is in the interest of the broadcaster to provide the service. FOX has a habbit of putting new shows in between existing popular shows to get people to take a look at them. They'd even shuffle the schedule to get you to take a look (I stopped watching the Simpsons because it kept moving around). Since you pass through 11-1 on your way to check the weather, there is a chance that something will catch your eye on the network. You may not be the type to channel surf, but I suspect it may help some people "discover" their other stuff. That's all.
Slashdot Mirror
I was thinking the same thing. Forget seeing outside the glasses. Increase the FOV and add a head tracker. Use independant feeds for the 2 displays and call Sony. I'm still waiting for the mass-market VR gear and games that are now practical.
The way to stop this nonsense is cleary to make wildass fun of it in a beer commercial. Remember all those idiots that used to say "what's up" in passing in the hallway? That all stopped after the WWHhhaazzzuuuuup beer commercials. If you make them feel like a jackass on a BEER commercial, they'll stop in short order. Or should I say: If you apply a poor quality rating to the suppliers deliverables in a widely accepted public forum, the resultant reconfiguration of the parameters is likely to change toward a positive outcome.
I did re-read your last post. I think I understand your point perfectly - you're concerned about what happens to your code that gets appended to anothers work, as opposed to someone improving upon that other work (a pure derivative). I think you understand the GPL side of things too. I just felt like trying the debating game with this topic today even though it's usually pointless. It's been fun. Thanks for that.
No one is compelling you to do anything. If you chose to build on a GPLed work, you have chosen to provide people your code upon request when you distribute it (quite selfless). GPL doesn't force anyone to make that choice.
No, I really meant BSD. BSD allows people to deny others freedom. A little example:
A writes code and releases it under a license (either BSD or GPL).
B uses code from A in his project and provides binary to people(and source if A had chosen GPL). C uses the software from B.
If A had used BSD license, B can deprived C of access to full source code for the software C is using (which is derived from or includes part of A). If A had chosen the GPL, C would be ensured those rights granted to B by A. GPL is about providing everyone the freedom to modify and redistribute the software they use, not just the middle-man (B). BSD offers B the option to deny C that freedom. GPL means B has to offer C the same freedom that B had. That's what I mean when I said the BSD license offers the freedom to deny peoples freedom. This is in fact a restriction the GPL places on B when he redistributes code. B is not allowed to deny C the same freedoms that B enjoys if A uses GPL.
B can of course opt not to use GPLed code from A.
That's not a licensing difference. That's your employer being rather generous. Or perhaps not, since they can incorporate the work you do at home into the stuff you do at work - you've given them permission because your home stuff is BSD licensed? A lot of employers would think the stuff you do at home belongs to them - especially if it's "the same subject matter". They might just have a fit if think you're just working for free and then discover you're giving away half of their product. Unless all the right people in the company have signed off on this, you're on really thin ice.
"What if I think providing the source code for my project is stupid and a waste of space?"
Then you shouldn't release your code under a BSD or GPL license. People who use the GPL license do not want you to incorporate their code into such a work - i.e. they'd see that as being a leech.
It's denying that developer the freedom to choose an alternate license for the original GPLed code. Actually, it doesn't deny him anything (copyright law does) it just doesn't grant it where BSD does. He may choose whateven license he likes for his own code. If he combines them, then it's a derived work and the original author (of the GPLed part) is just ensuring that all derivatives of his code are kept free (in the sense that he sees it).
I should not have said anything. GPL vs BSD licenses is an argument that should not be started. Most people familiar with them understand the arguments on both sides and feel one or the other is "more free". That said, I will take one more shot at it ;-)
"The BSD license does not deny anyone anything. You said it yourself "BSD allows". GPL disallows."
If you think using other peoples code is any sort of privlege or freedom, then BSD allows people to deny others that freedom. OK, the license doesn't deny peoples freedom - it just grants people the freedom to deny others that freedom. RMS phrased this very nicely in an interview, but I can't find it right now. It really is the crux of the debate, and it seems to be a very polarizing issue.
So you seem to think that giving away code is a "selfless act". And I'd be inclined to agree. If that's what you think, then GPL promotes further selfless behavior, while BSD allows the selfish to well... be selfish. Now as the author, releasing under which license offers more selflessness?
"Those who deny freedom to others deserve it not for themselves." -- Abraham Lincoln
GPL enforces that sentiment. BSD does not.
The binaries are a derivative of the source and are therefore bound by the GPL. If you distribute binaries, you are required by GPL to offer full source code to those binaries. That means if you use unaltered GPLed source files along with some of your own, the full source code (including yours) must be offered under the GPL to people whom you give the binary.
"but doesn't the "Compiler Boundry Rule" serve as a rule-of-thumb..."
Almost. It's the "Linker Boundary Rule". Source files get turned into object files, which are then linked into an executable. The executable is a derivative of all the source files. The LGPL allows you to use dynamic linking (at runtime) to the GPLed code from non-GPLed code. LGPL was apparently a compromise, which I think has been deprecated by the FSF now that a critical mass of Free Software is available. People still use LGPL, but that's just to allow commercial interests to link their code (i.e. to stroke their ego with more users). I expect LGPL to go away very slowly (you can release derived code under regular GPL).
You are correct that you can include a GPLed database server on a CD with your own application and not have to GPL the application because they are really 2 seperate works that happen to come on the same CD.
A roughly disk shaped galaxy is very different from a uniform spherical distribution of mater. If we try to slice a galaxy like an onion, we get rings of matter - not shells. The gravitational force acting on a body inside a ring of uniform construction is NOT zero as it is inside a spherical shell of uniform density. Nor does a disk act like all its mass is concentrated in the center. This is another one of the mistaken assumptions of some of the dark matter proponents.
If you were refering to the dark matter sphere, then yes, you can adjust the density of the dark sphere as a function of radius to create any galactic rotation curve you like (with some limitations).
"It's obvious that the galactic rotation rates are wrong given only visible matter."
As I said before, the shape of the rotation curve should not taper off if there was only regular matter in a galaxy. That assumption has no basis in mathematics, yet it's the basis for the DM argument. You're right, the math isn't that hard, but Kepler doesn't apply and niether do uniform shells of matter.
My whole point is that the evidence for dark matter is based on some mathematical generalizations that don't carry over to multi-body disk shaped galaxies.
Based on what model does a sphere make perfect sense? Assuming dark matter exists, my intuition tells me it is either a) attracted to regular matter or b) attracted to itself. Otherwise it would not clump. If either of these is correct, a sphere is not the correct shape. If the DM is not rotating, the cloud would collapse and not have a large radius. If it is rotating, it will bulge in the middle and not remain a spherical distribution. I will maintain that there is nothing natural or intuitive about a spherical blob of dark matter with an embedded galaxy. I'd agree that there may be undetected "normal" matter in galaxies, but I'm not so sure that's needed either. I just wish people would stop claiming the expected rotation curve is based on Keplers laws and not Newtons with many bodies. Anyone who thinks 2 body Kepler equations are valid for entire galaxies needs is not qualified to be working in that area. As far as I can tell, this flawed assumption was (and still is) the basis for "dark matter".
I looked into that whole thing. Most of the people who make that claim refer to Keplers laws of motion for orbiting bodies. If you assume the stars orbit a heavy core and don't interact with each other you get a galactic rotation curve that tapers off with radius. Real measured curves are nearly flat, so they conclude some "dark matter" that has some really unintuitive properties (see below). My own calculations of a rotation curve for a uniform flat disk of stars using interactions between all stars shows velocity increasing roughly linearly all the way out, and increasing even faster toward the edge. I don't think we should be suprised that observation lies somewhere between these two models. My distribution of stars is not accurate.
Stupid properties of dark matter: The interaction with regular matter must be asymetric. Why? Because they model it as a sphere of dark matter enclosing a disk shaped galaxy to get the expected rotation curve. If dark matter interacted with itself and visible matter in the same way visible matter interacts with itself, they should have the same distribution. I think they just observed that a big sphere of stuff would make their flawed model match reality and said "oh there must be this goofy thing here". Remember, there are NO direct observations of dark matter (or energy).
The tragedy of Einstein is that he convinced physicists that strange nonintuitive things are a part of the universe. This encourages the promotion of nifty off-the-wall sounding theories that make headlines to get funding.
I've said it here before: The only dark matter is between the astrophisicists ears.
There is a reason to think that too. If you need someones OK to use their network, how exactly are you supposed to get it? i.e. you need to access their network to know weather it's OK.
I think by RL standards you have a _very_ clear-cut case of pre-meditated theft.
I think given the number of places that deliberately provide free access (some whole cities) and the fact that WEP exists and is encouraged for security, we have a clear case of someone getting upset when they were too lazy to get a better understanding of how things work. They should also realize that they were lucky this guy wasn't hacking them, and that someone who was being malicious would probably not have been caught.
BTW, your involuntary sharing of food example is just stupid. Food is finite. Unused bandwith does not get used at a later time. Sharing bandwidth when you want to max it out IS annoying, and one should take steps to keep people out.
Well, there's the whole YRO section which usually complains about governments. And then articles complaining about the censorship that would be frowned upon right?
I would guess slashdot is one of the sites blocked in China. Is this correct? Anyone here from China?
If you're refering to the Michigan poletown plant, Michigan has since switched its position on the issue. The state no longer force private->private transfers. That's not to say they couldn't start again. As a result of this supreme court case, one michigan rep is trying to get the state constitution ammended to disallow it by law in the future.
Implementing a similar "tuple sort" on a CPU with the same restrictions would be much faster than generic qsort. They claim the Intel C++ qsort is optimized using hyper threading and SSE instructions (they don't specify so I assume it's just a compiler setting), but I don't see that offering the same advantages of restricting it to a "tuple sort".
Theirs is more useful than I first thought, but I still think the CPU should be able to match it by imposing similar restrictions.
BTW, I'm also a strong believer in heapsort vs quicksort. Most comparisons I've seen show quicksort about 1.5x to 2x faster, but they also omit some optimizations of heapsort that I use. Heapsort is guaranteed to be O(n*log(n)) where quicksort is probably that fast, but degrades to O(n^2) if you're unlucky.
What I meant is that the GPU version is NOT generic like the qsort() CPU version. By a "less generic" version on the CPU I meant one that is optimized for a specific data type like "float". The standard quicksort does not sort a bunch of numbers, it sorts a list of pointers to things (like structs for example) when you provide it a pointer to a compare function that compares the things.
One can easily increase the performance of qsort by removing the generic function call to a compare funciton, and replacing it with a straight compare. That alone should make it run quite a bit faster (probably more than the 2x it's trailing in this comparison).
If the GPU sort is sorting actual data rather than pointers to data, it is even less generic than I assumed and the CPU version could be sped up again.
So my big assumption is that the GPU sort they implemented does not work on generic structures.
If it does sort pointers to structures (making an ass of me), and is only limited in the actual compare operation (i.e. no strings in the structures) then the CPU version can still be sped up by hard coding the offset (or passing it in) to a numeric element in a structure and removing the generic function call. The significance of this in terms of performance should not be underestimated, and would not be unfair since the GPU version has a similar restriction.
Optimizing for a particular case on a piece of hardware and then comparing it to unoptimized code on another piece of hardware is not a fair comparison.
To me, all they've done is show that the GPU really isn't as general purpose as the fanboys keep telling us. They didn't implement a generic sort because it was too hard. Even doing what they did is fairly tricky and required writing new code rather than running the same code on both platforms.
Sure, Novell was still around when they decided to go the Linux route. I wouldn't have called them "relevant" at that point though - at least not in the way one thinks of MS today. BTW, my employer still uses Groupwise and some other Novell stuff. It takes a long time for something to go away completely, but that's doesn't mean it's still "relevant". SCO is still around too :-) For me personally, MS is already irrelevant - they have nothing to offer that I can't get a GPLed (read free in this context) equivalent for.
Relevance is probably an S-curve function of market share and MS is just not to the steep part yet.
So it's "hard coded" for a couple types. The standard qsort has you pass a function pointer to a comparison routine so it can sort anything. Standard qsort also sorts a list of pointers to the items - I bet this GPU sort works directly on the data. Implementing a less generic version on the CPU is likely to result in it being faster than the GPU sort.
The blurb at the end about increasing GPU speed with each generation is crap too. Both CPU and GPU performance are now limited by power dissipation issues.
It remains more effective to write stuff like this on the CPU rather than code for 2 different devices. Imposing the same limitations on generic C code will usually result in the same or better performance on the CPU. These "general purpose GPU" programs always seem to illustrate why the GPU is not general purpose.
I just run a pair of home-built speakers. Each has a 15" woofer, 6" (or is it 8?) midrange and two horns. All reasonably matched for power and frequency response. Amp is some old POS 80W per channel thing a friend gave me. It just kicks ass. Subwoofers tend to resonate at a particular frequency and just produce a thump rather than reproducing the correct sound. Big woofers respond down to ~20Hz correctly. The only thing that might be an improvement is having rears, but I don't have a theater setup - just music.
Favorite test/demo track: The Launch by Boston
2) Increasing the blade speed is only possible until its air speed = mach 1 and then things shake apart. Lots of helicopters can go over 100MPH, it's getting to 350MPH (mach 0.5) that's a problem. You have forward motion of 350mph and with a "stationary" blade tip on one side you get 700MPH blade tip on the other side. Then you're done.
Even fixed wing aircraft with a propeller keep their tip speed at or below about 0.9 mach. Above that speed really bad things start to happen.
If by "Natural" you mean the funny Microsoft "broke in half" keyboard, I'd recommend you try it. I got one, and IMHO it's the only good thing Microsoft ever made. You don't need to learn anything, as the layout is still QWERTY. If you know how to type properly it's a no-brainer. If you've got some bad habbits like typing "Y" with the wrong hand, you'll be forced to fix that. It annoyed my wife, but didn't slow her down and she's a writer. I prefer the improved wrist position. YMMV but I'd recommend at least trying the thing for a week.
Right. Now if someone could produce the pictures from OUI we could just look and you'd know which one it is. Having written the first functional Cinematronics emulator, and having talked to both Tim and Scott, I've always wanted to see the original picture. At least it's a good excuse to check out some porn ;-)
I tried to say that even if the weather subchannel contains no commercials it is in the interest of the broadcaster to provide the service. FOX has a habbit of putting new shows in between existing popular shows to get people to take a look at them. They'd even shuffle the schedule to get you to take a look (I stopped watching the Simpsons because it kept moving around). Since you pass through 11-1 on your way to check the weather, there is a chance that something will catch your eye on the network. You may not be the type to channel surf, but I suspect it may help some people "discover" their other stuff. That's all.