Don't get me wrong. I'm not advocating war. I'm advocating a counter-terrorist strike or twenty, targeted at known terrorist groups, and removing just those people, with as much care and good planning as they showed this morning.
There's something fundamentally wrong with this line of reasoning, which is mostly what prompted me to write the message at the head of this thread. You are advocating a counter strike. Let's assume you are successful and do actually cause some damage to these groups? Do you think they'll stand there a do nothing about it? One of the things today's attack proved is that this is doable. These people are hoping the U.S.A. counter strikes. This is not war against a country. There's no country to declare war to (at least not atm, but I don't think this will change). This is terrorism. You can only target groups, but that's the problem. It's not like you bomb a building and that's it. To put this in terms people in this group can understand, it's just like MS trying to attack Linux. They can't because there's no single entity where your attack can be directed to. Think about what the U.S.A. did in Panama, and notice there are clear differences. You could break into some guy's house and bring him to the U.S.A. and prosecute and excecute him there, but you still have a group.
You also have to notice something else: the purpose of the attack is to enrage, to create chaos, to make the people claim for revenge. This is not a military installation that was attacked, that probably would have failed. The target was a civilian target because that stood a chance of success. This attack is bait. Don't swallow it.
And you implicitely noted something else: this is the first time is a lot of years that war has been brought to U.S.A.'s backyard, which is perhaps the reason why people in other parts of the world see it differently: they still remember having war on their backyards.
Shit. Kabul, Afghanistan, seems to be under missile attack.
For the record: Negotiating with terrorists DOES NOT WORK. It's like negotiating with a hydra: Set peace with one, and the others will still attack. You're making the broad and stupid assumption that there is such a thing as a "terrorist organization" with rules that all of the terrorists follow.
[...]
I'm not finding a lot of historical precedent that bodes well for the safety of the US if we don't track these folks down and wipe them out.
So, I'm the one making the stupid assumption, uh? Which country are you going to declare war against? War is an armed conflict between two sovereign nations. Precisely because this is a terrorist act you can't point your guns to some country and be done with it. In doing so, the U.S.A. will be drawn into starting a war, much like you were drawn into taking parts in a war because of Pearl Harbor. Justice is to be made, but war is not the way.
Not responding to this attack will not lead to peace.
But a military attack on the suspects won't either. Just look at this forum which is supposed to have lots of highly educated people that can really assess the benefits derived from U.S.A.'s previous military responses. They are roaring for blood.
I never said this has to go unanswered. But wiping some country out of the map isn't the correct answer. Today fifty thousand died. How many does it take to get even? Fifty thousand, too? Five hundred thousand? Five million? All of them?
FWIW, I'm not a citizen of the U.S. nor do I live there.
Violence induces more violence. Retaliation will only lead to more deaths. If you are a citizen of the U.S. of America, please write your representative right now and ask him to join a plea for peace. Historically the U.S. reaction to this kind of attack is to counter strike. It's highly probably that it's already being planned or even carried on. That will solve nothing. You might get even, but that achieves nothing. The death will not come back and the attack has been already recorded on the books of history. At this point in time, counter attacking is irrational and puts not only the lives of U.S. citizens at risk, but those of lots of people all arround the world, too.
FWIW, I'm not a citizen of the U.S. nor do I live there.
Violence induces more violence. Retaliation will only lead to more deaths. If you are a citizen of the U.S. of America, please write your representative right now and ask him to join a plea for peace. Historically the U.S. reaction to this kind of attack is to counter strike. It's highly probably that it's already being planned or even carried on. That will solve nothing. You might get even, but that achieves nothing. The death will not come back and the attack has been already recorded on the books of history. At this point in time, counter attacking is irrational and puts not only the lives of U.S. citizens at risk, but those of lots of people all arround the world, too.
Starting at 2300 m/s and ending at 2600 m/s over 80 m gives you an acceleration of (2600**2-2300**2)/(2*80) ~ 9200 m/s**2, which is ~ 1000 g. OTOH, from 0 m/s to 2300 m/s over 40 m is 66000 m/s**2 (~6700 g), but nothing actually says the acceleration was uniform so 10000g peak inside the barrel is not that far fetched.
For example, go to The Universe on the Table, read the story (it's good)... keep on reading... there! You are on the footnote. Do you see that editor's note where it says the author produced the story you just read especially for The Inifite Matrix and that's part of a larger work, The Periodic Table of Science Fiction? Well, there there's another story, the one that corresponds to Mg (you just read H), called Under's Game. Ah. That got your attention, didn't it? Go read it. It's hilarious (if you read the original work by OSC)
I don't know. What's the tradeoff? Lots of chicken droppings on your field? The soil can use some nutrients, I'm sure. Can the chickens do it? Have you tried or is it a hunch of yours? Who eats more? The two oxen or the chickens? What about the sleeping place? Sure, those are a lot of chickens, but they don't have a problem if you sit them side to side and they don't have a problem with sleeping in three or four rows. Manageability, yes, that's a problem... now we see a real tradeoff. Being able to replace a chicken if it dies (and chickens are cheap) or one (possibly both) of your oxen, which are not cheap, versus manageability of the two oxen. Sure, feeding the chickens is also a problem, as well as collecting the eggs wow! byproducts! suddenly I can do something with the chickens that I couldn't do with the oxen.
A related site, which I find a bit more interesting, is the clusters database. Particularly noteworthy are three PC clusters that cross the teraflops line (peak performance, mind you, but still impressive).
My question is a bit tainted by my personal opinion in this matter, but I'm sure Sam will be able to provide a different (and surely interesting) point of view.
At the risk of being a tad Linux centric, does the availability or unavailability of hardware specifications and technical documentation on graphics hardware affect a Linux game developer and why? With two concrete cases in mind, on one hand NVIDIA who provides binary drivers and "high level" documentation about their cards, and on the other hand, ATI for which there's source code for the drivers with support for half of the features the hardware offers, but neither openly accesible hardware documentation nor much "high level" docs, how do you think this can affect the future development of games for Linux (proprietary or otherwise)?
those suckers with the 1600x1200 screens and the GeForce 2 video card are still super pimp
It'd be pimp if a) you could restart the X server without locking the machine solid and b) NVIDIA would finally get a clue, admit that the bloody GPU rag goes along with documenting how said GPU works (IOW, would you buy a CPU for which there's no technical documentation?)
Why? Wouldn't the computers work just as well with no display card period, booting off of a network card that images them
and the image on the wall would come from... where exactly? Those 20 graphic cards are rendering the image fed to the LCD projectors that, uhm, project them into the wall.
Um...how do you build half a display?
It's not a screen, it's 20 projected images on the wall. Just remove 10 and you have "half" a display.
Hi, in case you guys want to spend some time answering questions here, how are you syncing the graphics cards across the wall? SGI claims you need an Infine Reality to achieve decent syncronization, which otherwise is very noticeable. Personally I've seen one CAVE type of installation driven by PCs syncronized over a RS-323 port and I can't honestly say I noticed the projectors being out of sync.
Which takes me to the other question: are you using or do you plan to implement active stereo projection or do you want to install another 20 projectors and use a passive system?
This story actually gives me a chance to bitch about OpenGL! None of these new features are a part of the standard OpenGL. "Extensions! Extensions!" you shout. However, due to the differences between hardware, you'll end up with ATI and NVIDIA versions of the same extensions, since the ARB won't touch such new/untested features.
You are boring, you know? (And moderators couldn't spot a troll even if they were standing under a bridge) We already went over this at least twice, and you are bringing out the same dried up arguments again. You still haven't said what prevents NVIDIA and ATI and the ARB from sitting at the table and coming up with an uniform API for such an extension. Last time your cried, just like know, that the "hardware" is too different. Bullshit. Direct3D is no better. You just have to decide on a freaking API and let the vendors implement that. Read elsewhere in this discussion about programmers not supporting DirectX N yet. How's that better than OpenGL's extension mechanism? It's not better. You still have to rewrite stuff. The only difference is that some companies that will remain unnamed have placed stupid patents arround interfaces, not features, interfaces and other companies have to come up and implement their own. If you want to bitch at someone, bitch at the companies that patent interfaces and stop trolling about OpenGL not supporting current technology.
It's a known fact that maya civilization in tha andes [...]
No, it's not, because the Maya, in the Gulf of Yucatán, didn't live anywhere near to Los Andes, mountain range in the south of the continent. You are thinking about the Inca. Don't you watch Disney movies?
[quoting Ebert] his hero survives two bumpy crash-landings that look about as realistic as the effects in his "Mars Attacks!"
Ooooooh... I was getting dissapointed. For a moment I thought Burton gave in and did a, *gasp*, 100% Hollywood movie. This quote from Ebert makes me want to see the movie. One of Tim Burton's strengths is that his movies stand appart by actually making a point. I loved Mars Attacks, it's a B-movie with Hollywood budget. I can almost hear Burton scream "Look at me! Watch me make a B-movie marketed as a normal one!" B-movies have the compeling attribute of not caring about being a box office hit or not. That also means you can't get big budgets for such a thing. Which is good. People making B-movies have fun doing them, and that shows. Sure, there are lots of crappy B-movies, but a large number of them can be appreciated as film making, you can see the hand of a passionate director at work (contrast with any recent Spielberg works -- I was a fan of him, until he peaked and went free fall), you get to see wonderful photography where the camera crew works with the equipement they have instead of being used by it (contrast, e.g., Speed, The Mummy), you don't usually get to see good acting, but you get to see people trying to act good, instead of
yelling "I'm good and I beleive it" (recent 007s, for example).
I was curious about Burton's vision of PotA, both because I love his work and because I watched the original when I was arround 12 or so, and I liked the film even back then. I watched it again a couple of times, and I liked it both more and less. In particular I don't like some of the acting, it looks unnatural, acted. The apes are one of the best elements in the original film. They are not only beleivable, they aren't just people in customes, but they make you think actual apes could have evolved in that direction.
While you could, theoretically, specify a different "identity" file for automated scripts, it offers little benefit since SSH doesn't provide a way of restricting rights based on the public key.
Actually, you can use another identity and restrict what it does. On the server machine you can put an entry for the second key on your.authorized_keys which reads like: (blame/-code for the ugly formatting)
from="mybox.domain.org",command="ls",no-port-forwa rding,no-X11-forwarding,no-agent-forwarding,no-pty the usual stuff here
And presto, you have a passphraseless passwordless authorization that can perform a specific action on the remote host only if the connection comes from a trusted box. There are other options, you can read it all in sshd(8).
I think gcc-3 meets or exceeds the steep C++ support that these clients require
The C/C++ user's journal ran a very interesting roundup regarding this. The STL situation seems a bit on the bad side, but the compiler itself seems to be pretty much standards conformant. Having to swith between SGI's MIPSpro, HP's aCC and gcc is sometimes painful but you can usually fix one compiler's complaining and still have it work with the other two.
So the question is, how long before the cheap renderfarm is joined by the cheap rasterisation farm, stuffed with nothing but GeForce2s, Celerons and RAM? Anyone?
There's still a lot of algorithms that you can't easily map to hardware (in other words, stuff that's much easier to program in software only). There's also the problem about the ammount of memory on the graphics card being much lower than the system's memory. Even if you can rasterize textured triangles much faster in hardware than in software, you still want to apply lots of texture layers to your objects, you want to work with larger and much more detailed textures or in general you need more realism than what current hardware-based algorithms can deliver.
But yes, you are right, hardware accelerated render farms will be something common in the future. The massively parallel pieces of silicon that current GPUs are is something you can't ignore. I have done some work on hardware accelerated distributed rendering, but alas, the website is not up yet. For interactive applications is not as good as I had hoped (instead of increasing the speed I can increase the problem size with a little performance gain on the side) but for applications where you want your stuff rendered faster (100x or 1000x faster), it's definitely a go.
The OpenGL case is really sweet: you can use GLX PBuffers (I understand WGL supports something similar too, as will DX-something, if not the current one) to render off-screen. In practical terms that means the size of your image is limited by the size of the available memory on the card. If you don't have GLX 1.3 support (ATI can I have some docs please, I want to understand how the card performs memory management in order to write such an extension for the Radeon under XFree86) you can still use tiling and get the same result (slower maybe, but anything faster than pure software is a gain here)
It's only incidental that Linux runs on commidity x86 hardware rather than having to run IRIX on ultra-expensive Origin machines. What aspects of Linux makes it an OS particularly suited to a renderfarm or in 3D work in general?
Read the article, it answers precisely this question. First consider what the task at hand is: rendering animations for the silver screen, that means hardware accelerated OpenGL rendering doesn't cut it, it's highly probable that you want to do software rendering. You have shadows, volumetric effects, particle systems, whatever, the point is you want to render this at a very high resolution and with a very high level of detail. For that kind of task, hardware rendering, either SGI or PC based, doesn't cut it or doesn't make sense (why would you want to render this interactively? noone is interacting with the scene). Morale: you want faster machines with lots of memory. Compare a farm of SGIs with a farm of PCs with the same raw processing power and the same ammount of memory. If you ignore the memory bandwidth issue, PCs beat SGIs hands down (if you doubt this try to figure out what SGI is doing right now and why they are getting into the IA64 market)
Then there's the "it's Unix" factor. If you are a Unix house (and have been for a lot of time -- DreamWorks is) you don't want to forget about all your in house tools and move to Windows to profit from the better price/performance ratio of PCs. You want to keep using Unix. Linux, for practical purposes, is that. Why not other PC Unices? (Solaris or the BSDs) Linux has better support, both in terms of hardware and software. It's easier to get people skillfull in Linux than the other alternatives, and the number is increasing, which means it's probably a bit cheaper, too.
Regarding the compiler issue someone else mentioned, there are several alternatives, like the Portland Group compilers, to name one. If it is an issue, the studio can invest in this. I'm sure they have evaluated the cost/benefit of this. If they say GCC cuts it for them...
And last, the above post is an obvious troll... moderators, are you awake?
Good point, except for one thing. Taco's review was not so much butting in (IMHO) as augmenting.
He can come down from Mount Olympus and make a, *gasp*, post, can't he? If he wants to argument or counterargument he can do so like everyone else, can't he?
Do you base your data mostly on marketing analysis or do you actually go a pay a consultor to scan machines on the net? If there are scans involved, how do you pick the IP blocks to be scanned and what's the uncertainty associated with such a method (and how is this uncertainty guessed)? If there are no scans involved, why not? If this is "maket analysis", can you defined that for me? Which factors are involved? And a different question: who's the target market for this kind of study? How much does such a thing cost?
[...] "self-made"? Numbers 215, 396, and 413 [...]
He, I've seen and worked with 215. Pretty non awe inducing group of 96 dual Pentium III boxes using a Myrinet interconnect. You can find some pictures at the Kepler homepage. It's amazing what good lighting and a good photographer can do:-) That aside, the system itself is pretty cool (and damn fast, even if there's some contention among its users for computing time)
Slashdot Mirror
There's something fundamentally wrong with this line of reasoning, which is mostly what prompted me to write the message at the head of this thread. You are advocating a counter strike. Let's assume you are successful and do actually cause some damage to these groups? Do you think they'll stand there a do nothing about it? One of the things today's attack proved is that this is doable. These people are hoping the U.S.A. counter strikes. This is not war against a country. There's no country to declare war to (at least not atm, but I don't think this will change). This is terrorism. You can only target groups, but that's the problem. It's not like you bomb a building and that's it. To put this in terms people in this group can understand, it's just like MS trying to attack Linux. They can't because there's no single entity where your attack can be directed to. Think about what the U.S.A. did in Panama, and notice there are clear differences. You could break into some guy's house and bring him to the U.S.A. and prosecute and excecute him there, but you still have a group.
You also have to notice something else: the purpose of the attack is to enrage, to create chaos, to make the people claim for revenge. This is not a military installation that was attacked, that probably would have failed. The target was a civilian target because that stood a chance of success. This attack is bait. Don't swallow it.
And you implicitely noted something else: this is the first time is a lot of years that war has been brought to U.S.A.'s backyard, which is perhaps the reason why people in other parts of the world see it differently: they still remember having war on their backyards.
Shit. Kabul, Afghanistan, seems to be under missile attack.
So, I'm the one making the stupid assumption, uh? Which country are you going to declare war against? War is an armed conflict between two sovereign nations. Precisely because this is a terrorist act you can't point your guns to some country and be done with it. In doing so, the U.S.A. will be drawn into starting a war, much like you were drawn into taking parts in a war because of Pearl Harbor. Justice is to be made, but war is not the way.
But a military attack on the suspects won't either. Just look at this forum which is supposed to have lots of highly educated people that can really assess the benefits derived from U.S.A.'s previous military responses. They are roaring for blood.
I never said this has to go unanswered. But wiping some country out of the map isn't the correct answer. Today fifty thousand died. How many does it take to get even? Fifty thousand, too? Five hundred thousand? Five million? All of them?
So, your answer is for both sides to bulldoze each other and call the last man standing the winner to enjoy whatever is left afterwards?
FWIW, I'm not a citizen of the U.S. nor do I live there.
Violence induces more violence. Retaliation will only lead to more deaths. If you are a citizen of the U.S. of America, please write your representative right now and ask him to join a plea for peace. Historically the U.S. reaction to this kind of attack is to counter strike. It's highly probably that it's already being planned or even carried on. That will solve nothing. You might get even, but that achieves nothing. The death will not come back and the attack has been already recorded on the books of history. At this point in time, counter attacking is irrational and puts not only the lives of U.S. citizens at risk, but those of lots of people all arround the world, too.
FWIW, I'm not a citizen of the U.S. nor do I live there.
Violence induces more violence. Retaliation will only lead to more deaths. If you are a citizen of the U.S. of America, please write your representative right now and ask him to join a plea for peace. Historically the U.S. reaction to this kind of attack is to counter strike. It's highly probably that it's already being planned or even carried on. That will solve nothing. You might get even, but that achieves nothing. The death will not come back and the attack has been already recorded on the books of history. At this point in time, counter attacking is irrational and puts not only the lives of U.S. citizens at risk, but those of lots of people all arround the world, too.
Starting at 2300 m/s and ending at 2600 m/s over 80 m gives you an acceleration of (2600**2-2300**2)/(2*80) ~ 9200 m/s**2, which is ~ 1000 g. OTOH, from 0 m/s to 2300 m/s over 40 m is 66000 m/s**2 (~6700 g), but nothing actually says the acceleration was uniform so 10000g peak inside the barrel is not that far fetched.
260 ft is 79.25 m, 30 ms is 30 ms, so that's an average speed of 2.641 km/s or 9508 km/h. The initial velocity of 5325 mph is 2.380 km/s or 8570 km/h
Wow.
For example, go to The Universe on the Table, read the story (it's good)... keep on reading... there! You are on the footnote. Do you see that editor's note where it says the author produced the story you just read especially for The Inifite Matrix and that's part of a larger work, The Periodic Table of Science Fiction? Well, there there's another story, the one that corresponds to Mg (you just read H), called Under's Game. Ah. That got your attention, didn't it? Go read it. It's hilarious (if you read the original work by OSC)
I don't know. What's the tradeoff? Lots of chicken droppings on your field? The soil can use some nutrients, I'm sure. Can the chickens do it? Have you tried or is it a hunch of yours? Who eats more? The two oxen or the chickens? What about the sleeping place? Sure, those are a lot of chickens, but they don't have a problem if you sit them side to side and they don't have a problem with sleeping in three or four rows. Manageability, yes, that's a problem... now we see a real tradeoff. Being able to replace a chicken if it dies (and chickens are cheap) or one (possibly both) of your oxen, which are not cheap, versus manageability of the two oxen. Sure, feeding the chickens is also a problem, as well as collecting the eggs wow! byproducts! suddenly I can do something with the chickens that I couldn't do with the oxen.
I like this analogy.
A related site, which I find a bit more interesting, is the clusters database. Particularly noteworthy are three PC clusters that cross the teraflops line (peak performance, mind you, but still impressive).
My question is a bit tainted by my personal opinion in this matter, but I'm sure Sam will be able to provide a different (and surely interesting) point of view.
At the risk of being a tad Linux centric, does the availability or unavailability of hardware specifications and technical documentation on graphics hardware affect a Linux game developer and why? With two concrete cases in mind, on one hand NVIDIA who provides binary drivers and "high level" documentation about their cards, and on the other hand, ATI for which there's source code for the drivers with support for half of the features the hardware offers, but neither openly accesible hardware documentation nor much "high level" docs, how do you think this can affect the future development of games for Linux (proprietary or otherwise)?
Thanks
It'd be pimp if a) you could restart the X server without locking the machine solid and b) NVIDIA would finally get a clue, admit that the bloody GPU rag goes along with documenting how said GPU works (IOW, would you buy a CPU for which there's no technical documentation?)
and the image on the wall would come from ... where exactly? Those 20 graphic cards are rendering the image fed to the LCD projectors that, uhm, project them into the wall.
It's not a screen, it's 20 projected images on the wall. Just remove 10 and you have "half" a display.
Hi, in case you guys want to spend some time answering questions here, how are you syncing the graphics cards across the wall? SGI claims you need an Infine Reality to achieve decent syncronization, which otherwise is very noticeable. Personally I've seen one CAVE type of installation driven by PCs syncronized over a RS-323 port and I can't honestly say I noticed the projectors being out of sync.
Which takes me to the other question: are you using or do you plan to implement active stereo projection or do you want to install another 20 projectors and use a passive system?
You are boring, you know? (And moderators couldn't spot a troll even if they were standing under a bridge) We already went over this at least twice, and you are bringing out the same dried up arguments again. You still haven't said what prevents NVIDIA and ATI and the ARB from sitting at the table and coming up with an uniform API for such an extension. Last time your cried, just like know, that the "hardware" is too different. Bullshit. Direct3D is no better. You just have to decide on a freaking API and let the vendors implement that. Read elsewhere in this discussion about programmers not supporting DirectX N yet. How's that better than OpenGL's extension mechanism? It's not better. You still have to rewrite stuff. The only difference is that some companies that will remain unnamed have placed stupid patents arround interfaces, not features, interfaces and other companies have to come up and implement their own. If you want to bitch at someone, bitch at the companies that patent interfaces and stop trolling about OpenGL not supporting current technology.
No, it's not, because the Maya, in the Gulf of Yucatán, didn't live anywhere near to Los Andes, mountain range in the south of the continent. You are thinking about the Inca. Don't you watch Disney movies?
Ooooooh... I was getting dissapointed. For a moment I thought Burton gave in and did a, *gasp*, 100% Hollywood movie. This quote from Ebert makes me want to see the movie. One of Tim Burton's strengths is that his movies stand appart by actually making a point. I loved Mars Attacks, it's a B-movie with Hollywood budget. I can almost hear Burton scream "Look at me! Watch me make a B-movie marketed as a normal one!" B-movies have the compeling attribute of not caring about being a box office hit or not. That also means you can't get big budgets for such a thing. Which is good. People making B-movies have fun doing them, and that shows. Sure, there are lots of crappy B-movies, but a large number of them can be appreciated as film making, you can see the hand of a passionate director at work (contrast with any recent Spielberg works -- I was a fan of him, until he peaked and went free fall), you get to see wonderful photography where the camera crew works with the equipement they have instead of being used by it (contrast, e.g., Speed, The Mummy), you don't usually get to see good acting, but you get to see people trying to act good, instead of yelling "I'm good and I beleive it" (recent 007s, for example).
I was curious about Burton's vision of PotA, both because I love his work and because I watched the original when I was arround 12 or so, and I liked the film even back then. I watched it again a couple of times, and I liked it both more and less. In particular I don't like some of the acting, it looks unnatural, acted. The apes are one of the best elements in the original film. They are not only beleivable, they aren't just people in customes, but they make you think actual apes could have evolved in that direction.
Actually, you can use another identity and restrict what it does. On the server machine you can put an entry for the second key on your .authorized_keys which reads like: (blame /-code for the ugly formatting)
And presto, you have a passphraseless passwordless authorization that can perform a specific action on the remote host only if the connection comes from a trusted box. There are other options, you can read it all in sshd(8).
The C/C++ user's journal ran a very interesting roundup regarding this. The STL situation seems a bit on the bad side, but the compiler itself seems to be pretty much standards conformant. Having to swith between SGI's MIPSpro, HP's aCC and gcc is sometimes painful but you can usually fix one compiler's complaining and still have it work with the other two.
There's still a lot of algorithms that you can't easily map to hardware (in other words, stuff that's much easier to program in software only). There's also the problem about the ammount of memory on the graphics card being much lower than the system's memory. Even if you can rasterize textured triangles much faster in hardware than in software, you still want to apply lots of texture layers to your objects, you want to work with larger and much more detailed textures or in general you need more realism than what current hardware-based algorithms can deliver.
But yes, you are right, hardware accelerated render farms will be something common in the future. The massively parallel pieces of silicon that current GPUs are is something you can't ignore. I have done some work on hardware accelerated distributed rendering, but alas, the website is not up yet. For interactive applications is not as good as I had hoped (instead of increasing the speed I can increase the problem size with a little performance gain on the side) but for applications where you want your stuff rendered faster (100x or 1000x faster), it's definitely a go.
The OpenGL case is really sweet: you can use GLX PBuffers (I understand WGL supports something similar too, as will DX-something, if not the current one) to render off-screen. In practical terms that means the size of your image is limited by the size of the available memory on the card. If you don't have GLX 1.3 support (ATI can I have some docs please, I want to understand how the card performs memory management in order to write such an extension for the Radeon under XFree86) you can still use tiling and get the same result (slower maybe, but anything faster than pure software is a gain here)
Read the article, it answers precisely this question. First consider what the task at hand is: rendering animations for the silver screen, that means hardware accelerated OpenGL rendering doesn't cut it, it's highly probable that you want to do software rendering. You have shadows, volumetric effects, particle systems, whatever, the point is you want to render this at a very high resolution and with a very high level of detail. For that kind of task, hardware rendering, either SGI or PC based, doesn't cut it or doesn't make sense (why would you want to render this interactively? noone is interacting with the scene). Morale: you want faster machines with lots of memory. Compare a farm of SGIs with a farm of PCs with the same raw processing power and the same ammount of memory. If you ignore the memory bandwidth issue, PCs beat SGIs hands down (if you doubt this try to figure out what SGI is doing right now and why they are getting into the IA64 market)
Then there's the "it's Unix" factor. If you are a Unix house (and have been for a lot of time -- DreamWorks is) you don't want to forget about all your in house tools and move to Windows to profit from the better price/performance ratio of PCs. You want to keep using Unix. Linux, for practical purposes, is that. Why not other PC Unices? (Solaris or the BSDs) Linux has better support, both in terms of hardware and software. It's easier to get people skillfull in Linux than the other alternatives, and the number is increasing, which means it's probably a bit cheaper, too.
Regarding the compiler issue someone else mentioned, there are several alternatives, like the Portland Group compilers, to name one. If it is an issue, the studio can invest in this. I'm sure they have evaluated the cost/benefit of this. If they say GCC cuts it for them...
And last, the above post is an obvious troll... moderators, are you awake?
He can come down from Mount Olympus and make a, *gasp*, post, can't he? If he wants to argument or counterargument he can do so like everyone else, can't he?
Do you base your data mostly on marketing analysis or do you actually go a pay a consultor to scan machines on the net? If there are scans involved, how do you pick the IP blocks to be scanned and what's the uncertainty associated with such a method (and how is this uncertainty guessed)? If there are no scans involved, why not? If this is "maket analysis", can you defined that for me? Which factors are involved? And a different question: who's the target market for this kind of study? How much does such a thing cost?
He, I've seen and worked with 215. Pretty non awe inducing group of 96 dual Pentium III boxes using a Myrinet interconnect. You can find some pictures at the Kepler homepage. It's amazing what good lighting and a good photographer can do :-) That aside, the system itself is pretty cool (and damn fast, even if there's some contention among its users for computing time)