Your right, the Ars Technica guy did say that. I was making a guess based on the fact that this week they published a paper on their new massively parallel ray tracer.
Global illumination does need help. I still think that raytracing is the way to go though, and I think something more flexible than the Renderman spec is needed as well. With some of the new renderers, you might get the best of both worlds (sorry, haven't been able to try Jig, Entropy or RenderDotC yet, so I can't say for sure). You get raytracing, and you get compatibility with existing software (probably a minor issue though) through the rib interface, but you can also right more in depth extensions. Say representing hair as RiCurves in the rib stream, but then raytrace is as a volume instead of just a surface.
But, photo realistic 3D rendering isn't my main focus. I mainly work on 2D imaging and real time 3D for training tools.
Despite the cover shot of Shrek, most of the article was about the cell animation tools that Dreamworks ported to use on Linux workstations rather than about rendering and tools used for Shrek. I distinctly get the impression that their cell tools weren't written using OpenGL well, or they wouldn't have felt that it was so easy to port. I bet with a little cleverness, they could have gotten a lot more speed by using OpenGL effectively (a lot of people don't realize that OpenGL is also great for 2D work). But, maybe really effective use of OpenGL would have required upgrading the Octanes (like if they bought Octanes without texture ram).
Err, most of the rendering being done on linux for titanic was compositing rendering, not 3D rendering. They ported their custom written compositing software to linux for the job.
>The only renderer used for feature films is
>PRman, with some help from BMRT for scenes that
>really can't do without raytracing.
You are forgetting Mental Ray. It has been used on quite a number of movies, and it has been around a long time. Buf Companie uses Mental Ray, as do other companies. Further, more and more companies are looking at other renderers like Jig, Entropy, and RenderdotC because as great as the PRMan standard has been all of these years, high end work is starting to require something more flexible.
Also, raytracing seems to be making a big comeback these days. It makes adding things like subsurface scattering and other specialty surfaces a lot easier, and also makes it easier to incorporate volumetric objects (hair is one thing that is nice to be able to do volumetrically). I believe that Final Fantasy was raytraced, and I'm pretty sure that Shrek was also, and I know that everything BlueSky does is raytraced.
Premier and AfterEffects aren't at all tied to a specific OS or hardware platform. Further, Premier is never used for films so it doesn't matter whether it makes it to linux. After Effects is used on some films, but I don't it will make it. But there are many superior (but also more costly) solutions on unix. For instance, there is Shake (NT, Linux, and Irix) and Pihrana (Irix) Flint and Flame (Irix), and Inferno (also Irix). Of those substitutes listed, all except for Shake are pretty closely tied to the hardware, so porting isn't going to be simple in any way shape or form. Besides, Linux doesn't really run on suitable hardware yet for those programs anyway (sure the CPU is fast enough, but those programs are tightly tied to SGIs video hardware, and while PCs might be able to keep up in blasting triangles, they still have a long way to go for color fidelity and texturing speed).
But the most important thing to remeber is that so many places use a lot of custom software, especially on the parts of the production pipeline that they are looking at linux for. Dreamworks animation (the part that does cell animation like The Road to Eldorado and Prince of Egypt) has replaced a lot of SGI workstations with linux machines, but PDS is quite a distance from replaceing their SGI workstations with linux machines. Instead they are replaceing the racks of O2s and Origins used for rendering using nearly completely proprietary software.
The same goes for Pixar (except they uses Suns on the server side running completely custom software), Hammerhead (O2s on desktop and an Athlon renderfarm running PRMan and custom software), digital domain (PRMan, and custom compositing software), and almost any other big company. The vast majority of machines in "hollywood" (are any of the big effects companies actually in hollywood?) are machines for rendering, and for rendering, virtually all of the software is there already.
There are several problems with your plan. First, second hand P2s will have stuff in them that you don't need, like video cards, sound cards, etc. Second, the selves that you mention most likely won't hold up well to the weight of the machines, and you will need a whole room to hold them. Also, the networking hardware will probably be a lot more than $2k if you want actual switches instead of hubs (although you might be able to work around not having switches, but it would be cheaper to buy the switches then to get someone to code around them not being there).
Now, if you use rack mounted machines, you will save a lot of space. It is possible to get special racks that have a large power supply and trays with room for just the mobo and harddrive. This will let you fit 128 machines into about 2 racks. Each machine will need at least 256 megs of ram, and at least a four gig hard drive (increase ram and harddrive depending on complexity, each frame in toystory requires a gig to render, and several gigs to hold the geometry and texture files). You probably can assemble each node for $200 to $300 using Athlons and IDE drives. But, for better reliability (which pays for itself in requiring less maintainence staff), switch to SCSI and consider going to Intel. Proper switches will costs at least $5000 most likely. If you outfit each node with a harddrive large enough to hold all the geometry and textures, you could consider just downloading everything to every machine, and using hubs, since move the rendered frames back to the file server won't hog that much traffic. But, as I said, why bother?
Excuse me. While it is true that the first two intel machines from SGI were workstations (linux runs on them, but it can't use the frame buffer), the next batch of machines, like the SGI 1400, SGI 1200, etc, are for rendering and serving tasks. These machines are mainly for linux (although initially at least NT was also supported). It wasn't until more recently (last year?) that SGI actually shipped a linux workstation.
You have to be careful about saying that things can be purchased for linux, even if it is sort of true. For instance, Houdini has been on linux for quite some time, but initially it was sold only bundled with HPs FX linux machines. These machines are notable because the use the HP-UX xserver on really high end hardware instead of using XFree like SGI does. I don't know what commercial user software runs well on SGIs yet. The SGI linux video hardware is not as good as HPs, and it doesn't look likely to change anytime soon either, which is too bad. The Quatro is fast, but it still misses important things like 10 and 16bit color depths, and really acurate rendering. Shake does everything in software, so that isn't an issue, but almost Flint/Flame, and Pihrana use the hardware for rendering, and so acuracy and color depth are extremely important with those programs.
Re:As long as I can connect...
on
Taming the Web
·
· Score: 1
Uhm, I think they already regulate the sale of lasers. They definately regulate the use of lasers.
You can get a good idea by taking your car to a test track and seeing how fast you can get it up to. Isn't that what everyone does when they get a new car?
If there are so many of you doing it, then petition the township, and they can help make it legal. I don't think that they can actually make it legal, but they can force the utility to allow it with the threate of giving the contract to a different utility. Uhm, this assumes that we are talking about utility based high speed services, like Verizon DSL or Comcast Cable.
The thought of revoking comcasts contract has come up quite frequently with local politicos do to comcast being so phenominally bad at living up to their promises.
If Verizon thinks that I'm going to get a DSL account for each computer in my house, they can just go screw themselves. I will not, nope, ugh, ugh. If they try to make me, I'll back to channel bonded dial ups.
2048 seems to pretty much be standard. Some people do shots in 4096. Toy Story was at a lower resolution, something like 1400 pixels across.
To render it on a quadro, I'd imagine that they would also have to cut texture resolutions, but that article just didn't say enough about what happened to be usefull.
In altivec, multiple instructions can be executed at once, and each instruction works on 4 to 16 numbers at once. A cray on the other had also executes multiple instructions at once, but instead of only operating on 4 to 16 numbers per instruction, an instruction can affect up to 64k numbers. This obviously does not happen in one clock cycle but it does happen fast enough that one 300mhz processor is faster than several gigahertz processors, especially when you look at weighing in memory access times.
Crays (except possibly the CS6400, the machine the Sun E10k is based on), sort of use a 64bit word. I say sort of because it holds a 64bit floating point or integer number, but it stores it in up to 80bits, depending on machine. It uses the extra space for error correction.
So, an 8megawork cray is equivalent to a 64megabyte PC (memory wise that is), except it really has 80 megs.
Cray also makes distributed clusters BTW.
But anyway, just to clarify, when we say that real Crays are vector machines, people keep pointing out Altivec. I would just like to comment that while altivec has 128bit vectors (anothers words, it can hold 4 32bit numbers in one vector register, or 8 16 numbers), Crays have vectors that can hold up to 64 thousand 64bit elements. Of course, it takes many clock cycles to add or multiply vectors, but it is still faster than the altivec way.
I don't know what a T1 currently costs. However, I'd say that you probably don't want more than 15 people per T1, so it would be pretty costly for everyone.
I don't think a DSLAM is needed. Just two regular SDSL modems for every user. There are documents out there about doing this.
The person who gets the T1 really should have a proper machine room to support the routers (16+) and other machines.
I'm pretty sure that they aren't blocking outgoing connections to port 25, because just recently, they said in a public notice that if I wanted to send mail from other email addresses than theirs, then I would need to use other email servers. But, I'm not home, so I can't test this right now.
Still, this is a strong argument for getting one friend to get a T1, then everyone chips in to pay for the T1 and roll their own DSL connections (hint: dry wires, also known as security lines, from the phone company is cheap, like $20 a month). I wish I had more friends nearby so that I could do this.
Judges are just to interpret the laws, not rule wether they are just or not. If there is a law they don't like (say the DMCA), they can't do anything about it.
Being pissed off just makes it easier for the loser to get the ruling overturned on appeal.
I used to think mac's sucked until i worked on them and supported them - they dont - the G4 is a mind blowing machine for what its intended for - trust me it can and does piss on any wintel or IBM compatible equipment in the fields of Graphics manipulation, Desktop Publishing, video editing and related functions
I don't know. I mean, G4s are nice, but in no way shape or form can they compare well to the best that IBM compatible machines have to offer. I mean, what could be more IBM compatible than an IBM? With that in mind, check out this. This bad boy running gimp or many other graphics apps will kick a G4 every which way, and it will do so with out breaking a sweat. And the G4s Velocity Engine is helpless to prevent the onslaught. So, screw those Apple toys, and get yourself a real IBM compatible machine.
(BTW, this is a joke, unless you actually have a spare $50k for that IBM, in which case go for it).
I don't take my notebook to church (do take palm pilot, mainly because I take it everywhere). But on any given sunday, I usually see at least 3 notebooks in the pews, plus a bunch of palm pilots and windows CE machines.
Oh yeah, and anyone who spends 700 fucking dollars on a chair deserves to go broke. IMHFO. For that price it had better massage my back, do my taxes, and wipe my ass automatically.
If you spend most of your time in a chair, then $700 is not to much to spend. It is better to have a good chair, monitor, keyboard, and mouse (or other pointer input) than it is to have the latest and fastest CPU, or the largest disks.
Now, I personally don't have $700 for a chair (Heck, I had a hard time with $700 for a car), but I haven't been too skimpy either, and in my home office, the chair, monitor, keyboard, and pointing devices cost as more than the machine itself did.
Slashdot Mirror
Your right, the Ars Technica guy did say that. I was making a guess based on the fact that this week they published a paper on their new massively parallel ray tracer.
Global illumination does need help. I still think that raytracing is the way to go though, and I think something more flexible than the Renderman spec is needed as well. With some of the new renderers, you might get the best of both worlds (sorry, haven't been able to try Jig, Entropy or RenderDotC yet, so I can't say for sure). You get raytracing, and you get compatibility with existing software (probably a minor issue though) through the rib interface, but you can also right more in depth extensions. Say representing hair as RiCurves in the rib stream, but then raytrace is as a volume instead of just a surface.
But, photo realistic 3D rendering isn't my main focus. I mainly work on 2D imaging and real time 3D for training tools.
Despite the cover shot of Shrek, most of the article was about the cell animation tools that Dreamworks ported to use on Linux workstations rather than about rendering and tools used for Shrek. I distinctly get the impression that their cell tools weren't written using OpenGL well, or they wouldn't have felt that it was so easy to port. I bet with a little cleverness, they could have gotten a lot more speed by using OpenGL effectively (a lot of people don't realize that OpenGL is also great for 2D work). But, maybe really effective use of OpenGL would have required upgrading the Octanes (like if they bought Octanes without texture ram).
Err, most of the rendering being done on linux for titanic was compositing rendering, not 3D rendering. They ported their custom written compositing software to linux for the job.
>The only renderer used for feature films is
>PRman, with some help from BMRT for scenes that
>really can't do without raytracing.
You are forgetting Mental Ray. It has been used on quite a number of movies, and it has been around a long time. Buf Companie uses Mental Ray, as do other companies. Further, more and more companies are looking at other renderers like Jig, Entropy, and RenderdotC because as great as the PRMan standard has been all of these years, high end work is starting to require something more flexible.
Also, raytracing seems to be making a big comeback these days. It makes adding things like subsurface scattering and other specialty surfaces a lot easier, and also makes it easier to incorporate volumetric objects (hair is one thing that is nice to be able to do volumetrically). I believe that Final Fantasy was raytraced, and I'm pretty sure that Shrek was also, and I know that everything BlueSky does is raytraced.
Premier and AfterEffects aren't at all tied to a specific OS or hardware platform. Further, Premier is never used for films so it doesn't matter whether it makes it to linux. After Effects is used on some films, but I don't it will make it. But there are many superior (but also more costly) solutions on unix. For instance, there is Shake (NT, Linux, and Irix) and Pihrana (Irix) Flint and Flame (Irix), and Inferno (also Irix). Of those substitutes listed, all except for Shake are pretty closely tied to the hardware, so porting isn't going to be simple in any way shape or form. Besides, Linux doesn't really run on suitable hardware yet for those programs anyway (sure the CPU is fast enough, but those programs are tightly tied to SGIs video hardware, and while PCs might be able to keep up in blasting triangles, they still have a long way to go for color fidelity and texturing speed).
But the most important thing to remeber is that so many places use a lot of custom software, especially on the parts of the production pipeline that they are looking at linux for. Dreamworks animation (the part that does cell animation like The Road to Eldorado and Prince of Egypt) has replaced a lot of SGI workstations with linux machines, but PDS is quite a distance from replaceing their SGI workstations with linux machines. Instead they are replaceing the racks of O2s and Origins used for rendering using nearly completely proprietary software.
The same goes for Pixar (except they uses Suns on the server side running completely custom software), Hammerhead (O2s on desktop and an Athlon renderfarm running PRMan and custom software), digital domain (PRMan, and custom compositing software), and almost any other big company. The vast majority of machines in "hollywood" (are any of the big effects companies actually in hollywood?) are machines for rendering, and for rendering, virtually all of the software is there already.
Only if you are Jesse Ventura. ;)
For those who didn't know, he was on Leno last night talking about how he never gets tickets now that he is governer.
There are several problems with your plan. First, second hand P2s will have stuff in them that you don't need, like video cards, sound cards, etc. Second, the selves that you mention most likely won't hold up well to the weight of the machines, and you will need a whole room to hold them. Also, the networking hardware will probably be a lot more than $2k if you want actual switches instead of hubs (although you might be able to work around not having switches, but it would be cheaper to buy the switches then to get someone to code around them not being there).
Now, if you use rack mounted machines, you will save a lot of space. It is possible to get special racks that have a large power supply and trays with room for just the mobo and harddrive. This will let you fit 128 machines into about 2 racks. Each machine will need at least 256 megs of ram, and at least a four gig hard drive (increase ram and harddrive depending on complexity, each frame in toystory requires a gig to render, and several gigs to hold the geometry and texture files). You probably can assemble each node for $200 to $300 using Athlons and IDE drives. But, for better reliability (which pays for itself in requiring less maintainence staff), switch to SCSI and consider going to Intel. Proper switches will costs at least $5000 most likely. If you outfit each node with a harddrive large enough to hold all the geometry and textures, you could consider just downloading everything to every machine, and using hubs, since move the rendered frames back to the file server won't hog that much traffic. But, as I said, why bother?
Excuse me. While it is true that the first two intel machines from SGI were workstations (linux runs on them, but it can't use the frame buffer), the next batch of machines, like the SGI 1400, SGI 1200, etc, are for rendering and serving tasks. These machines are mainly for linux (although initially at least NT was also supported). It wasn't until more recently (last year?) that SGI actually shipped a linux workstation.
You have to be careful about saying that things can be purchased for linux, even if it is sort of true. For instance, Houdini has been on linux for quite some time, but initially it was sold only bundled with HPs FX linux machines. These machines are notable because the use the HP-UX xserver on really high end hardware instead of using XFree like SGI does. I don't know what commercial user software runs well on SGIs yet. The SGI linux video hardware is not as good as HPs, and it doesn't look likely to change anytime soon either, which is too bad. The Quatro is fast, but it still misses important things like 10 and 16bit color depths, and really acurate rendering. Shake does everything in software, so that isn't an issue, but almost Flint/Flame, and Pihrana use the hardware for rendering, and so acuracy and color depth are extremely important with those programs.
Uhm, I think they already regulate the sale of lasers. They definately regulate the use of lasers.
You can get a good idea by taking your car to a test track and seeing how fast you can get it up to. Isn't that what everyone does when they get a new car?
If there are so many of you doing it, then petition the township, and they can help make it legal. I don't think that they can actually make it legal, but they can force the utility to allow it with the threate of giving the contract to a different utility. Uhm, this assumes that we are talking about utility based high speed services, like Verizon DSL or Comcast Cable.
The thought of revoking comcasts contract has come up quite frequently with local politicos do to comcast being so phenominally bad at living up to their promises.
If Verizon thinks that I'm going to get a DSL account for each computer in my house, they can just go screw themselves. I will not, nope, ugh, ugh. If they try to make me, I'll back to channel bonded dial ups.
2048 seems to pretty much be standard. Some people do shots in 4096. Toy Story was at a lower resolution, something like 1400 pixels across.
To render it on a quadro, I'd imagine that they would also have to cut texture resolutions, but that article just didn't say enough about what happened to be usefull.
> Did I mention the cross-compilers for SH/MIPS/etc cost a *lot*?
Bah. GCC is free. Be civilized and use that. Many other people do. Like the European Space Agency, TiVo, and others.
Because their 500mhz and gigahertz machines from the early and mid 90s didn't sell.
In altivec, multiple instructions can be executed at once, and each instruction works on 4 to 16 numbers at once. A cray on the other had also executes multiple instructions at once, but instead of only operating on 4 to 16 numbers per instruction, an instruction can affect up to 64k numbers. This obviously does not happen in one clock cycle but it does happen fast enough that one 300mhz processor is faster than several gigahertz processors, especially when you look at weighing in memory access times.
Crays (except possibly the CS6400, the machine the Sun E10k is based on), sort of use a 64bit word. I say sort of because it holds a 64bit floating point or integer number, but it stores it in up to 80bits, depending on machine. It uses the extra space for error correction.
So, an 8megawork cray is equivalent to a 64megabyte PC (memory wise that is), except it really has 80 megs.
Yes, but will yer Mac continue to outperform the Cray at running the Gimp?
Darn it, now I'm going to have to go try this out and see...
Cray also makes distributed clusters BTW. But anyway, just to clarify, when we say that real Crays are vector machines, people keep pointing out Altivec. I would just like to comment that while altivec has 128bit vectors (anothers words, it can hold 4 32bit numbers in one vector register, or 8 16 numbers), Crays have vectors that can hold up to 64 thousand 64bit elements. Of course, it takes many clock cycles to add or multiply vectors, but it is still faster than the altivec way.
I don't know what a T1 currently costs. However, I'd say that you probably don't want more than 15 people per T1, so it would be pretty costly for everyone.
I don't think a DSLAM is needed. Just two regular SDSL modems for every user. There are documents out there about doing this.
The person who gets the T1 really should have a proper machine room to support the routers (16+) and other machines.
I'm pretty sure that they aren't blocking outgoing connections to port 25, because just recently, they said in a public notice that if I wanted to send mail from other email addresses than theirs, then I would need to use other email servers. But, I'm not home, so I can't test this right now.
Still, this is a strong argument for getting one friend to get a T1, then everyone chips in to pay for the T1 and roll their own DSL connections (hint: dry wires, also known as security lines, from the phone company is cheap, like $20 a month). I wish I had more friends nearby so that I could do this.
Judges are just to interpret the laws, not rule wether they are just or not. If there is a law they don't like (say the DMCA), they can't do anything about it.
Being pissed off just makes it easier for the loser to get the ruling overturned on appeal.
I don't know. I mean, G4s are nice, but in no way shape or form can they compare well to the best that IBM compatible machines have to offer. I mean, what could be more IBM compatible than an IBM? With that in mind, check out this. This bad boy running gimp or many other graphics apps will kick a G4 every which way, and it will do so with out breaking a sweat. And the G4s Velocity Engine is helpless to prevent the onslaught. So, screw those Apple toys, and get yourself a real IBM compatible machine.
(BTW, this is a joke, unless you actually have a spare $50k for that IBM, in which case go for it).
Well, obviously, I read /.
I don't take my notebook to church (do take palm pilot, mainly because I take it everywhere). But on any given sunday, I usually see at least 3 notebooks in the pews, plus a bunch of palm pilots and windows CE machines.
If you spend most of your time in a chair, then $700 is not to much to spend. It is better to have a good chair, monitor, keyboard, and mouse (or other pointer input) than it is to have the latest and fastest CPU, or the largest disks.
Now, I personally don't have $700 for a chair (Heck, I had a hard time with $700 for a car), but I haven't been too skimpy either, and in my home office, the chair, monitor, keyboard, and pointing devices cost as more than the machine itself did.