Pacific Data Images is very quickly becoming a Linux shop (as is Pixar... though Pixar plans to stick with Sun/UltraSPARC for rendering). From what I understand, PDI used several hundred SGI Origin 200 (the 3U/4U rackmount model) running IRIX as well as seveal hundred HP boxes running linux to render Shrek. After the movie was completed, they sold their SGI Origins and replaced them with even newer x86/Linux boxes. Most of the modeling for Shrek was done with SGI Octane and SGI Octane2 workstations and the final compositing was done with Onyx2 machines. Though as you pointed out, they did experiment with Linux for some scenes and for a great deal of the rendering. According to an Alias|Wavefront press release (an SGI-owned company), PDI is replacing half of their Octanes with SGI and HP x86 Linux workstations.
I would say that Industrial Light and Magic is probably one of the world's largest SGI companies, if not the largest. Two years ago they had around 900 employees (many traditional artists and programmers) but still I think they even do their rendering on SGI's although they might be starting to switch now.
ILM has always used a wide variety of systems (Windows, MacOS, Linux, and even Amiga in projects as recently as two years ago). But the bulk of their work continues to be on the SGI MIPS/IRIX platform (modeled on Octane2 workstations, previewed on Onyx2/Onyx3000, rendered on Origin2000/Origin3000, final compositing and editing on Onyx again). ILM was recently featured on TechTV's "FreshGear" and was showing off their renderfarm, essentially 1150 MIPS R12000 and R14000 CPUs spread across 4 different Origin systems. They have they money to stick with SGI and do get the rewards of using SGI's nice hardware (such as 512 CPUs per single origin... easier to maintain, etc). They also, of course, get generous discounts from SGI.
AFAIK, SGI's largest customer in terms of sheer CPUs and bandwidth is the NSA (the USA's National Security Agency).
According to the press release, we're talking about 40 Octanes and 4 Origin 200s.
Those were the inital machines used in 1997/1998 for the FF movie. Over time they added more Octanes, and even several Origin 2000 (now "SGI 2000") and Onyx2 systems to the mix. From what I understand, SS has more SGI systems throughout the entire company than Industrial Light and Magic has, an unfair compairison, as ILM only works on movies, they don't make 3D grames (where most of SS's SGIs are put to use for mockups and modeling).
SGI's stockprice is in the toilet because they really haven't updated their hardware in over 3 years nor have adjusted their pricing or sales channels. The Origin/Onyx 3000 can scale to 512 CPUs through a cable-based backplaneless design with gobs of thruput (716.0 GB/sec) but that doesn't mean a thing with the InfiniteReality3 graphics virtually unchanged since the original InfiniteReality in 1995 and CPU modules selling for about $75,000 each. SGI is it's own worst enemy.
I always thought it was interesting how most of Pixar's "competition" actually uses Pixar software in the process of rendering their movies.... though I can understand the popularity of Renderman, it's very tuned and highly extendable through a clean plugin/shader architecture. http://www.pixar.com/products/rende rman/products/
I agree the the Blender Book is a handy and informative tutorial and reference. I've paged through the v.1.8 and v.2.0 at the local Barnes and Noble. However, I'll be damned if I'll pay $40 US for the thing. It's not even spiral bound. There is NO pdf available (compare to the Linux Documention Project) and the material is copyrighted. Nor is Blender itself GPL at the moment.
Blender is neat and (mostly) free, but I think I'll pass on this one for awhile.
(See Parent)
Teehee! Well put... I too have noticed GNOME evolve into something that almost seems to have been developed and marketed from Redmond. Interesting how a group's tolerance levels shift over time. But then again, it was only about 24 months ago when the average distro (with all the goodies installed and running) was still zippy on a 486/100 or P60.
Me? I use XFCE, but find KDE to be nifty as well. My next box will probably run either Blackbox or KDE most of the time.
(6.4 was almost a rogue port of 6.2 to support the 10k proc)
The R10K (MIPS R10000) was supported as far back as "IRIX 5.3 Including R10K" in 1995. But I agree about 6.4 (as well as 6.3 for O2)... rogue port indeed. Luckilly 6.5 has been great, especially the quarterly updates that slowly roll in new features and fix the bugs. Heavily tested, too (SGI stays about 1.5 quarters ahead of their users, heavily testing each new release on their own machines first).
Good point... even if these can get out to homes at reasonable prices, this is gonna be worse shared bandwidth than cable modem
technology; cables are 4mbps average (around here anyway), sharing a couple T3's or maybe an OC3 for larger area isn't that bad - but
one of these copper links is the speed of just one of those kinda-expensive OC3's, and i don't think the phone companies are going to be
too eager to install huge pipes to fully take advantage of a 155mbps link to each home...
VERY wel put! The nearby ISP here pays $13,000 per month for each of their upstream DS3/T3 connections (45mbps to UUNET/Worldcom). OC3 is 155mbps and is in the neighborhood of $35,000 per month.
Well put, and a good example, too. The major ISP/telco (a cooperative) in a town near me serves about 750 dialup and DSL customers. Their upstream bandwidth is pretty substantial (2x DS3 [45mbps] and 2x DS1 [1.5mbps]). I don't know what they pay for the DS1s, but both DS3s are connected to UUNET/Worldcom and cost $13,000 each per month to operate. Anyway, getting back to my point, this telco made the mistake of selling 2.2mbps ADSL for under $200/month. Within three weeks all of their upstream bandwidth was clogged. Customers complained but were not willing to pay any more, nor settle for cheaper, slower connections. All wanted 2.2mbps full time for $200/month.
Your situation sounds a lot like my former home. I lived in a new neighborhood in a city of 125,000 people. The only problem was, the telco didn't have enough lines running from that side of town to the central office (home of their switching equipment) on the other, older, side of town. So, they ended up heavily multiplexing the telephone lines. Even though we lived only a mile from a nearby telephone building, we couldn't connect to the ISP (5 miles away) above carrier 19200 - 28800. And that was *after* the telco conditioned the line for us. We ended up getting wireless internet via a 24db antenna pointed towards my ISP's antenna on the top of a nearby (2 miles away) office building.
Not all telephone lines are CAT3... my home (built in 1994) has two-pair (4 conductors) and about 8 twists per foot (somewhere in the middle of cat3 and cat5). My older home had flat (no twists) two-pair running in the walls. I have no idea what sort of cable comes from the telco to the pedestal in my yard, but from the pedistal to the box on the side of my house is six-pair (12 conductors) with probably 8 twists per foot and it's thick (22 guage versus the 24 guage normally used for telephone and network wiring).
Why does Slashdot keep using the old cube logo for SGI? The glory days of SiliconGraphics (and thus, the cube logo) are long gone. The x86/PC days are here along with the fitting 'sgi' logo. Quit using the cube.
I can see that they're standing beside many racks of SGI Origin 3000 gear (whether it's a single machine or many smaller machines depends on the cabled configuration -- O3K uses a mesh of cables rather than a backplane for its third-generation ccNUMA architecture).
At any rate, I'm curious as to who company they'll be buying their new system from. The only clue I can gather is their mention of 5120 CPUs to churn out 40 TFLOPS... that's 7.8 GFLOP per CPU which pretty much rules out SGI or even the latest Pentium/Xeon, Itanium, UltraSPARC III, or Alpha (unless one uses a very Apple-esq SIMD benchmark but I would imagine they want and need something more flexible for true performance with SISD, SIMD, and MIMD combined, rather than just bragging rights for a high single benchmark and top spot on Top500). My guess is they're going with either the not-yet-released Cray SV2 (which will combine the parallel strengths of the T3E with the vector strenghts of the SV1ex) or NEC's successor to the SX-5 (SX-5 is marketed by Cray in the USA, but under its creator, NEC, in other countries).
what about ClarisOffice (which today is known as AppleOffice)
You mean ClarisWorks / AppleWorks. AppleWorks was originally available for the Apple ][ series of machines in the late 1970s... and is now available for Mac OS X. As far as the "Office" name, I don't know how far back it goes, but the software for the Lisa (Apple's original GUI machine in 1982 - 1984) was Lisa Office System. The Macintosh shipped in 1984 with apps such as MacWrite, MacPaint, and MacDraw (all Apple apps). Microsoft made a Mac GUI version of Word later that year.
Amen for that. Living near the Dallas area (it's no Silicon Valley, but it did have its share of dot cons) I've seen my share of $200 SGI Octanes, $300 Origins, and $600 Onyx2s. You're not kidding about having to beat the dealers, dotcon closeouts are often a total madhouse. Especially when the original vendor of the still-under-warranty equipment shows up to buy back their goods for pennies on the millions (only to resell them as "remanufactured" for 350x the firesale price).
That may sound like an odd question, but if you look at SGI's Linux offerings over the past 12 month, you'll see that they resemble a (poor) attempt at being a VA Linux wannabe.
Also note that some of SGI's Linux servers were OEMed from VA.
It will be interesting to see how this affects SGI over the next 6 months... will SGI's Linux offerings be better or worse as a result?
First of all, it proves how big business (big scared businesses) can drive the little guy out of his own business.
It also shows that the future of hardware is -not- from commerical organizations, but rather from truly open, not-for-profit vendors. The only way we can continue to grow is to abandon the old idea that somone has to get rich by selling things at an inflated price. VA has proven that a vendor need not make a profit to better the world. Expect the concept of free, open hardware (and not just open software) to grow over the next few years.
...and nothing more... but as a commerical software vendor with your source totally closed, you're able to integrate GPLed code without anyone knowing about it. I'd keep a close eye on Sun over the next 12 months.
Will this become part of Red Hat Professional Server? I've heard a lot of good things about PostgreSQL and am very happy that Red Hat is going to start working with it! I use RHPS on a daily basis and will buy the next version even more quickly if Red Hat DB/Red Hat PostgreSQL will be included.
Slashdot Mirror
Pacific Data Images is very quickly becoming a Linux shop (as is Pixar... though Pixar plans to stick with Sun/UltraSPARC for rendering). From what I understand, PDI used several hundred SGI Origin 200 (the 3U/4U rackmount model) running IRIX as well as seveal hundred HP boxes running linux to render Shrek. After the movie was completed, they sold their SGI Origins and replaced them with even newer x86/Linux boxes. Most of the modeling for Shrek was done with SGI Octane and SGI Octane2 workstations and the final compositing was done with Onyx2 machines. Though as you pointed out, they did experiment with Linux for some scenes and for a great deal of the rendering. According to an Alias|Wavefront press release (an SGI-owned company), PDI is replacing half of their Octanes with SGI and HP x86 Linux workstations.
I would say that Industrial Light and Magic is probably one of the world's largest SGI companies, if not the largest. Two years ago they had around 900 employees (many traditional artists and programmers) but still I think they even do their rendering on SGI's although they might be starting to switch now.
ILM has always used a wide variety of systems (Windows, MacOS, Linux, and even Amiga in projects as recently as two years ago). But the bulk of their work continues to be on the SGI MIPS/IRIX platform (modeled on Octane2 workstations, previewed on Onyx2/Onyx3000, rendered on Origin2000/Origin3000, final compositing and editing on Onyx again). ILM was recently featured on TechTV's "FreshGear" and was showing off their renderfarm, essentially 1150 MIPS R12000 and R14000 CPUs spread across 4 different Origin systems. They have they money to stick with SGI and do get the rewards of using SGI's nice hardware (such as 512 CPUs per single origin... easier to maintain, etc). They also, of course, get generous discounts from SGI.
AFAIK, SGI's largest customer in terms of sheer CPUs and bandwidth is the NSA (the USA's National Security Agency).
According to the press release, we're talking about 40 Octanes and 4 Origin 200s.
Those were the inital machines used in 1997/1998 for the FF movie. Over time they added more Octanes, and even several Origin 2000 (now "SGI 2000") and Onyx2 systems to the mix. From what I understand, SS has more SGI systems throughout the entire company than Industrial Light and Magic has, an unfair compairison, as ILM only works on movies, they don't make 3D grames (where most of SS's SGIs are put to use for mockups and modeling).
SGI's stockprice is in the toilet because they really haven't updated their hardware in over 3 years nor have adjusted their pricing or sales channels. The Origin/Onyx 3000 can scale to 512 CPUs through a cable-based backplaneless design with gobs of thruput (716.0 GB/sec) but that doesn't mean a thing with the InfiniteReality3 graphics virtually unchanged since the original InfiniteReality in 1995 and CPU modules selling for about $75,000 each. SGI is it's own worst enemy.
I always thought it was interesting how most of Pixar's "competition" actually uses Pixar software in the process of rendering their movies.... though I can understand the popularity of Renderman, it's very tuned and highly extendable through a clean plugin/shader architecture.e rman/products/
http://www.pixar.com/products/rend
Kernel Hogan!!
I agree the the Blender Book is a handy and informative tutorial and reference. I've paged through the v.1.8 and v.2.0 at the local Barnes and Noble. However, I'll be damned if I'll pay $40 US for the thing. It's not even spiral bound. There is NO pdf available (compare to the Linux Documention Project) and the material is copyrighted. Nor is Blender itself GPL at the moment.
Blender is neat and (mostly) free, but I think I'll pass on this one for awhile.
(See Parent)
Teehee! Well put... I too have noticed GNOME evolve into something that almost seems to have been developed and marketed from Redmond. Interesting how a group's tolerance levels shift over time. But then again, it was only about 24 months ago when the average distro (with all the goodies installed and running) was still zippy on a 486/100 or P60.
Me? I use XFCE, but find KDE to be nifty as well. My next box will probably run either Blackbox or KDE most of the time.
Or how much they apparently pay per month:
basicly $0.59/second = $1.5 Million/month
Guess he's lucky he's only being sued for 1/3 of a month of internet access?? : )
(6.4 was almost a rogue port of 6.2 to support the 10k proc)
The R10K (MIPS R10000) was supported as far back as "IRIX 5.3 Including R10K" in 1995. But I agree about 6.4 (as well as 6.3 for O2)... rogue port indeed. Luckilly 6.5 has been great, especially the quarterly updates that slowly roll in new features and fix the bugs. Heavily tested, too (SGI stays about 1.5 quarters ahead of their users, heavily testing each new release on their own machines first).
I [heart] IRIX 6.5
(now at 6.5.12!)
Good point... even if these can get out to homes at reasonable prices, this is gonna be worse shared bandwidth than cable modem technology; cables are 4mbps average (around here anyway), sharing a couple T3's or maybe an OC3 for larger area isn't that bad - but one of these copper links is the speed of just one of those kinda-expensive OC3's, and i don't think the phone companies are going to be too eager to install huge pipes to fully take advantage of a 155mbps link to each home...
VERY wel put! The nearby ISP here pays $13,000 per month for each of their upstream DS3/T3 connections (45mbps to UUNET/Worldcom). OC3 is 155mbps and is in the neighborhood of $35,000 per month.
Well put, and a good example, too. The major ISP/telco (a cooperative) in a town near me serves about 750 dialup and DSL customers. Their upstream bandwidth is pretty substantial (2x DS3 [45mbps] and 2x DS1 [1.5mbps]). I don't know what they pay for the DS1s, but both DS3s are connected to UUNET/Worldcom and cost $13,000 each per month to operate. Anyway, getting back to my point, this telco made the mistake of selling 2.2mbps ADSL for under $200/month. Within three weeks all of their upstream bandwidth was clogged. Customers complained but were not willing to pay any more, nor settle for cheaper, slower connections. All wanted 2.2mbps full time for $200/month.
Your situation sounds a lot like my former home. I lived in a new neighborhood in a city of 125,000 people. The only problem was, the telco didn't have enough lines running from that side of town to the central office (home of their switching equipment) on the other, older, side of town. So, they ended up heavily multiplexing the telephone lines. Even though we lived only a mile from a nearby telephone building, we couldn't connect to the ISP (5 miles away) above carrier 19200 - 28800. And that was *after* the telco conditioned the line for us. We ended up getting wireless internet via a 24db antenna pointed towards my ISP's antenna on the top of a nearby (2 miles away) office building.
Not all telephone lines are CAT3... my home (built in 1994) has two-pair (4 conductors) and about 8 twists per foot (somewhere in the middle of cat3 and cat5). My older home had flat (no twists) two-pair running in the walls. I have no idea what sort of cable comes from the telco to the pedestal in my yard, but from the pedistal to the box on the side of my house is six-pair (12 conductors) with probably 8 twists per foot and it's thick (22 guage versus the 24 guage normally used for telephone and network wiring).
Why does Slashdot keep using the old cube logo for SGI? The glory days of SiliconGraphics (and thus, the cube logo) are long gone. The x86/PC days are here along with the fitting 'sgi' logo. Quit using the cube.
Write (and GPL!) the apps you've always wanted. I wish I was retired...
Looking at the article:
0 3C LIM.html
http://archives.nytimes.com/2001/07/03/science/
I can see that they're standing beside many racks of SGI Origin 3000 gear (whether it's a single machine or many smaller machines depends on the cabled configuration -- O3K uses a mesh of cables rather than a backplane for its third-generation ccNUMA architecture).
At any rate, I'm curious as to who company they'll be buying their new system from. The only clue I can gather is their mention of 5120 CPUs to churn out 40 TFLOPS... that's 7.8 GFLOP per CPU which pretty much rules out SGI or even the latest Pentium/Xeon, Itanium, UltraSPARC III, or Alpha (unless one uses a very Apple-esq SIMD benchmark but I would imagine they want and need something more flexible for true performance with SISD, SIMD, and MIMD combined, rather than just bragging rights for a high single benchmark and top spot on Top500). My guess is they're going with either the not-yet-released Cray SV2 (which will combine the parallel strengths of the T3E with the vector strenghts of the SV1ex) or NEC's successor to the SX-5 (SX-5 is marketed by Cray in the USA, but under its creator, NEC, in other countries).
Anyone have additional insight?
...to blame it all of the damned butterflys...
what about ClarisOffice (which today is known as AppleOffice)
You mean ClarisWorks / AppleWorks. AppleWorks was originally available for the Apple ][ series of machines in the late 1970s... and is now available for Mac OS X. As far as the "Office" name, I don't know how far back it goes, but the software for the Lisa (Apple's original GUI machine in 1982 - 1984) was Lisa Office System. The Macintosh shipped in 1984 with apps such as MacWrite, MacPaint, and MacDraw (all Apple apps). Microsoft made a Mac GUI version of Word later that year.
Amen for that. Living near the Dallas area (it's no Silicon Valley, but it did have its share of dot cons) I've seen my share of $200 SGI Octanes, $300 Origins, and $600 Onyx2s. You're not kidding about having to beat the dealers, dotcon closeouts are often a total madhouse. Especially when the original vendor of the still-under-warranty equipment shows up to buy back their goods for pennies on the millions (only to resell them as "remanufactured" for 350x the firesale price).
Schweeet! I'm already wanting to order one.... truly the first cube computer I'd ever buy!
How will this affect SGI?
That may sound like an odd question, but if you look at SGI's Linux offerings over the past 12 month, you'll see that they resemble a (poor) attempt at being a VA Linux wannabe.
Also note that some of SGI's Linux servers were OEMed from VA.
It will be interesting to see how this affects SGI over the next 6 months... will SGI's Linux offerings be better or worse as a result?
First of all, it proves how big business (big scared businesses) can drive the little guy out of his own business.
It also shows that the future of hardware is -not- from commerical organizations, but rather from truly open, not-for-profit vendors. The only way we can continue to grow is to abandon the old idea that somone has to get rich by selling things at an inflated price. VA has proven that a vendor need not make a profit to better the world. Expect the concept of free, open hardware (and not just open software) to grow over the next few years.
After reading this:
3 281,00.html
.NET. I'm not sure if I should laugh or be scared.
http://www.zdnet.com/zdnn/stories/news/0,4586,509
(yes, that's right, Microsoft and others will spend a collective $1,000,000,000.00 marketing Windows XP)
I have to wonder how much they'll market
Get ready to fight the FUD.
...and nothing more... but as a commerical software vendor with your source totally closed, you're able to integrate GPLed code without anyone knowing about it. I'd keep a close eye on Sun over the next 12 months.
Will this become part of Red Hat Professional Server? I've heard a lot of good things about PostgreSQL and am very happy that Red Hat is going to start working with it! I use RHPS on a daily basis and will buy the next version even more quickly if Red Hat DB/Red Hat PostgreSQL will be included.
Very sweet news!!!