The current generation of SGI NUMAflex based machines use a mesh of full duplex 3.2 GByte/sec interconnects. That's 25.6 Gbit/sec.
That's way more than 3 times. Plus the latency is several orders of magnitude less.
The tradeoff is cost. A fully populated rack (32 Itanium2 CPUs or 128 MIPS R1x000 CPUs) starts at $1M can can easily run upwards of $4M. If your task is CPU bound, then a homebrew cluster will be almost as good. If your task is I/O bound, you can't beat the Origin. Until the Cray X1 ships, anyway.
Also keep in mind that while an Origin system can be partitioned, they are typically run as one single image system. The beasts easily expand from 2 CPUs up to 512 (even 1024 with special support from SGI). The cross-system memory latency increases with the larger configurations, but the net cross-section bandwidth/thruput increases linearlly with the CPU count. Very efficent design.
Pretty sweet machine. Again, until the Cray X1 ships!:)
If you're in the Silicon Valley you'll want some industrial or retro funiture, but if your company is located anywhere else, you'll want the "me too" look that only Aeron chairs can provide.
I know we're just joking about these requirements, but they're scarily familar to three NOCs I've been involved with. I think there must be some unwritten ruleset that goes something along these lines: 1) 50% of NOC budget must be spent on funriture and flat-panel displays. 2) Trendy lighting in NOC must seriously interfere with trendy displays. (example: if room is equiped with halogen spot lighting, at least one non-movable light should be aimed at a projection screen). 3) NOC must be located in the most inconvenient area of the most inconvenient building. 4) Actual NOC computers must be running the latest, untested wiz-bang buggy software on the latest, untested wiz-band buggy hardware. 5) Half of the NOC staff must be completely unskilled, impersonal, and unwashed. 6) The other half othe NOC staff must be anal, uptight, and permanently pissed off.
Server Room layout is another story... but does match rule #2 quite well... the perfect server room is often located as far from the building's loading dock as possible. With a proper pallet jack, it should take at least 30 minutes to haul a crated SGI Origin or Sun Enterprise server from the loading dock to the server room. Smaller items should take no less than 15 minutes. Shaky ramps, cramped elevators, and narrow hallways are a plus.
You probably mean ENIAC, the huge digital beast built for the Army at the University of Pennsylvania. It was something like 30 tons and had roughly 20K vacuum tubes. Slower than your TI calculator, but being able to do 5000 ops per second wasn't too shabby for 1945.
UNIVAC, on the other hand, was a successful commercial computer (though only after Mauchly and Eckert sold out to Remington-Rand... the company that later became the LZW/GIF bastards, Unisys). The various UNIVACs fit into a single (large) enclosure and had a snazzy operators console. Very SciFi looking. Again, slow by todays standards, but quite a speed demon back in the day.... 40kbps tape storage, 2.25 MHz logic units (still vacuum tube based), sustained performance of over 100,000 ops per second.
Hmm, now as I proof-read, I realize you *are* talking about the UNIVAC-1. I'll post this anyway, heh.
Did anyone ever open up Field and Stream again?... last I heard it had been closed for awhile. Minot also has Applebees, which spread thruout North Dakota like the plague in the early 1990s.
And then there's the almost monopolistic hold on the Bismarck/Minot grocery stores by the Barlow family...
All in all, it's a pretty nice area. But only if you like peace and quiet. It's not a place for the "d00d wherez da scene??" crowd.
I know, of course that OS X allows multiple users, and its support for multimedia is much better, but does it allow them simultaneously?
Yes, but not the way you describe. Mac OS X, with its unix guts, allows for many folks to be logged in at one time... remotely. But only one person can be logged in at the graphical console at once. A quick check shows 4 (idle) folks ssh'ed into our Mac OS X server (an xServe running 10.2).
That's one are where Microsoft has the Mac (and Linux/Unix) world beat. But it shouldn't be too hard to hack in a rough equivalant. Sucks a lot of ram, though... but I guess that can be paged, considering the session is pretty much frozen.
Maybe you should clarify that you're talking about 8 bit and 16 bit *per component* not per pixel. 48 bit per pixel color and heavy lifting has been an SGI MIPS/IRIX strong point for a long time. Some folks use the 18 wheeler analogy... a high-RPM sports car will beat a big diesel 18 wheeler in a drag race... unless each is pulling a 60 ton trailer.
The newest SGIs support USB and a few (maybe just O3K?) have firewire. On the big iron, USB has been a *really* nice upgrade from PS/2, especially on machines configured with multiple graphics pipes and multiple users. Rather than installing extra BaseIO modules for additional PS/2 hookups for additional users, you can now just plug in as many keys/mice as you'd like, bind each set to a certain number of graphics pipes. Helps us keep our Onyx 3800 flexible... most of the time it's running each of its three graphics pipes seperately... we have a config that'll drive three sets of keys/mice for three users, one graphics pipe driving two monitors per user. But when we need the power, we have one user driving all three pipes on a single multi-projector panoram screen. It's not totally plug and play, but it's a lot easier than it used to be.
He should have included Apple's x86 version of Rhapsody (developer release 1 or 2 of Mac OS X from several years ago). Either that or Darwin x86, which is available from Apple's website.
InfiniteReality 4 can be ordered with up to 11 GB of "gfx ram". 1 GB is dedicated for textures, the remaining 10 GB is for the various buffers (frambuffer, puffers, etc).
Mac just released OS X.2 and while it is BSD and not really UNIX
Who's Mac? I think you mean Apple (or really, NeXT). BSD is very much pure UNIX, it's Berkeley's fork of AT&T's UNIX from long time ago. BSD has been a major influence on "modern" unices, take a look at some of the UNIX family trees / timelines. Though SunOS 5.x (Solaris 2.5 and newer) has a lot of AT&T System V in it these days, it was once almost pure BSD... as were most west coast flavors of UNIX.
IRIX is alive and kicking, it gets a major overhaul every quarter, as does most of its major subcomponents. Over the past 24 months, revisions of IRIX 6.5 have rolled in seamless support for SN2MIPS (O3K/O300/Fuel), VPro/Odyssey graphics, InfiniteReality3&4 graphics, as well as a major compiler and runtime upgrade (MIPSpro 7.3), clustered/distrbuted filesystem (CXFS), and gobs of media and gl libraries. The help system was recently totally overhauled, CXFS continues to get major upgrades (and more cross platform support). Recently announced were still more major overhauls... MIPSpro compilers will be upgraded again in October, Performer is getting a huge upgrade by the end of the year, Java is being beefed up, the freeware.sgi.com archive is growing. Plus there are a lot of things that just work... I wish other vendors had *half* of the performance profiling tools that IRIX includes. SGI kit isn't cheap, and parts of it aren't too flashy, but it's hard to beat if you need the torque.
The version numbers don't increment much, but there's been a lot of behind-the-scenes work and gobs of improvements addressed across the board. The rest of the industry has been moving so fast that one no longer needs an SGI to do most 3D or video work, but should you need several pipes of graphics for a simulator or need to shuffle several streams of uncompressed 1080i, then there's no better platform. It's a tough industry, and SGI's financials suck, but they're not doing too bad.
Yeah, I wish SGI would update their desktop a bit more, but then I also wish Cray would implement some major changes in UNICOS userland too... but it looks like I'll have to wait for both.
I've noticed that the Wintel world loves to use "detection" in many of the buzzwords (such as motion detection)... Occlusion Detection may be slightly more familar to you as Occlusion Culling or just the "Cull" step in the render pipeline. It's nothing new, in fact the time saving step is what allowed oldschool 3D hardware (such as an SGI RealityEngine) to obtain decent performance back in the day.
Yep, Gibson writes gui Win32 windows apps in pure x86 assembly. He's nuts, but his apps are tiny and run fast. Lots of good resources there.
New meaning to "Red" in Red Hat
on
KDE Gets The Hat
·
· Score: 2
*Sigh* I really wish Debian was a bit more with it... I think I may try Sarge soon.
And what's up with the X Window System?
on
KDE Gets The Hat
·
· Score: 2, Funny
Command-line console utilities and the various shells are still far from perfection... so why are all of the lemmings moving to GUIs? They're just slow and inefficent anyway! NeWS was bad, X11 is worse. Windows and Mac OS are even less flexible.
If you have to make an X app, please do us all a favor and use "clean" straight xlib, stay away from the bloat of Motif, GTK, and Qt.
choice / customization is a *GOOD* thing
on
KDE Gets The Hat
·
· Score: 2
I fully agree... this is why I use opensource in the first place. In fact, I only know of one or two power users that *don't* customize the daylights out of their work environment anyway. I like my distros to do what they were intended to do -- give me an easy way to install the latest goodies in a semi-stable state. I want the distro to leave out the politics and as much preconfiguration as possible --- I want to make the choices, not some committee.
The way I do things: pick the right tool for the right job and after some use, customize (interface, menus, scripts, modules, filesystem layout, etc) to best fit my needs. Linux works great for most of my needs, though I also use SGI, Sun, Mac, and Wintel systems for specific tasks.
I haven't been keeping up with Myrinet... until they license their design to other manufacturers and/or drop their prices significantly, I'm not interested. If I wanted to be locked into a product built by only one company, I would have bought a Cray or SGI in the first place.
I'm interested in doing something very similar for my new home, but I would like to have HD capability right off the bat. There are already many satellite stations broadcasting HD material, and I'd like to make use of that... especially since most of my monitors support HD resolutions. Aside from my Sony Wega in the living room, every other display is on a Mac or PC. I already have Cat5e to most rooms and have a nice managed 3com switch bringing it all together.
There have been many good replies to this thread, though most are talking theory, some experimental at best. A camera module with the ability to capture 12,000 high resolution frames per second is bound to cost a fortune, and I really doubt there will be much competition for a long time. Perhaps a cheaper alternative would be to purchase several currently-available high speed CCD/CMOS camera modules and use a series of mirrors and lenses to allow the cameras to work together in a round-robin fashion to achieve the a much higher framerate. This would certainly keep the project from being locked into proprietary hardware -- be it a single interface type, manufacturer, or other monopolistic attribute. The idea of "parallel" items is nothing new, we've already seen success with drives, clusters, and even an array of projectors to create a high resolution projected wall. Just a thought...
As I pointed out in another post, a good option would be to use several gigE interconnects to connect the CCD module to the many nodes of a beowulf cluster. Besides, you're going to need a cluster to manage that much data anyway.
Slashdot Mirror
The current generation of SGI NUMAflex based machines use a mesh of full duplex 3.2 GByte/sec interconnects. That's 25.6 Gbit/sec.
:)
That's way more than 3 times. Plus the latency is several orders of magnitude less.
The tradeoff is cost. A fully populated rack (32 Itanium2 CPUs or 128 MIPS R1x000 CPUs) starts at $1M can can easily run upwards of $4M. If your task is CPU bound, then a homebrew cluster will be almost as good. If your task is I/O bound, you can't beat the Origin. Until the Cray X1 ships, anyway.
Also keep in mind that while an Origin system can be partitioned, they are typically run as one single image system. The beasts easily expand from 2 CPUs up to 512 (even 1024 with special support from SGI). The cross-system memory latency increases with the larger configurations, but the net cross-section bandwidth/thruput increases linearlly with the CPU count. Very efficent design.
Pretty sweet machine. Again, until the Cray X1 ships!
487227651.28491117
Works great on my iBook. It's overpriced, but it works better for me than Gimp or Corel PhotoPaint 10.
I bought the boxed, retail version of Warcraft III. It included a Mac version which runs great on my iBook.
If you're in the Silicon Valley you'll want some industrial or retro funiture, but if your company is located anywhere else, you'll want the "me too" look that only Aeron chairs can provide.
I know we're just joking about these requirements, but they're scarily familar to three NOCs I've been involved with. I think there must be some unwritten ruleset that goes something along these lines:
1) 50% of NOC budget must be spent on funriture and flat-panel displays.
2) Trendy lighting in NOC must seriously interfere with trendy displays. (example: if room is equiped with halogen spot lighting, at least one non-movable light should be aimed at a projection screen).
3) NOC must be located in the most inconvenient area of the most inconvenient building.
4) Actual NOC computers must be running the latest, untested wiz-bang buggy software on the latest, untested wiz-band buggy hardware.
5) Half of the NOC staff must be completely unskilled, impersonal, and unwashed.
6) The other half othe NOC staff must be anal, uptight, and permanently pissed off.
Server Room layout is another story... but does match rule #2 quite well... the perfect server room is often located as far from the building's loading dock as possible. With a proper pallet jack, it should take at least 30 minutes to haul a crated SGI Origin or Sun Enterprise server from the loading dock to the server room. Smaller items should take no less than 15 minutes. Shaky ramps, cramped elevators, and narrow hallways are a plus.
I have the original UNIVAC in my spare house
You probably mean ENIAC, the huge digital beast built for the Army at the University of Pennsylvania. It was something like 30 tons and had roughly 20K vacuum tubes. Slower than your TI calculator, but being able to do 5000 ops per second wasn't too shabby for 1945.
UNIVAC, on the other hand, was a successful commercial computer (though only after Mauchly and Eckert sold out to Remington-Rand... the company that later became the LZW/GIF bastards, Unisys). The various UNIVACs fit into a single (large) enclosure and had a snazzy operators console. Very SciFi looking. Again, slow by todays standards, but quite a speed demon back in the day.... 40kbps tape storage, 2.25 MHz logic units (still vacuum tube based), sustained performance of over 100,000 ops per second.
Hmm, now as I proof-read, I realize you *are* talking about the UNIVAC-1. I'll post this anyway, heh.
Did anyone ever open up Field and Stream again?... last I heard it had been closed for awhile. Minot also has Applebees, which spread thruout North Dakota like the plague in the early 1990s.
And then there's the almost monopolistic hold on the Bismarck/Minot grocery stores by the Barlow family...
All in all, it's a pretty nice area. But only if you like peace and quiet. It's not a place for the "d00d wherez da scene??" crowd.
Hooray for the Dakota Carrier Network!
Although I don't like the review itself - I think it is a bit dry and not very journalistic
Unlike, of course, the fine prose found on Slashdot.
I know, of course that OS X allows multiple users, and its support for multimedia is much better, but does it allow them simultaneously?
Yes, but not the way you describe. Mac OS X, with its unix guts, allows for many folks to be logged in at one time... remotely. But only one person can be logged in at the graphical console at once. A quick check shows 4 (idle) folks ssh'ed into our Mac OS X server (an xServe running 10.2).
That's one are where Microsoft has the Mac (and Linux/Unix) world beat. But it shouldn't be too hard to hack in a rough equivalant. Sucks a lot of ram, though... but I guess that can be paged, considering the session is pretty much frozen.
Maybe you should clarify that you're talking about 8 bit and 16 bit *per component* not per pixel. 48 bit per pixel color and heavy lifting has been an SGI MIPS/IRIX strong point for a long time. Some folks use the 18 wheeler analogy... a high-RPM sports car will beat a big diesel 18 wheeler in a drag race... unless each is pulling a 60 ton trailer.
The newest SGIs support USB and a few (maybe just O3K?) have firewire. On the big iron, USB has been a *really* nice upgrade from PS/2, especially on machines configured with multiple graphics pipes and multiple users. Rather than installing extra BaseIO modules for additional PS/2 hookups for additional users, you can now just plug in as many keys/mice as you'd like, bind each set to a certain number of graphics pipes. Helps us keep our Onyx 3800 flexible... most of the time it's running each of its three graphics pipes seperately... we have a config that'll drive three sets of keys/mice for three users, one graphics pipe driving two monitors per user. But when we need the power, we have one user driving all three pipes on a single multi-projector panoram screen. It's not totally plug and play, but it's a lot easier than it used to be.
He should have included Apple's x86 version of Rhapsody (developer release 1 or 2 of Mac OS X from several years ago). Either that or Darwin x86, which is available from Apple's website.
InfiniteReality 4 can be ordered with up to 11 GB of "gfx ram". 1 GB is dedicated for textures, the remaining 10 GB is for the various buffers (frambuffer, puffers, etc).
> Uhh...O3ks are SN1 machines, not SN2 unless I'm
> missing something. As the O2k are SN0 machines.
That's a good point. IRIX is doomed.
Mac just released OS X.2 and while it is BSD and not really UNIX
Who's Mac? I think you mean Apple (or really, NeXT). BSD is very much pure UNIX, it's Berkeley's fork of AT&T's UNIX from long time ago. BSD has been a major influence on "modern" unices, take a look at some of the UNIX family trees / timelines. Though SunOS 5.x (Solaris 2.5 and newer) has a lot of AT&T System V in it these days, it was once almost pure BSD... as were most west coast flavors of UNIX.
IRIX is alive and kicking, it gets a major overhaul every quarter, as does most of its major subcomponents. Over the past 24 months, revisions of IRIX 6.5 have rolled in seamless support for SN2MIPS (O3K/O300/Fuel), VPro/Odyssey graphics, InfiniteReality3&4 graphics, as well as a major compiler and runtime upgrade (MIPSpro 7.3), clustered/distrbuted filesystem (CXFS), and gobs of media and gl libraries. The help system was recently totally overhauled, CXFS continues to get major upgrades (and more cross platform support). Recently announced were still more major overhauls... MIPSpro compilers will be upgraded again in October, Performer is getting a huge upgrade by the end of the year, Java is being beefed up, the freeware.sgi.com archive is growing. Plus there are a lot of things that just work... I wish other vendors had *half* of the performance profiling tools that IRIX includes. SGI kit isn't cheap, and parts of it aren't too flashy, but it's hard to beat if you need the torque.
The version numbers don't increment much, but there's been a lot of behind-the-scenes work and gobs of improvements addressed across the board. The rest of the industry has been moving so fast that one no longer needs an SGI to do most 3D or video work, but should you need several pipes of graphics for a simulator or need to shuffle several streams of uncompressed 1080i, then there's no better platform. It's a tough industry, and SGI's financials suck, but they're not doing too bad.
Yeah, I wish SGI would update their desktop a bit more, but then I also wish Cray would implement some major changes in UNICOS userland too... but it looks like I'll have to wait for both.
I've noticed that the Wintel world loves to use "detection" in many of the buzzwords (such as motion detection)... Occlusion Detection may be slightly more familar to you as Occlusion Culling or just the "Cull" step in the render pipeline. It's nothing new, in fact the time saving step is what allowed oldschool 3D hardware (such as an SGI RealityEngine) to obtain decent performance back in the day.
http://grc.com/smgassembly.htm
Yep, Gibson writes gui Win32 windows apps in pure x86 assembly. He's nuts, but his apps are tiny and run fast. Lots of good resources there.
*Sigh*
I really wish Debian was a bit more with it... I think I may try Sarge soon.
Command-line console utilities and the various shells are still far from perfection... so why are all of the lemmings moving to GUIs? They're just slow and inefficent anyway! NeWS was bad, X11 is worse. Windows and Mac OS are even less flexible.
If you have to make an X app, please do us all a favor and use "clean" straight xlib, stay away from the bloat of Motif, GTK, and Qt.
I fully agree... this is why I use opensource in the first place. In fact, I only know of one or two power users that *don't* customize the daylights out of their work environment anyway. I like my distros to do what they were intended to do -- give me an easy way to install the latest goodies in a semi-stable state. I want the distro to leave out the politics and as much preconfiguration as possible --- I want to make the choices, not some committee.
The way I do things: pick the right tool for the right job and after some use, customize (interface, menus, scripts, modules, filesystem layout, etc) to best fit my needs. Linux works great for most of my needs, though I also use SGI, Sun, Mac, and Wintel systems for specific tasks.
I haven't been keeping up with Myrinet... until they license their design to other manufacturers and/or drop their prices significantly, I'm not interested. If I wanted to be locked into a product built by only one company, I would have bought a Cray or SGI in the first place.
I'm interested in doing something very similar for my new home, but I would like to have HD capability right off the bat. There are already many satellite stations broadcasting HD material, and I'd like to make use of that... especially since most of my monitors support HD resolutions. Aside from my Sony Wega in the living room, every other display is on a Mac or PC. I already have Cat5e to most rooms and have a nice managed 3com switch bringing it all together.
There have been many good replies to this thread, though most are talking theory, some experimental at best. A camera module with the ability to capture 12,000 high resolution frames per second is bound to cost a fortune, and I really doubt there will be much competition for a long time. Perhaps a cheaper alternative would be to purchase several currently-available high speed CCD/CMOS camera modules and use a series of mirrors and lenses to allow the cameras to work together in a round-robin fashion to achieve the a much higher framerate. This would certainly keep the project from being locked into proprietary hardware -- be it a single interface type, manufacturer, or other monopolistic attribute.
The idea of "parallel" items is nothing new, we've already seen success with drives, clusters, and even an array of projectors to create a high resolution projected wall.
Just a thought...
As I pointed out in another post, a good option would be to use several gigE interconnects to connect the CCD module to the many nodes of a beowulf cluster. Besides, you're going to need a cluster to manage that much data anyway.