Even SGI Reality engines (including infinite reality) used this method, although I think you have up to four raster managers. The hitch was you wouldn't get the extra texture memory for the reasons pointed out in other posts. 64MB was the limit.
Good point, you're right. Though there was more texture ram as time went on. InfiniteReality 3 had 256 MB texture ram per raster manager board, InfiniteReality 4 has 1 GB texture ram per board. (Plus 2.5 GB of general purpose framebuffer per board).
ATI's Windows drivers are offically updated once in awhile, and are generally rock solid, but there are occasionally problems that aren't resolved for months at a time.
I meant to say NVIDIA.
ATI updates their WHQL Windows driver about once a month, somtimes more often than that.
NVIDIA updates their offical driver once in awhile. Thus occasional problems aren't resoloved for months at a time.
Both companies make good cards and good drivers these days... so buy what you can afford. Personally, I love my Radeon 9600 Pro. But my GeForce 4 Ti 4200 is still a good performer too.
parallel graphics pipelines
on
Video Card History
·
· Score: 3, Informative
It is possible to scale performance that way, but the result will be less than double the frame rate, simply because the time to generate a frame does not scale linearly with resolution.
To do 60 frames per second, you have roughly 16ms to generate a frame. A couple of those ms will be gobbled up with I/O transactions and various wait states, so you're already at the point where double the power is only going to result in 1.75x the performance. This will also be highly dependant on how well the 3D code can be parallelized (are there a lot of read_pixels callbacks that require both GPUs and both banks of memory to talk to each other? etc).
This has actually been done by SGI for awhile now. A couple years ago they took their Origin 3000 architecture and stuck on dozens of V12 GPUs and tiled the graphics for higher performance. That concept has been tweaked for their Onyx4 sytems... one large single computer with up to 34 ATI FireGL X1 GPUs. 16 GPUs work on each display in a 4x4 grid. Each GPU generates its 400x300 piece and 16 of those are composited in real time to make up a 1600x1200 display. I believe the biggest such machine to date has 32 GPUs powering two zippy fast 1600x1200 displays and 2 GPUs driving an additional 4 lesser powered displays. SGI gets quite a speedup by doing it that way, with 16 GPUs per display, but there's also a lot of overhead (even more in SGI's case, with 34 AGP 8X busses in said system). Their implementation of OpenGL and OpenGL Performer is tweaked for this, though.
So yeah, it can be done, but the fact that the GPUs will spend a significant amount of time doing non-rendering tasks (I/O, waiting for data, copying the final result to frame buffer, etc) means that you won't see a nice linear scaling. The cost of making custom hardware and custom drivers also adds up. With top-end PC 3D accelerators costing $400 already, I can't picture many users shelling out $1000+ for a dual GPU card.
Forceware -- the name sounds evil, in a bad way
on
Video Card History
·
· Score: 1
Is it just me or does Forceware sound like a copy of MS Office shoved in yer face at work/school?
I'm happy with ATI & Linux
on
Video Card History
·
· Score: 2, Informative
ATI generally releases an new WHQL Windows driver about once a month and a new Linux driver about every 6 weeks. I've had no problems with their XFree86 4.3 driver. They don't have a FreeBSD driver, though, but I guess a PowerBook would give somewhat of the same experience (BSD-based OS, XFree86-based X envrionment, Radeon 9600, plus Quartz/DisplayPDF and access to Mac apps). Mac OS X also has the ATI (and nVIDIA) drivers built-in and are updated with the software update utility.
ATI's Windows drivers are offically updated once in awhile, and are generally rock solid, but there are occasionally problems that aren't resolved for months at a time.
Pretty damn good at the time. The bundled h/w accelerated version of Descent was mind boggling!
The S3 ViRGE also suffered from poor drivers in its later days. I really wish I knew just how well it could have stacked up against the Voodoo. I remember the patched version of Descent you mentioned! The software version looked good, the hardware version was amazing! There were also some other 3D demos on the bonus CD, one was a walk thru a flame torch - lit castle. Probably looks like crap by todays standards, but back then it was our own low-resolution personal SGI Onyx!
The Permedia 1 was also a fun card. Not too great with later games, but it ran Quake2 and the OpenGL version of Quake1 really well. It was also one of the few cards of its day to run most OpenGL routines quite well (as opposed to the horribly unbalanced mix of slow and fast that the consumer cards were before the TNT and RagePro).
I think what finally brought 3-D graphics acceleration into the mainstream was the introduction of graphics card chipsets that could combine decent 3-D acceleration with fast 2-D graphics all at once.
ATI's RageII/RageII+/RageII+DVD, the forerunner to the RagePro, was a decent all-in-one performer for certain games. The drivers were AWFUL, though, so its hard to tell if the problems were with the silicon itself. I don't think there was ever even a DVD Player that made use of the RageII+DVD. ATI's head was far up its rump back then.
The first wave of Beige Apple G3s and the very first revision of the iMac used the RageII+. It was fine for a few games (Nanosaur, Bugdom, LAPD FutureCop, Quake1) but even with the texture ram upgrade simm, it wasn't enough to run Quake2 or Quake3 with any decent performance. The original Unreal, which ran fine on a Voodoo1, was painfully slow on RageII.
Rage128 was such a breath of fresh air! Zippy fast Quake3, butter-smooth Heavy Metal:FAKK2.
Re:Confusion with later Voodoo cards?
on
Video Card History
·
· Score: 2, Informative
Which is a pretty simple way to get double the performance. I wonder why noone's done this recently
They are, in the chip itself, sorta. Modern all-in-one GPUs have multiple texture pipelines, which does split some of the load on the silicon level. It's not SLI, but it's the same concept.
The problem is SLI only doubles the fillrate. Both 3D chipsets need to work on the exact same dataset. SLI was a great boost back when 3D hardware was slow at texturing. These days the hardware can pump a couple thousand frames per second worth of textures, it's fancy multipass rendering and dynamic shaders (and to some extent, the geometry) that take up all of the frame generation time. SLI could speed some of this up, but it wouldn't help with most of the bottlenecks. It would be like putting new tires on a car that needs an engine tuneup.
ScanLine Interleaving. To get over the fillrate bottleneck, one card pumped pixels for even-numbered scan lines, the other worked on the odd-numbered scan lines. Back in the days when a dual PII/400 and dual Voodoo2 was the gamer's ultimate machine. There were even a few companies that stuck two Voodoo2 chipsets on a single card.
A lot of professional/expensive 3D systems before the Voodoo2 used a similar technique. If one of the texture ram modules comes loose on an SGI Indigo2 MaximumImpact, textured models will suddenly lose half of their scanlines!
"Video Cards" started in 1996?
on
Video Card History
·
· Score: 2, Interesting
Does anyone else notice that this "Video Card" history starts off with about the 3rd consumer 3D accelerator? They didn't even mention the groundbreaking Rendition Verite. Nor any of the non-PC 3D systems that came before it (Jim Clark / SGI's Geometry Engine based systems in 1983 or the image processors from Evans & Southerland).
And if it's a Video Card history, why no mention of EGA/CGA?
Sounds more like "the 3D accelerator world since the Voodoo" history. It's articles like this that make me wish the slashdot editors would remember they have some readers that are older than high school age.
My guess, especially given the offical reply from McD, is that they are planning something like this, just not with Apple. I wouldn't be suprised to hear that they'll be giving away 1 Billion MusicMatch or Napster songs.
It's a 32 processor SGI Origin 2000 / Onyx2 with InfiniteReality graphics. Two full racks of Origin 2000 = 32 processors, and half a rack of Onyx2 = 2 graphic pipelines.
Late 1997, early 1998 technology. SGI is currently selling Origin 3900 and Onyx4. Origin 4000 is rumored to be demoed at SC2003 later this month.
My personal opionion is that google and netscape are very different if for no reason other than the fact that netscape consistently put out defective products. It wasn't until the very end (4.71?) that netscape finally got rid of most of the crashes and memory leaks.
Netscape was OK up to and including the 2.02 release. It was all downhill from 3.0 on. I switched to MSIE 3.0 on Windows (remember that swirly background pattern behind the buttons?) and the equiv MSIE release for Mac OS. I stuck with Netscape Navigator only on Unix.
I've used 4.79 and 4.8, they're still awful, though a tad faster on old hardware than even Mozilla 1.5. (But, in the end, I still use Mozilla/MozillaFirebird is it doesn't segfault every 15 minutes).
These days I run Mozilla/MozillaFirebird on everything but OS X, where I use Safari.
I am speaking from experience when I tell you that building a large cluster from desktops is just not a good way to go. They take up a hell of a lot more room, they put out a lot more heat, and the remote management capabilites are degraded.
Desktops take up more room, correct. And yes, the desktop G5 does not have a console serial port like the xServe does. But seriously, how many modern clusters do you see with a terminal server connecting to each of the node's serial port? These days it's all install-and-run. OS X is UNIX... you can do a lot with a remote shell. These folks will never need to sit down at a GUI for each node. If you look at their setup photos, you'll see that they even removed the gfx card from each node.
And... desktops DO NOT put out more heat that a similar rackmount unit. The hard drives are the same, the processors are the same. A larger case does not create more heat. More heat may be expelled due to better fans, but that is a GOOD THING, you don't want your board, ram, and processors to cook. The only difference between the two is the power supply. Slim rackmount machines generally have smaller power supplies. But, with modern switching power supplies, there is nearly no difference in power consumption (and, by the laws of thermodynamics, heat output).
Once you go rack, you never go back. I much prefer a rack of 1U units that are built to be used in cluster situations.
Yes and no. A rack of 1U servers is small, compact, snazzy looking, and neat. But, you also increase the number of processors per square foot, which can be a cooling issue. With a concentration of heat in that area, more cool air will need to be directed to the rack.
I guess VT also has the luxury of running CPU intensive tasks. Those machines can only 8 GB RAM while other offerings can hold 16 GB and if they start to swap....ouch, not having SCSI drives will hurt.
4 GB per processor is pretty good for the current HPC world. A lot of monster supercomputer are still sold with 2 - 4 GB per processor. The G5 can unoffically support 16 GB via 2 GB DIMMs, but Apple has not certified this. SCSI drives are great for a big RAID, fibrechannel is even better. But for the drive in each node, IDE is fine. Even Google uses IDE drives in their nodes (which they use as a distributed filesystem too!).
All in all this setup is very impressive when just considering CPU performance. Wonder what is going to happen when a proffessor needs to run a few hundred jobs that use 10 or so GB of RAM each.
The prof will have to re-write his code to use less ram per processor. This is a cluster afterall, and code for clusters have to work with a fixed amount of ram per node. This is not a Cray X1, SunFire15K, or SGI Origin with high thruput, low latency global shared memory. Very very few supercomputers, and even fewer clusters, have 10 GB of ram per processor. Even 8 GB per proc is pretty rare today.
If the thread did need that much ram, it would be possible to pool memory between several nodes, it wouldn't be too fast, though (but still WAY faster than swapping to any harddrive). I believe they're currently getting a little over 800 MBytes/sec real-world thruput via the 20gbit full duplex Inifniband interconnects.
Just FWIW, they are claiming power usage of 1.5MW for this cluster of 2200 processors. Cray just released the numbers for their upcoming Red Storm cluster with over 10,000 AMD Opteron processors, just slightly less than 2.0MW.
Ugh, this is getting old.
Red Storm, the machine by itself itself, uses 2.0MW.
Big Mac and all of its networking gear uses less than 0.75MW. The supercomputing center itself (building, air conditioning, UPS battery charging equipment, and the 1100 G5s) is fed by a 1.5MW substation feed. They're still not even maxing out the substation.
The latest, fastest Opterons (not the scaled down low-power Opteron for blade servers) consume 53 watts at full clock. PowerPC 970 @ 2 GHz consumes 48 watts. The U2 and K3 motherboard chipset on the dual G5s uses just as much power as the PowerPC 970 "G5" processors. Hell, the power supply in a dual processor G5 system is 550 watts. 550 x 1100 machines = 0.61MW.
Well, we'll see.. MS's current offerings suck ass. I recently had to install XP after I had been using Linux for years exclusively. I can tell you, I was not thrilled with the experience (pardon the pun). I spent more time trying to get rid of MS's bug laden software such as IE, Outlook, Media Player and replace them with something sane than I would trying to compile and install extra software for Linux.. and the hard part was making sure those bastard programs never popped up again in the course of my day.. because I was not given a choice to de-install them permanently.
Wow, all these years I took granted that I could surf the web without even the thought about having a slimy piece of spyware installing itself on my machine. I truly feel sorry for all of Bill Gates' minions.
It's pretty amazing what people put up with, isn't it.
90% of the computing world would be happy with Firebird, OpenOffice (or even just AbiWord), and a good webmail account. But they instead pump money into Microsoft for Windows/IE and MS Office.
This leads me to a question that maybe someone could answer... is there a good AbiWord-like word processor for un*x that DOESN'T try to be Word clone? If so, I'd be interested in trying it out.
Yeah, we're all bashing Longhorn and Microsoft's ways right now... but come 2006, Microsoft will win once again as millions of businesses and tens of millions of homes upgrade to Longhorn. Even the most hardcore geek will have Longhorn on a partition within a year of its release.
We bashed Win98, but ended up using it anyway. Ditto for Win2K and XP.
Sure, our servers will still run un*x, but as long as we keep using Windows at work for "compatibility" and "familiarity"... and a single innocent Windows box at home "just for games", Microsoft will keep winning.
This is not a flame or a troll, but just a prediction based upon the past. I would like to be proven wrong, though...
Does MS assume everyone is running their displays at 1600x1200? I thought OS X wasted screen space, but this with the taskbar, sidebar, and superdeluxebig++ window frames is even worse.
Hey grandpa, they let you have laptops at the nursing home?
Hehe! But seriously, you must not have any experience with crazy numeric meteorology folks. In a lot of similar textbooks you'll find FORTRAN listings for many of their most efficent algorithms. Many of which would be a complete bear to reproduce in C.
I've been using Panther for about 24hours and really haven't run into any bugs other than having to reinstall MS Office X.
If you haven't already, I would highly recommend filing bug reports with Apple (go to the feedback section of their OS X website). Rumor has it Apple is currently collecting bugs to fix for 10.3.1.
Also, did you do a fresh install or an upgrade. My roommate and I did fresh installs on our machines and really haven't run into any of the bugs you've reported. Moving icons and mounting servers have worked fine for us. Though we don't use IMAP or Bluetooth.
Ehh... Carmack isn't developing Doom3 on MacOS. He has been pretty vocal about his love for Visual Studio 6. However, Doom3 does not make much use of DirectX (all of the gfx use OpenGL, for example) so he has made a few builds on Mac OS X and Linux over the past couple years.
In fact, the first demo of Doom3 (and the first demo of the GeForce 3 too) was on Mac OS X as part of one of Steve Jobs's keynote speeches.
Doom3 will be available for Mac OS X... but it's not being developed on it.
The Max OS X Panther 10.3 box includes 4 CDs... three for 10.3 and it's accessories (keep in mind these three CDs include localizations for 12 languages)... and a development environment CD containing compilers, various SDKs, and the feature-filled xCode IDE.
It's a bit alien to those not used to the NeXT way, but it only took my roommate about 15 minutes to find his way around. Both of us have already converted most of our projects to xCode.
Overall, I'm quite happy with it, but I've found a few bugs. yes, I've reported at least one to apple
Fire up Safari (or click on the feedback icon) and file the rest fo the bug reports. Apple is currently culling the feedback looking for bugs to fix for 10.3.1, every little bit helps!
Slashdot Mirror
Even SGI Reality engines (including infinite reality) used this method, although I think you have up to four raster managers. The hitch was you wouldn't get the extra texture memory for the reasons pointed out in other posts. 64MB was the limit.
Good point, you're right. Though there was more texture ram as time went on. InfiniteReality 3 had 256 MB texture ram per raster manager board, InfiniteReality 4 has 1 GB texture ram per board. (Plus 2.5 GB of general purpose framebuffer per board).
ATI's Windows drivers are offically updated once in awhile, and are generally rock solid, but there are occasionally problems that aren't resolved for months at a time.
I meant to say NVIDIA.
ATI updates their WHQL Windows driver about once a month, somtimes more often than that.
NVIDIA updates their offical driver once in awhile. Thus occasional problems aren't resoloved for months at a time.
Both companies make good cards and good drivers these days... so buy what you can afford. Personally, I love my Radeon 9600 Pro. But my GeForce 4 Ti 4200 is still a good performer too.
It is possible to scale performance that way, but the result will be less than double the frame rate, simply because the time to generate a frame does not scale linearly with resolution.
To do 60 frames per second, you have roughly 16ms to generate a frame. A couple of those ms will be gobbled up with I/O transactions and various wait states, so you're already at the point where double the power is only going to result in 1.75x the performance. This will also be highly dependant on how well the 3D code can be parallelized (are there a lot of read_pixels callbacks that require both GPUs and both banks of memory to talk to each other? etc).
This has actually been done by SGI for awhile now. A couple years ago they took their Origin 3000 architecture and stuck on dozens of V12 GPUs and tiled the graphics for higher performance. That concept has been tweaked for their Onyx4 sytems... one large single computer with up to 34 ATI FireGL X1 GPUs. 16 GPUs work on each display in a 4x4 grid. Each GPU generates its 400x300 piece and 16 of those are composited in real time to make up a 1600x1200 display. I believe the biggest such machine to date has 32 GPUs powering two zippy fast 1600x1200 displays and 2 GPUs driving an additional 4 lesser powered displays. SGI gets quite a speedup by doing it that way, with 16 GPUs per display, but there's also a lot of overhead (even more in SGI's case, with 34 AGP 8X busses in said system). Their implementation of OpenGL and OpenGL Performer is tweaked for this, though.
So yeah, it can be done, but the fact that the GPUs will spend a significant amount of time doing non-rendering tasks (I/O, waiting for data, copying the final result to frame buffer, etc) means that you won't see a nice linear scaling. The cost of making custom hardware and custom drivers also adds up. With top-end PC 3D accelerators costing $400 already, I can't picture many users shelling out $1000+ for a dual GPU card.
Is it just me or does Forceware sound like a copy of MS Office shoved in yer face at work/school?
http://www.ati.com/support/driver.html
ATI generally releases an new WHQL Windows driver about once a month and a new Linux driver about every 6 weeks. I've had no problems with their XFree86 4.3 driver. They don't have a FreeBSD driver, though, but I guess a PowerBook would give somewhat of the same experience (BSD-based OS, XFree86-based X envrionment, Radeon 9600, plus Quartz/DisplayPDF and access to Mac apps). Mac OS X also has the ATI (and nVIDIA) drivers built-in and are updated with the software update utility.
ATI's Windows drivers are offically updated once in awhile, and are generally rock solid, but there are occasionally problems that aren't resolved for months at a time.
Pretty damn good at the time. The bundled h/w accelerated version of Descent was mind boggling!
The S3 ViRGE also suffered from poor drivers in its later days. I really wish I knew just how well it could have stacked up against the Voodoo. I remember the patched version of Descent you mentioned! The software version looked good, the hardware version was amazing! There were also some other 3D demos on the bonus CD, one was a walk thru a flame torch - lit castle. Probably looks like crap by todays standards, but back then it was our own low-resolution personal SGI Onyx!
The Permedia 1 was also a fun card. Not too great with later games, but it ran Quake2 and the OpenGL version of Quake1 really well. It was also one of the few cards of its day to run most OpenGL routines quite well (as opposed to the horribly unbalanced mix of slow and fast that the consumer cards were before the TNT and RagePro).
I think what finally brought 3-D graphics acceleration into the mainstream was the introduction of graphics card chipsets that could combine decent 3-D acceleration with fast 2-D graphics all at once.
ATI's RageII/RageII+/RageII+DVD, the forerunner to the RagePro, was a decent all-in-one performer for certain games. The drivers were AWFUL, though, so its hard to tell if the problems were with the silicon itself. I don't think there was ever even a DVD Player that made use of the RageII+DVD. ATI's head was far up its rump back then.
The first wave of Beige Apple G3s and the very first revision of the iMac used the RageII+. It was fine for a few games (Nanosaur, Bugdom, LAPD FutureCop, Quake1) but even with the texture ram upgrade simm, it wasn't enough to run Quake2 or Quake3 with any decent performance. The original Unreal, which ran fine on a Voodoo1, was painfully slow on RageII.
Rage128 was such a breath of fresh air! Zippy fast Quake3, butter-smooth Heavy Metal:FAKK2.
Which is a pretty simple way to get double the performance. I wonder why noone's done this recently
They are, in the chip itself, sorta. Modern all-in-one GPUs have multiple texture pipelines, which does split some of the load on the silicon level. It's not SLI, but it's the same concept.
The problem is SLI only doubles the fillrate. Both 3D chipsets need to work on the exact same dataset. SLI was a great boost back when 3D hardware was slow at texturing. These days the hardware can pump a couple thousand frames per second worth of textures, it's fancy multipass rendering and dynamic shaders (and to some extent, the geometry) that take up all of the frame generation time. SLI could speed some of this up, but it wouldn't help with most of the bottlenecks. It would be like putting new tires on a car that needs an engine tuneup.
ScanLine Interleaving. To get over the fillrate bottleneck, one card pumped pixels for even-numbered scan lines, the other worked on the odd-numbered scan lines. Back in the days when a dual PII/400 and dual Voodoo2 was the gamer's ultimate machine. There were even a few companies that stuck two Voodoo2 chipsets on a single card.
A lot of professional/expensive 3D systems before the Voodoo2 used a similar technique. If one of the texture ram modules comes loose on an SGI Indigo2 MaximumImpact, textured models will suddenly lose half of their scanlines!
Does anyone else notice that this "Video Card" history starts off with about the 3rd consumer 3D accelerator? They didn't even mention the groundbreaking Rendition Verite. Nor any of the non-PC 3D systems that came before it (Jim Clark / SGI's Geometry Engine based systems in 1983 or the image processors from Evans & Southerland).
And if it's a Video Card history, why no mention of EGA/CGA?
Sounds more like "the 3D accelerator world since the Voodoo" history. It's articles like this that make me wish the slashdot editors would remember they have some readers that are older than high school age.
[end rant]
My guess, especially given the offical reply from McD, is that they are planning something like this, just not with Apple. I wouldn't be suprised to hear that they'll be giving away 1 Billion MusicMatch or Napster songs.
It's a 32 processor SGI Origin 2000 / Onyx2 with InfiniteReality graphics. Two full racks of Origin 2000 = 32 processors, and half a rack of Onyx2 = 2 graphic pipelines.
Late 1997, early 1998 technology. SGI is currently selling Origin 3900 and Onyx4. Origin 4000 is rumored to be demoed at SC2003 later this month.
MSIE on OS X is ass. Use Safari or Camino.
MSIE on Windows is annoying without a third-party popup blocker, and even then, there are still too many security holes.
Do what I do, use Mozilla or MozillaFirebird on everything but OS X... on that platform, run Safari.
My personal opionion is that google and netscape are very different if for no reason other than the fact that netscape consistently put out defective products. It wasn't until the very end (4.71?) that netscape finally got rid of most of the crashes and memory leaks.
Netscape was OK up to and including the 2.02 release. It was all downhill from 3.0 on. I switched to MSIE 3.0 on Windows (remember that swirly background pattern behind the buttons?) and the equiv MSIE release for Mac OS. I stuck with Netscape Navigator only on Unix.
I've used 4.79 and 4.8, they're still awful, though a tad faster on old hardware than even Mozilla 1.5. (But, in the end, I still use Mozilla/MozillaFirebird is it doesn't segfault every 15 minutes).
These days I run Mozilla/MozillaFirebird on everything but OS X, where I use Safari.
I am speaking from experience when I tell you that building a large cluster from desktops is just not a good way to go. They take up a hell of a lot more room, they put out a lot more heat, and the remote management capabilites are degraded.
Desktops take up more room, correct. And yes, the desktop G5 does not have a console serial port like the xServe does. But seriously, how many modern clusters do you see with a terminal server connecting to each of the node's serial port? These days it's all install-and-run. OS X is UNIX... you can do a lot with a remote shell. These folks will never need to sit down at a GUI for each node. If you look at their setup photos, you'll see that they even removed the gfx card from each node.
And... desktops DO NOT put out more heat that a similar rackmount unit. The hard drives are the same, the processors are the same. A larger case does not create more heat. More heat may be expelled due to better fans, but that is a GOOD THING, you don't want your board, ram, and processors to cook. The only difference between the two is the power supply. Slim rackmount machines generally have smaller power supplies. But, with modern switching power supplies, there is nearly no difference in power consumption (and, by the laws of thermodynamics, heat output).
Once you go rack, you never go back. I much prefer a rack of 1U units that are built to be used in cluster situations.
Yes and no. A rack of 1U servers is small, compact, snazzy looking, and neat. But, you also increase the number of processors per square foot, which can be a cooling issue. With a concentration of heat in that area, more cool air will need to be directed to the rack.
I guess VT also has the luxury of running CPU intensive tasks. Those machines can only 8 GB RAM while other offerings can hold 16 GB and if they start to swap....ouch, not having SCSI drives will hurt.
4 GB per processor is pretty good for the current HPC world. A lot of monster supercomputer are still sold with 2 - 4 GB per processor. The G5 can unoffically support 16 GB via 2 GB DIMMs, but Apple has not certified this. SCSI drives are great for a big RAID, fibrechannel is even better. But for the drive in each node, IDE is fine. Even Google uses IDE drives in their nodes (which they use as a distributed filesystem too!).
All in all this setup is very impressive when just considering CPU performance. Wonder what is going to happen when a proffessor needs to run a few hundred jobs that use 10 or so GB of RAM each.
The prof will have to re-write his code to use less ram per processor. This is a cluster afterall, and code for clusters have to work with a fixed amount of ram per node. This is not a Cray X1, SunFire15K, or SGI Origin with high thruput, low latency global shared memory. Very very few supercomputers, and even fewer clusters, have 10 GB of ram per processor. Even 8 GB per proc is pretty rare today.
If the thread did need that much ram, it would be possible to pool memory between several nodes, it wouldn't be too fast, though (but still WAY faster than swapping to any harddrive). I believe they're currently getting a little over 800 MBytes/sec real-world thruput via the 20gbit full duplex Inifniband interconnects.
Just FWIW, they are claiming power usage of 1.5MW for this cluster of 2200 processors. Cray just released the numbers for their upcoming Red Storm cluster with over 10,000 AMD Opteron processors, just slightly less than 2.0MW.
Ugh, this is getting old.
Red Storm, the machine by itself itself, uses 2.0MW.
Big Mac and all of its networking gear uses less than 0.75MW. The supercomputing center itself (building, air conditioning, UPS battery charging equipment, and the 1100 G5s) is fed by a 1.5MW substation feed. They're still not even maxing out the substation.
The latest, fastest Opterons (not the scaled down low-power Opteron for blade servers) consume 53 watts at full clock. PowerPC 970 @ 2 GHz consumes 48 watts. The U2 and K3 motherboard chipset on the dual G5s uses just as much power as the PowerPC 970 "G5" processors. Hell, the power supply in a dual processor G5 system is 550 watts. 550 x 1100 machines = 0.61MW.
Well, we'll see.. MS's current offerings suck ass. I recently had to install XP after I had been using Linux for years exclusively. I can tell you, I was not thrilled with the experience (pardon the pun). I spent more time trying to get rid of MS's bug laden software such as IE, Outlook, Media Player and replace them with something sane than I would trying to compile and install extra software for Linux.. and the hard part was making sure those bastard programs never popped up again in the course of my day.. because I was not given a choice to de-install them permanently.
Wow, all these years I took granted that I could surf the web without even the thought about having a slimy piece of spyware installing itself on my machine. I truly feel sorry for all of Bill Gates' minions.
It's pretty amazing what people put up with, isn't it.
90% of the computing world would be happy with Firebird, OpenOffice (or even just AbiWord), and a good webmail account. But they instead pump money into Microsoft for Windows/IE and MS Office.
This leads me to a question that maybe someone could answer... is there a good AbiWord-like word processor for un*x that DOESN'T try to be Word clone? If so, I'd be interested in trying it out.
Prepare to lose all karma...
Yeah, we're all bashing Longhorn and Microsoft's ways right now... but come 2006, Microsoft will win once again as millions of businesses and tens of millions of homes upgrade to Longhorn. Even the most hardcore geek will have Longhorn on a partition within a year of its release.
We bashed Win98, but ended up using it anyway. Ditto for Win2K and XP.
Sure, our servers will still run un*x, but as long as we keep using Windows at work for "compatibility" and "familiarity"... and a single innocent Windows box at home "just for games", Microsoft will keep winning.
This is not a flame or a troll, but just a prediction based upon the past. I would like to be proven wrong, though...
Does MS assume everyone is running their displays at 1600x1200? I thought OS X wasted screen space, but this with the taskbar, sidebar, and superdeluxebig++ window frames is even worse.
"It's funny because it's true".
Hey grandpa, they let you have laptops at the nursing home?
Hehe! But seriously, you must not have any experience with crazy numeric meteorology folks. In a lot of similar textbooks you'll find FORTRAN listings for many of their most efficent algorithms. Many of which would be a complete bear to reproduce in C.
I've been using Panther for about 24hours and really haven't run into any bugs other than having to reinstall MS Office X.
If you haven't already, I would highly recommend filing bug reports with Apple (go to the feedback section of their OS X website). Rumor has it Apple is currently collecting bugs to fix for 10.3.1.
Also, did you do a fresh install or an upgrade. My roommate and I did fresh installs on our machines and really haven't run into any of the bugs you've reported. Moving icons and mounting servers have worked fine for us. Though we don't use IMAP or Bluetooth.
Ehh... Carmack isn't developing Doom3 on MacOS. He has been pretty vocal about his love for Visual Studio 6. However, Doom3 does not make much use of DirectX (all of the gfx use OpenGL, for example) so he has made a few builds on Mac OS X and Linux over the past couple years.
In fact, the first demo of Doom3 (and the first demo of the GeForce 3 too) was on Mac OS X as part of one of Steve Jobs's keynote speeches.
Doom3 will be available for Mac OS X... but it's not being developed on it.
The Max OS X Panther 10.3 box includes 4 CDs... three for 10.3 and it's accessories (keep in mind these three CDs include localizations for 12 languages)... and a development environment CD containing compilers, various SDKs, and the feature-filled xCode IDE.
It's a bit alien to those not used to the NeXT way, but it only took my roommate about 15 minutes to find his way around. Both of us have already converted most of our projects to xCode.
Included in the box (what a cool black box it is, too!) is a development environment CD (compilers, APIs, SDKs, and the xcode IDE).
I'm happy to see Apple still giving the development tools away for free.
Overall, I'm quite happy with it, but I've found a few bugs. yes, I've reported at least one to apple
Fire up Safari (or click on the feedback icon) and file the rest fo the bug reports. Apple is currently culling the feedback looking for bugs to fix for 10.3.1, every little bit helps!