[QuadG5Man]

With those big G5 cases could we ever see Quad G5 proccessors in one box?

Just wondering, would love to buy one ;-)

brian

[Lateralus]

I'm sure it is entirely possible from an architectural stand point. As of now, there is one processor per daughtercard and each daughtercard has an independant bus ranging from 800MHz to 1GHz. In the Dual G4s, there was two processors on the one daughtercard in the system and both processors shared that 100/133/167MHz bus speed. So, if two processors could be fit on one daughtercard with the G4, and forced to share the comparatively shitty 1xxMHz bus speed, then I am sure that it is entirely possible to have two processors per daughtercard with the G5. And having two processors share the bus speed would not be all that bad whatsoever since there is plenty of bus bandwidth, unlike the G4s.

However, I doubt we will see Apple ever introduce a quad machine. For starters, I doubt they want to thread the OS for 4 processors. Also, the cost of the 4 processors would be quite high, forcing Apple to give the machine a price tag that would make it out of reach of the average consumer.

And you have to remember that the only reason Apple became as Pro-Dual as they are is because the G4 got stuck at 500MHz while Intel and AMD ramped their chips to 1GHz and beyond. So, the only thing Apple could do to help the situation was put two 500MHz chips together. And since the G4 never scaled well, and always kept the Mac behind the PC side in terms of clock speed, the dual philosophy stuck. And now with the G5, the clock speed gap has been closed considerably, but it still exists. Though because of the raw bandwidth of the G5, it competes well against every Intel offering, which the G4 did not. I suspect that the only reason Apple went Dual with the G5 was because of the fact that so many software manufacturers, including Apple themself, had made such dual friendly software.

[gururafiki]

Originally posted by PowerMacMan:
I'm sure it is entirely possible from an architectural stand point. As of now, there is one processor per daughtercard and each daughtercard has an independant bus ranging from 800MHz to 1GHz. In the Dual G4s, there was two processors on the one daughtercard in the system and both processors shared that 100/133/167MHz bus speed. So, if two processors could be fit on one daughtercard with the G4, and forced to share the comparatively shitty 1xxMHz bus speed, then I am sure that it is entirely possible to have two processors per daughtercard with the G5. And having two processors share the bus speed would not be all that bad whatsoever since there is plenty of bus bandwidth, unlike the G4s.

However, I doubt we will see Apple ever introduce a quad machine. For starters, I doubt they want to thread the OS for 4 processors. Also, the cost of the 4 processors would be quite high, forcing Apple to give the machine a price tag that would make it out of reach of the average consumer.

And you have to remember that the only reason Apple became as Pro-Dual as they are is because the G4 got stuck at 500MHz while Intel and AMD ramped their chips to 1GHz and beyond. So, the only thing Apple could do to help the situation was put two 500MHz chips together. And since the G4 never scaled well, and always kept the Mac behind the PC side in terms of clock speed, the dual philosophy stuck. And now with the G5, the clock speed gap has been closed considerably, but it still exists. Though because of the raw bandwidth of the G5, it competes well against every Intel offering, which the G4 did not. I suspect that the only reason Apple went Dual with the G5 was because of the fact that so many software manufacturers, including Apple themself, had made such dual friendly software.

Now I feel smarter. Thank you.

[reader50]

We have not heard any rumors about the X-Station lately, but it might be a Quad if it's real. Back during the clone days, DayStar made a quad 604e box. At the time, it was the ultimate Mac.

There has been talk that the 976 will be available in dual-core, a pair of those would fit in the G5 case and would work just like a normal Quad.

Being faster than x86 isn't going to grow market share too much, being *way faster* might. For example, if Apple introduced consumer software that could not be ported to x86 because no existing box was powerful enough.

Steve once mentioned such an app, a photo editor that could clear up out-of-focus pictures. You need to ray-trace each pixel back through a virtual lens, then attempt to refocus the virtual lens to produce a focused picture. Lots of ray tracing, and successive adjusting of the virtual lens to find the sharpest result. Right now, nothing on the desktop is powerful enough, but a quad 980 might do the trick.

If everyone wanted a must-have app and Apple had the only boxes that could run it, more Macs would sell. It was like that with the original AppleWorks - people bought Apple]['s in order to run the software. Same thing for page layout with PageMaker, people bought Macs because it could not be done on the PC at that time.

I think Apple will release whatever it can to grow market share, the 130nm 970 somewhat limited them to duals unless they wanted a vacuum cleaner on the desk. The 970FX with it's lower power consumption is a different ball game. We'll see.

[jcadam]

I wouldn't be surprised at all if IBM released quad-processor (or more) PPC970 machines (have they already? I haven't checked), but Apple.... nah.

You could just get a cluster of XServe G5s in lieu of an XStation .................

[Love Calm Quiet]

Marketing wise, with small demand for the Quads, wouldn't a better plan be for apple to continue encouraging the Distributed Processing idea?

Then apple could just encourage purchase of additional units (no extra apple design / inventory costs) - maybe stripped down (small HD, no superdrive) for power users...

Suppose for Photoshop, etc. you could just add another DPG5 to the office and have drain its CPU cycles when no one was using it!

[Eug]

I think Apple will release whatever it can to grow market share, the 130nm 970 somewhat limited them to duals unless they wanted a vacuum cleaner on the desk. The 970FX with it's lower power consumption is a different ball game.

Well, not really. It's likely that a 2.5 GHz G5 will use as much power as a 1.8 GHz G5 130 nm. And who knows how much power a 3 GHz G5 will take. I suspect the G5 Power Mac's cooling was designed specifically for the dual G5 3.0 GHz.

[bstone]

I recall seeing cards that had like 4 additional G4 procs on it once upon a time. Whatever happened to those?

[Ratm]

970s

[Scotttheking]

Originally posted by Ratm:
970s
<snip>

Erm, no.

[Eriamjh]

Would a quad G5 require RAM to be installed in Quads (four-at-a-time)?

[reader50]

No, it would work with the current dual-channel setup. But it would improve performance if they went with quad memory channels.

[djohnson]

Mmmmm.... 4 instances of seti@home or dnetc running at once...

[Superchicken]

Oh Quad Processor G5... to be a god among geeks...

There may be a market for it who knows.

[andrewhicks]

Well, Microsoft are putting 3 Power chips into the new Xbox. Maybe anything is possible.

[msuper69]

Originally posted by PowerMacMan:
...
However, I doubt we will see Apple ever introduce a quad machine. For starters, I doubt they want to thread the OS for 4 processors.
...

All of the documentation I have ever read about the OS X kernel indicates that it is multi (as in >2) cpu capable, including Apple's own developer connection site. They always say multiple processor capable w/o stating any current 2 cpu limitation.

Can anybody who is more knowledgeable verify this?

[redesigner]

the XRAID box, with processor cards in them instead of hard drives.. no need for a new case design eh?

each slot could possibly handle

1- dual-core G5 blade
2- hard drive
3- quad-GigE network card
4- admin blade

and the mind boggles at how sweet this would be, but I should prolly be sending this to MacOSRumors..

[redesigner]

when Avie was a PhD candidate designing Mach for his dissertation.. I was just a lowly freshman studying CS..

Mach was always designed as a multi-processor, multi-threaded OS. Why would anyone dumb such an elegant design down. Certainly not its own architect

[g3_brandon]

Originally posted by msuper69:
All of the documentation I have ever read about the OS X kernel indicates that it is multi (as in >2) cpu capable, including Apple's own developer connection site. They always say multiple processor capable w/o stating any current 2 cpu limitation.

Can anybody who is more knowledgeable verify this?

Yes, if I remember correctly, Apple advertised the OS when OS X first was released as having support for up to 64? processors in a single box. I could be wrong on that number, but I know its somewhere up there. I wonder what Virginia Tech had to do to OS X to make it work with all those processors.

[Vader�s Pinch of Death]

The current G5s seem over cooled. The fans are always at low unless you go nuts in photoshop or something, even then they only go up slightly.

Apple definitely made the first G5's with much faster and hotter chips in the future in mind.

[olePigeon]

Originally posted by Vader�s Pinch of Death:
The current G5s seem over cooled. The fans are always at low unless you go nuts in photoshop or something, even then they only go up slightly.

Apple definitely made the first G5's with much faster and hotter chips in the future in mind.

That was also with the 970. Now they should be shipping with the 970FX. The cooling seems overkill.

[Durandalus]

Originally posted by PowerMacMan:

However, I doubt we will see Apple ever introduce a quad machine. For starters, I doubt they want to thread the OS for 4 processors. Also, the cost of the 4 processors would be quite high, forcing Apple to give the machine a price tag that would make it out of reach of the average consumer.

Agree with you on the rest but that machine wouldn't be targeted at average consumers. Just look at Dual Opteron machines that are often much more expensive than a Dual G5. Or HPs dual Itanium Workstations. A machine with two measly 1.3GHz CPUs comes in at $ 9500. I think there's plenty of room for Apple. The question is how much engineering work they'd have to put into it and if they could sell enough machines. After all a dual G5 is fast enough for most people. And for video processing or 3D work you simply use some XServes. Most apps that need that kind of processing power can be run on a Cluster these days. Thinking of this: What's the probability of a Quad G5 XServe? There's a market for quad and more CPU servers.

[jamesa]

FWIW, I'd buy one.

-- james

[mitchell_pgh]

I would think that they would go after the Xserve with quad G5 first.

[nobitacu]

That would be great if they can come up with this and make it work. I'm thining about buying a G5 around the end of Oct. So, I'll be picking up what ever the top of the line G5 is than.

Ming

[djohnson]

I have to say that i would be willing to wait an extra 6 months to a year to afford a machine like that!

On the topic of the G5 case being overcooled as it is, I am thinking that Apple will want to continue to use this case for a few years. The next GX series is rumored to be dual cored and that would produce more heat. How about 2 dual cored chips? Shoot, how about 4 dual cored chips!!!!

[Nsee]

PCI based quad processor cards for G3/G4 did exist and still exist. A company called total impact makes them.
The total mPower is a G3/G4 multiprocessor card
Total impact's website is here.

[djohnson]

Hmm, mighty expensive though! $3200 for 4 400Mhz G3's... I think i would rather buy a dual G5 xserve or tower for that cost...

[SouthPaW1227]

Originally posted by jamesa:
FWIW, I'd buy one.

-- james

Me too. I'd sell basically anything it took to get one.

[alien]

Originally posted by g3_brandon:
Yes, if I remember correctly, Apple advertised the OS when OS X first was released as having support for up to 64? processors in a single box. I could be wrong on that number, but I know its somewhere up there. I wonder what Virginia Tech had to do to OS X to make it work with all those processors.

They didn't have to do much, except install MPI (Message Passing Library) on the G5s. MPI is widely used in sci-tech apps. Basically what happens is that a problem is re-programmed, and split out among many (often) identical programs. Since the apps VT use probably are programmed for MPI in the first place, this wouldn't be a big issue.

Each program will work only on a part of the problem, and send and receive data (using MPI) to/from the other programs as necessary. The important thing here is that the OS on each G5 doesn't change, it still only sees two processors. But the programs can communicate with each other across the network.

[BkueKanoodle]

Originally posted by Durandalus:
Agree with you on the rest but that machine wouldn't be targeted at average consumers. Just look at Dual Opteron machines that are often much more expensive than a Dual G5. Or HPs dual Itanium Workstations. A machine with two measly 1.3GHz CPUs comes in at $ 9500. I think there's plenty of room for Apple. The question is how much engineering work they'd have to put into it and if they could sell enough machines. After all a dual G5 is fast enough for most people. And for video processing or 3D work you simply use some XServes. Most apps that need that kind of processing power can be run on a Cluster these days. Thinking of this: What's the probability of a Quad G5 XServe? There's a market for quad and more CPU servers.

While you can buy/build a opteron DP for way less then the 9500 you stated. Having said that though there's a certain sweet spot in performance in Multi processor machines. You'll get more bang for your buck and better performance associated with 2 DP G5;s then you will with a single Quad Machine, because of the SMP overhead from running 4 procs.

[djohnson]

But where would the geek factor be in merely having 2 dual G5's?

[Catfish_Man]

Originally posted by Vader�s Pinch of Death:
The current G5s seem over cooled. The fans are always at low unless you go nuts in photoshop or something, even then they only go up slightly.

Apple definitely made the first G5's with much faster and hotter chips in the future in mind.

low == quiet. The G5 is designed to have its fans on low whenever possible.

[CIA]

Well, the current G5's are damn quiet. But if hotter/faster chips get put in there and those fans start ramping up all the time, you know people are gonna start bitchin' about how loud the machines are.

OSX can support many, many processors. For most pro-user tasks, many (ie. 4 and up) small slow processors work better then 1 fast one.
For general use, 1 fast one is better tho.

[andreas_g4]

Originally posted by reader50:
Steve once mentioned such an app, a photo editor that could clear up out-of-focus pictures. You need to ray-trace each pixel back through a virtual lens, then attempt to refocus the virtual lens to produce a focused picture. Lots of ray tracing, and successive adjusting of the virtual lens to find the sharpest result. Right now, nothing on the desktop is powerful enough, but a quad 980 might do the trick.

Do you have any additional info on this? Sounds pretty interesting.

[macgyvr64]

Is refocusing an blurry image even possible? How come you never hear of grid computers doing such a thing? They would certainly have the power. Plus...If I had an app that could do that, I'd wait a few days for it to work.

[reader50]

Originally posted by andreas_g4:
Do you have any additional info on this? Sounds pretty interesting.

It came up in an interview Steve gave ~1+ year ago, to an outside magazine (BusinessWeek?). The interview was linked from most Mac news sites, but I don't have the stories bookmarked. With enough digging, I could find it.

Steve did not go into technical details of how it would work, he was discussing iApps and mentioned that an app to refocus bad pictures was possible, but no current desktop system was anywhere close to powerful enough.

I first ran into the idea of refocusing through a virtual lens in the book "The Hunt for Red October". Clancy didn't go into too much detail there either, but enough - and the process makes sense.

Unfortunately, it involves all kinds of floating point calculations starting from upwards of 3 million pixels. Try a Terragens scene rendering sometime on a 2048 x 1536 scale for a taste of how long it takes to render a 3 megapixel photo. Then remember than most film photos are of far higher res than 3mp (20+?). Add in that you are not randomly generating patterns, but raytracing - where you have to first guess in what way the lens was misfocused.

First, raytrace back through an incorrect virtual lens, then raytrace forwards through a theoretically perfect lens. If the focus hasn't improved enough (or has gotten worse), repeat with a different starting lens. Repeat as needed to get the photo in the best possible focus. While you can make some guesses along the way, I would not be surprised if you had to loop through the refocus operation over 20 times.

Further, this will at best focus out-of-focus images. If the image is blurred because the camera moved during the shot, I can' t think of any easy way to clear it up. You'd have to treat the photo as a composite of many sharper photos, and try to subtract out the sharper frames. Since the value you want to subtract isn't known in advance (it's the frame you want to end up with), the problem would have to be solved with another long recursive loop. If something moved during the exposure, then you have the added problem that the intermediate frames will not be identical - the object that moved probably can't be fully focused in such a case.

A bad photo may suffer from both out-of-focus and camera movement. The app would have to sort that out. I don't think any current desktop system could do it overnight, or even in a few days. A grid computer costing a few $million could, but where is the market for people to refocus their photos for a $1,000 each with free overnight delivery? The NSA and similiar organizations are about the only likely customers.

[Richard Edgar]

What is described might be possible for a slightly out of focus image, but I doubt more than that. For a start, since a photograph is fundamentally 2D, it has lost information about the scene. So, creating a 'theoretical' scene to re-image is going to require guesswork. For a 'typical' photo (e.g. of people), this might be realistic, but the general case is not. Then there's the depth of field - some parts of a scene will be out of focus, even with a perfect lens. Finally, there is image noise. This is always present, and is typically very bad news for doing such calculations. Remember, NASA had to install correction optics in the Hubble - deconvolving the images wasn't possible (that was a while ago, and those images are very noisy, but still.....).

I wouldn't expect this sort of thing to work wonders. It would probably just be better than a simple Unsharp Mask.

[m@]

seam to remember removing motion blur is easy. take fourier transform of photo and take FT of a line, add or multiply together then inverse fourier transform. Based on none of the informaion actually beeing lost just distributrd in a uniform way, leading to the motion lines in the photo. However no sure how uniform the circles of confusion leading to an out of focus photo are.

[msuper69]

I'd like to know how one could retreive or generate information that was never there to begin with. That would certainly be a new paradigm: Creating something out of nothing.

Real life isn't like on TV, where the CIA or NSA is able to miraculously take a blurred photo and clean it up so that it's possible to see every hair on someone's head.

Not gonna happen. Not in this iteration of the Big Bang anyway. Perhaps in the next go-round when the laws of nature are changed.

[Richard Edgar]

seam to remember removing motion blur is easy. take fourier transform of photo and take FT of a line, add or multiply together then inverse fourier transform. Based on none of the informaion actually beeing lost just distributrd in a uniform way, leading to the motion lines in the photo

I bet that's on the assumption of zero noise. The real world isn't like that. For example, there's a large number of high order algorithms for integrating tabulated data. However, if you know your data is noisy, then the best thing to do is to just add everything together. The reason? The high order algorithms integrate the noise as well, and get dominated by it very rapidly.