Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Quad PowerMacs?
- Thread starter MacRumors
- Start date
- Sort by reaction score
But, will the gains be less dramatic on the G5/G6 architecture? What I mean is, as I understand it the P4 has a very long pipeline. That means more bubbles and with larger penalties for them. The G5/G6, having shorter pipelines, should have better efficiency to begin with, meaning that filling bubbles might not yield incredible gains. Yes/no?
It depends on your perspective. The instructions being asked of the P4 by Cinebench are just inefficient by more than 17-19%. The CPU is having a performance hit because of all the NOOPs in the pipeline. HT gives some of that back wich you already had. You can't get something for nothing, without HT you were being robbed by poor instruction sets.
i guess you have never heard of the Playstation or Xbox. they are sold at a loss.
i know, i know... they make it back (in theory) by licensing deals on the games and accessories, but each individual box is a loss for the company.
I suspect Apple does have a 4 processor machine in their lab and probably has had it for a while. When they might start selling one is another story.
My reasoning is that going beyond 2 processors in a machine requires a lot more work in cleaning up the OS Kernel. Also, any work done with that makes the kernel run that much better on a dual processor system.
As for how much one might cost, I suspect less than double a what a dual processor system costs. I'd say with the dual 2GHz G5 being $3000 US, a quad processor system would come in around $5000 US at most.
Who would buy one? Well anybody who does animation rendering, DVD encoding, etc.
Would I buy one? Probably. At times it wouldn't get bogged down, at other times I'd probably keep it very busy. As it stands, my next Powermac is liable to be a dual 3GHz G5 with at least 4GB of RAM. And yes, I really do need that much RAM. Heck, if I could add more RAM to my current Powermac that would be great.
I guess my one major point here is please remember that if you can't use something, someone else might be able to. Just buy what you need and can aford and let those who need bigger systems buy them
My reasoning is that going beyond 2 processors in a machine requires a lot more work in cleaning up the OS Kernel. Also, any work done with that makes the kernel run that much better on a dual processor system.
As for how much one might cost, I suspect less than double a what a dual processor system costs. I'd say with the dual 2GHz G5 being $3000 US, a quad processor system would come in around $5000 US at most.
Who would buy one? Well anybody who does animation rendering, DVD encoding, etc.
Would I buy one? Probably. At times it wouldn't get bogged down, at other times I'd probably keep it very busy. As it stands, my next Powermac is liable to be a dual 3GHz G5 with at least 4GB of RAM. And yes, I really do need that much RAM. Heck, if I could add more RAM to my current Powermac that would be great.
I guess my one major point here is please remember that if you can't use something, someone else might be able to. Just buy what you need and can aford and let those who need bigger systems buy them
Actually there might not be as much of a loss on them now. Manufacturing costs did go down and such.
However, when the main selling point is the hardware, Apple can't afford to take a loss on the sale.
My understanding is that with the G5 currently there is an incredible under usage of the processor's functional units. Not bubbles per se, but the fact that half the units go unused at times because there are just so many of them. So the G5 still has room to do more work and SMT would help greatly with that. And the Power5 (basis for the G6 more than likely) which will have SMT will also iirc have more execution units per core, so it would benefit even more from SMT.
If you look at the G5 now, it can dispatch up to 5 instructions per cycle, but has 8 (I think more) units that can be doing work, and which mostly do their work in a single cycle (probably some 75%+ of PPC instructions are 1 cycle execution). Having two threads pushing instructions through such a core would make for lots of speed gains.
Re: Not going to happen
As for what's faster, AMD is not sleeping and there will be plenty of makers of quad opteron systems.
I think a Quad G5 system is a good thing for Apple.
Actually in most cases, a quad processor system will be much faster than 2 dual proc xServes in a cluster.
As for what's faster, AMD is not sleeping and there will be plenty of makers of quad opteron systems.
I think a Quad G5 system is a good thing for Apple.
Although it would seem unlikely that Apple would really pursue a quad machine there are some temptations. Apple has not been resoundingly successful in the low-end mass market but does currently have some interesting leverage at the high end with the G5. There are numerous engineering tools such as Pro-E and Ideas which kill even a high end workstation as models get above a certain size. A lot of time is spent staring at an unmoving screen. And the unadulterated Unix OS with those machines is not all that user friendly when it comes to working with MS applications- at least as compared to OSX. With a stable OS, clear MS compatibility and a super-fast machine it is conceivable that Apple could actually make inroads in the high end workstation - if the developers could see some motivation to make their tools function on the Mac. Given the constant costly overhead of dealing with Windows viruses and the like, if there was a clear, confirmed performance advantage to the Mac system the potential exists for corporate EIS organizations to reconsider the Mac as a viable platform. This is probably a pipe dream but you never know. I think about Pixar.....
Right, but what tasks are you talking about? Rendering? Sure, when rendering a scene, C4D will devote one processor to half the image and the other half to the other processor (I used to use C4D all the time) which will cut the rendering time in half. But thats not really going to give you 3.6ghz processors, its still giving you 2ghz processors that divide the work up better. (multitask)
In Maya, 2 processors mean one processor for smoothshading and the other for wireframe, so in theory actual response time in the gui may be quicker. (read this one time at www.alias.com) Still, thats just divide and conquer.
Argue it all you want, but 2.5 + 2.5 + 2.5 +2.5 does not equal 10gHz. It would still be a 2ghz machine that multitasks so well that it might compete as a 6-7 ghz. Although, I would bet that the more processors you add, the less effect they would have. (law of diminishing returns)
Oh,and BTW, I wouldnt call Cinema4D a "high end app"
Re: OT:
How is Apple selling their OS at a loss if probably the number one reason people buy Apple's hardware is to run their OS on it? If everyone went out and pirated the Apple OS, it doesn't matter that much in the sense that they're running it on hardware that they paid Apple for. Now if everyone did that for windows, that's another story. I believe that's one of the main reason is that there is no anti-piracy protection in OS X... since there is no reason to. In the end, Apple made their $$$.
How is Apple selling their OS at a loss if probably the number one reason people buy Apple's hardware is to run their OS on it? If everyone went out and pirated the Apple OS, it doesn't matter that much in the sense that they're running it on hardware that they paid Apple for. Now if everyone did that for windows, that's another story. I believe that's one of the main reason is that there is no anti-piracy protection in OS X... since there is no reason to. In the end, Apple made their $$$.
I don't know about you but if I'm hosting Web Apps I'd like to let one CPU running Apache Tomcat/Cocoon lock down that CPU and spawn 100s even 1000s of threaded instances of Web Apps.
If I'm doing Real-Time Video Editing you have to lock down one CPU to do just or it's not in Real-Time.
There are myriad reasons for this but if you like multithreading via just a single CPU then so be it.
That's not multitasking, that's multithreading. In this case, the threads are split across processors. My point is that if your machine happens to spend 90% of it's time rendering (or calculating protein folding or whatever), then dual processors are just as good as a faster machine. If they do the job nearly twice as fast, you can term the internal workings however you want, but there's no difference to me, it's faster. But yes, a single 4 GHz G5 would be preferable, I agree (because of, as you mention, tasks with are not MT/MP-aware). I wish there were one to buy.
Well you might be surprised if you haven't used it for a while, it's quite capable now and is beginning to sport some very high-end features, some of which even Maya doesn't have. Streamlined workflow, a pretty and fast stock raytracer, and rock-solid stability are still its best features though, IMO. I find Cinema much faster for certain kinds of things and Maya better for others. But anyway, same deal with MP Maya rendering...Originally posted by agreenster
Oh,and BTW, I wouldnt call Cinema4D a "high end app"
OT:
This is way off topic, but here goes:
10 Million copies of OSX have sold as of Feb. Apple has sold some 7 million computers in the last two years. But we don't know their cost, but indirectly we can look at the retail sales numbers to get an idea. So 3 million retail boxes were sold most likely through dealers. Cheapest I saw OSX sold was $69 so divide that by 2 to get the actual box cost to dealers (standard mark up 100% for retail software). Actual profit for apple is probably 50 to 75% of that. So at best apple keeps $30 per box.
Lets be generous and say that apple makes $50 per box and be more generous that the cost is included in the cost of each machine. So that is at most $500 million in sales for the OS.
Apple bought NeXT from Jobs for $400 million just about 8 years ago.
Looking at the last 8 years of OS RD on the 10-K forms submitted by Apple to the SEC, they spent as little as, $10 but more like $20 million each year for OS development. That is an additional $160 million in costs.
Lets not forget the loss costs on developing Rhapsody that didn't sell one copy.
So maybe my statement of selling at a loss is far from damning but it is hardly week. Even with a generosity factor of 66%, giving an over estimate total os sales of $500 million, it doesn't cover the published $560 million for development.
The profitability of the OS is far from certain, hence me saying, I am going out on a limb. The point of the poster was to sell a cheap machine at a loss. If your highest volume machine is sold at a loss as is the software on it, how does Apple stay in business?
This is way off topic, but here goes:
10 Million copies of OSX have sold as of Feb. Apple has sold some 7 million computers in the last two years. But we don't know their cost, but indirectly we can look at the retail sales numbers to get an idea. So 3 million retail boxes were sold most likely through dealers. Cheapest I saw OSX sold was $69 so divide that by 2 to get the actual box cost to dealers (standard mark up 100% for retail software). Actual profit for apple is probably 50 to 75% of that. So at best apple keeps $30 per box.
Lets be generous and say that apple makes $50 per box and be more generous that the cost is included in the cost of each machine. So that is at most $500 million in sales for the OS.
Apple bought NeXT from Jobs for $400 million just about 8 years ago.
Looking at the last 8 years of OS RD on the 10-K forms submitted by Apple to the SEC, they spent as little as, $10 but more like $20 million each year for OS development. That is an additional $160 million in costs.
Lets not forget the loss costs on developing Rhapsody that didn't sell one copy.
So maybe my statement of selling at a loss is far from damning but it is hardly week. Even with a generosity factor of 66%, giving an over estimate total os sales of $500 million, it doesn't cover the published $560 million for development.
The profitability of the OS is far from certain, hence me saying, I am going out on a limb. The point of the poster was to sell a cheap machine at a loss. If your highest volume machine is sold at a loss as is the software on it, how does Apple stay in business?
Okay okay. I stand corrected. Multithreading...I agree that four processors are better than two. But its still a misnomer that the mHz of a quad machine = the speed of the single processor x 4.
And yes, I have heard some things about C4D, and that its made some improvements. (I havent used it since XL7) But its nowhere close to the industry standard: Maya.
C4D was one of the first to use "hyperNurbs," which is still great (called smooth Proxy in Maya) and I remember the renderer was pretty decent. But the interface and character setup was so much slower. Character animation in Maya FLIES.
But anyway...
Re: Re: Please, people, Think Different!
Actually, that's FAR from the most intelligent thing I've ever read on here.
Faster processors do lead to new technologies. Imagine something like iPhoto, or Garageband if we were still using machines that were on par with a 486.
On the other hand, the thought that a quad processor G5 would bring advanced to typical users is rediculous. Home users will Never have quad processor G5s unless they have really small, *ahem*.. thingies.. or maybe if their home Graphics or Research business would greatly benefit from it. [It would be a boon to all those home Computational Chemists. ;-)]
A Quad processor G5, though really swank, would be physically big. It would be very expensive. It would require much more power to run and it would generate much more waste heat. It'd even, very likely, generate more noise. It's not a machine that would find a niche in a typical home.
I agree that we would all benefit from faster machines, though I have to point out that for the typical user the current machines are currently FASTER than what is required by most people. Do you think that having 4GHz P4s or 3GHz G5s today would significanlty change most people's computing experience? Would they read their email faster or even browse the web faster? Would they type their Word docs faster? The fact is, todays hardware is ahead of the software in all but a few very processor intensive applications like Video. Even most photoshop operations are instantanious unless you are working on enormous production quality images, and then it's still blazingly fast. Long gone are the days were you'd apply a filter and walk across the street for a snack.
and.. ahem.
Bill Gates never said that we'd never need more than 640K. That's an internet myth.
Actually, that's FAR from the most intelligent thing I've ever read on here.
Faster processors do lead to new technologies. Imagine something like iPhoto, or Garageband if we were still using machines that were on par with a 486.
On the other hand, the thought that a quad processor G5 would bring advanced to typical users is rediculous. Home users will Never have quad processor G5s unless they have really small, *ahem*.. thingies.. or maybe if their home Graphics or Research business would greatly benefit from it. [It would be a boon to all those home Computational Chemists. ;-)]
A Quad processor G5, though really swank, would be physically big. It would be very expensive. It would require much more power to run and it would generate much more waste heat. It'd even, very likely, generate more noise. It's not a machine that would find a niche in a typical home.
I agree that we would all benefit from faster machines, though I have to point out that for the typical user the current machines are currently FASTER than what is required by most people. Do you think that having 4GHz P4s or 3GHz G5s today would significanlty change most people's computing experience? Would they read their email faster or even browse the web faster? Would they type their Word docs faster? The fact is, todays hardware is ahead of the software in all but a few very processor intensive applications like Video. Even most photoshop operations are instantanious unless you are working on enormous production quality images, and then it's still blazingly fast. Long gone are the days were you'd apply a filter and walk across the street for a snack.
and.. ahem.
Bill Gates never said that we'd never need more than 640K. That's an internet myth.
It really depends on what you want actually. You forget about Newtek's Lightwave 3D, wich has an encridible good renderer even exceding Maya's with a lot of professional features not implemented in many apps..
You have to put a lot of effort in Maya before you get good results. In Lightwave, it goes a little bit easier. (still hard)
All other 3D apps I know off, .. going from very good toys to pretty bad toys..
Not to mention even that C4D is way more expensive than LW 7.5 or 8.. All the seperate modules are build in in LW 8 for 1599 $ while you have to buy the Studio version of C4D to get the same (and it isn't that good as LW either) you pay 2500 $ !!
Re: Re: Re: Please, people, Think Different!
Never have quad processor G5s unless they have really small, *ahem*.. thingies.. or maybe if their home Graphics or Research business would greatly benefit from it. [It would be a boon to all those home Computational Chemists. ;-)]
A Quad processor G5, though really swank, would be physically big. It would be very expensive. It would require much more power to run and it would generate much more waste heat. It'd even, very likely, generate more noise. It's not a machine that would find a niche in a typical home.
[/QUOTE]
Try massive not big, typical QUAD chassis are 27" deep in a 4u/5u width, imagine 4 of those G5 heatsinks, also 2 PSU's (500W/600W each) as a 600W is not enough for a QUAD system and also alot more fans hence more noise. Iam planning on the QUAD Opteron end of the year but it will definetly be completly liquid cooled (CPUS, VIDEO, CHIPSET, PSU's) and will require alot of work to shut it up!
Never have quad processor G5s unless they have really small, *ahem*.. thingies.. or maybe if their home Graphics or Research business would greatly benefit from it. [It would be a boon to all those home Computational Chemists. ;-)]
A Quad processor G5, though really swank, would be physically big. It would be very expensive. It would require much more power to run and it would generate much more waste heat. It'd even, very likely, generate more noise. It's not a machine that would find a niche in a typical home.
[/QUOTE]
Try massive not big, typical QUAD chassis are 27" deep in a 4u/5u width, imagine 4 of those G5 heatsinks, also 2 PSU's (500W/600W each) as a 600W is not enough for a QUAD system and also alot more fans hence more noise. Iam planning on the QUAD Opteron end of the year but it will definetly be completly liquid cooled (CPUS, VIDEO, CHIPSET, PSU's) and will require alot of work to shut it up!
Re: Re: Re: Re: Please, people, Think Different!
It might take a little longer than a year, but IBM is not sleeping the way Motorola was.
Actually as the die size on the G5s (and future processors used by Apple) shrink, the amount of electricity they use and heat they generate gets reduced. Therefore the overall space requirements and powersupply needs are reduced. Also reduced is the amount of cooling needed per processor. So, I suspect a Quad processor G5 within a year might have the same space and power/cooling requirmements as the current Dual 2.0GHz G5 Powermac. I do know in terms of noise, the fan setup in the Powermac G5 is relatively quiet for what it is cooling.
It might take a little longer than a year, but IBM is not sleeping the way Motorola was.
Re: Hmmmm
Yep. The other point worth mentioning is that the PPC970 is currently used in a vanilla SMP configuration, which generally doesn't scale well beyond two processors anyway (memory contention). The Opteron uses HyperTransport as a ccNUMA fabric (thanks in part to the embedded memory controller), which makes it scale almost linearly to a lot more processors.
The two downsides are that it takes a ccNUMA aware operating system to really use it well (e.g. Linux 2.6) and accessing memory on other nodes is slower than vanilla SMP, though in practice a remote memory access on Opteron is about as fast as a local memory access on the PPC970.
Before they do anything else, I too would like to see IBM embed a memory controller and start working on using HyperTransport as a ccNUMA fabric.
Yep. The other point worth mentioning is that the PPC970 is currently used in a vanilla SMP configuration, which generally doesn't scale well beyond two processors anyway (memory contention). The Opteron uses HyperTransport as a ccNUMA fabric (thanks in part to the embedded memory controller), which makes it scale almost linearly to a lot more processors.
The two downsides are that it takes a ccNUMA aware operating system to really use it well (e.g. Linux 2.6) and accessing memory on other nodes is slower than vanilla SMP, though in practice a remote memory access on Opteron is about as fast as a local memory access on the PPC970.
Before they do anything else, I too would like to see IBM embed a memory controller and start working on using HyperTransport as a ccNUMA fabric.
A year, these people are thinking that its being released now? As for the die size yes it decreases but you need more powerfull cooling solutions as the contact area is smaller hence the heat is harder to remove so you need bigger heatsinks to absorb the heat faster so no, space requirements do not get smaller. As for the power requirement they will be high anyway due to the 90nm process (current leakage etc) and this is worse at 65/40nm eventually you will be dissipating more static power than dynamic. Power requirements will continue to increase as we move on, remember when the Althon was released a few years back we had 200W PSU's now I cant do without a 600W for a DUAL Opteron and with the QUAD requiring 1000W.
Re: Re: Hmmmm
Memory performance scales with number of CPUS in an Opteron system 6.4GB/s per CPU when NUMA is enabled. Unfortunetly this makes the CPUS with 3 HT links very expensive aka 800 series. If Apple wanted to make a realistic QUAD CPU machine this method would not be used only the standard SMP hence OSX would take advantage out of the box.
Memory performance scales with number of CPUS in an Opteron system 6.4GB/s per CPU when NUMA is enabled. Unfortunetly this makes the CPUS with 3 HT links very expensive aka 800 series. If Apple wanted to make a realistic QUAD CPU machine this method would not be used only the standard SMP hence OSX would take advantage out of the box.
Quad Powermac Lineup
How's this for a Quad Processor Powermac lineup?
Single 2.5 GHz: $1999
Dual 2.5 GHz: $2999
Quad 2.5 GHz: $3999
Quad Powermacs? Never gonna happen...
How's this for a Quad Processor Powermac lineup?
Single 2.5 GHz: $1999
Dual 2.5 GHz: $2999
Quad 2.5 GHz: $3999
Quad Powermacs? Never gonna happen...
Re: Re: Re: Hmmmm
It is a relative thing. Before the Opteron, you couldn't buy a scalable 4-way or 8-way server without spending MAJOR coin on Unix iron. The 4-way and 8-way Opteron boxes may be expensive, but they are still priced like high-end commodity PC servers, which is to say pretty damn reasonably.
The important point is that these boxes are as fast as anything with the same number of processors, whether it is proprietary Big Iron or not. In essence, you can now pay PC server prices and get as much or more scalability and performance as the classic proprietary Unix enterprise hardware that would normally cost you 10x as much as the Opteron equivalent. Which is why just about every Unix Big Iron vendor is making plans for an Opteron-based mid-range product line.
The Opteron represents a significant move up the performance hierarchy for commodity computing in the world of enterprise servers, and is now soundly in the mid-range enterprise arena.
It is a relative thing. Before the Opteron, you couldn't buy a scalable 4-way or 8-way server without spending MAJOR coin on Unix iron. The 4-way and 8-way Opteron boxes may be expensive, but they are still priced like high-end commodity PC servers, which is to say pretty damn reasonably.
The important point is that these boxes are as fast as anything with the same number of processors, whether it is proprietary Big Iron or not. In essence, you can now pay PC server prices and get as much or more scalability and performance as the classic proprietary Unix enterprise hardware that would normally cost you 10x as much as the Opteron equivalent. Which is why just about every Unix Big Iron vendor is making plans for an Opteron-based mid-range product line.
The Opteron represents a significant move up the performance hierarchy for commodity computing in the world of enterprise servers, and is now soundly in the mid-range enterprise arena.