Yeah I'm sorry but I just don't see where an application would suffer at all with a dual core implementation.
The only issue with the dual cores is the FSB being shared. They've doubled the L2 cache and the increased speed in "chip to chip" communication should more than make up for any latencies caused by the sharing of the FSB.
The more you look at it the more dual cores make perfect sense architecturally and financially. I think the impact to developers will be quite transparent.
Right, multicore is the future : at least for motorola, Intel, AMD and IBM ...
Wait a second. Dual-core chips are better than dual-processor chips. It would be just like having a more efficient dual-processor system. Why would you switch if they replaced something worse with something better?
read the rest of the thread and you'll get it.
I'm saying just dual core isn't enough. They need to be dual chips there as well.
I'm saying just dual core isn't enough. They need to be dual chips there as well.
There will be. The though of an affordable 4 socket Quad system was a mere fantasy. I don't think there is a mother less than $1000 that has 4 CPU sockets.
Apple's delimma on whether to offer 2 socke DC Powermacs will hinge on IBM yields. The question at hand is:
1. Does a Quad system provide enough sales to match the profit that can be obtained by selling 2 whole DC Powermac systems?
What Apple may want to do is keep the fastest CPU as a DC Powermac system and then take the processor that yields the highest and create a Quad system. Then say that the 2.5Ghz DC chips yield exceptionally well.
So let's say they can hit 3Ghz DC Q2 2005.
They should ship a $2799 DC 3Ghz system
and then follow that up with a $3499 Quad 2.5Ghz.
For some users the DC 3Ghz would be more beneficial to their needs while others would enjoy a Quad system better in their workflow.
I'm saying just dual core isn't enough. They need to be dual chips there as well.
OK. I'm sure they will make "quads" (two dual-core machines) eventually. But I also have to wonder how efficient they will be, given the state of today's software. How many threads can an app generate and at what point does adding processors/cores not make sense?
I agree, it appears that now that MHz is such a problem at .90 micrometers as well as heat. I would think that multi-threading would add some increase in performance, and some marketing mind share. Dual core two chip systems with multithreading. The multithreading may not get us much but I don't see the other low lying fruit that there is except for Altivec 2.0. And I almost forgot on chip memory controller, which I would think that a CPU such as the PowerPC 9xx with 200+ instructions in flight could really benefit from on chip memory controller and multithreading, and a great compiler.
Quote:
Originally posted by NovaVengeance
No, multi-threading still might see the light of day on a PPC chip.
Could you give an example of something that would perform *worse* on a dual core chip than on two separate chips? Is it just the possibility that they wouldn't up the FSB bandwidth correspondingly that makes you say this?
Applications at this point of course not, we don't have the chips to run the software on! The number one problem as I see it is the high potential for FSB saturation on the current implementation.
The other issue, now that we are talking about future chips, is the impact on hard single thread applications. This comes from my current belief that you will not be able to realize the same top speed in a dual core unit simply because of the heat issue. That is the near 2X increase in power disapation, though this may not be as bad as it sounds as IBM has a long way to go with power management. Of course if a single core chip is never produced with the technology IBM applies against the dual core unit, there will never be a possibility for comparison. The other reality is that there is very little in this world that can't be multithreaded to some extent, so this argument only applies to specific cases.
One must remember that the technology to build this dual core unit could be applied to building a very robust single core unit. Given the same amount of transitors very high performance could be achieved. This would be via SMT, AltVec2, a huge cache and other improvements. Of course they might not even call such a chip a 970, that sort of jump is probally reserved for the 980.
I would not be surprised to find that the first dual core chips end up delivered at a lower clock rate than is available in single core processors at the time. AMD has pretty much indicated that their first dual core chips will be a bit slower than the single core chips out at the time. Slower in the sense of clock rate, obviously system performance will be better, though some applications may be hampered.
All that being said if the new iMac is delivered with a dual core chip I'd be very surprised and extremely happy. This is exactly what Apple needs to deliver in its iMac3 to spark the market a bit. Dual core is a great solution for the low end of the market.
I think Peter Glaskowsky hits the same vein we're discussing here
According to Glaskowsky, customers running imaging and scientific applications developed for the Mac platform will appreciate the multicore design. "A dual-core chip is more effective than a single-core chip on problems that stress the computational resources of the chip, more than the front-side bus bandwidth. Because the 970FX has a very fast, efficient front-side bus, most Mac applications will favor the dual-core configuration."
Also of note in the article are the comments on the "Code Cracking" functions that appear new. I can see how this function fits better with the superior low latency benefits of dual core systems.
Also there seems to be a bit more confirmatin that the pipelines have indeed been lengthened somewhat which should help clocking.
I think this is good timing for Apple. I don't see FSB saturation happening until memory moves beyond DDR2 667. That gives Apple 12-18 months to milk the current setup and prepare for a 65nm 980 system which I believe is going to have Ondie Mem Controllers, SMT and dual cores as well.
The proof will be in the pudding if Apple starts to bang the "multithread your apps!" drum again. WWDC 2005 might have more sessions on threading and supporting SMP well in preparation for what seems to be "SMP nirvana" in 2006.
I think Peter Glaskowsky hits the same vein we're discussing here
According to Glaskowsky, customers running imaging and scientific applications developed for the Mac platform will appreciate the multicore design. "A dual-core chip is more effective than a single-core chip on problems that stress the computational resources of the chip, more than the front-side bus bandwidth. Because the 970FX has a very fast, efficient front-side bus, most Mac applications will favor the dual-core configuration."
Also of note in the article are the comments on the "Code Cracking" functions that appear new. I can see how this function fits better with the superior low latency benefits of dual core systems.
Also there seems to be a bit more confirmatin that the pipelines have indeed been lengthened somewhat which should help clocking.
I think this is good timing for Apple. I don't see FSB saturation happening until memory moves beyond DDR2 667. That gives Apple 12-18 months to milk the current setup and prepare for a 65nm 980 system which I believe is going to have Ondie Mem Controllers, SMT and dual cores as well.
The proof will be in the pudding if Apple starts to bang the "multithread your apps!" drum again. WWDC 2005 might have more sessions on threading and supporting SMP well in preparation for what seems to be "SMP nirvana" in 2006.
When you say milk the current setup I hope your only referring to 90nm process, and not 2.5GHz for 12 - 18 months.
I think we have extended our wishful thinking in this forum well past the line of sight. I think we'll see things like PCIe, and maybe a show like CBIT (or what ever that microprocessor showcase is) before we should go jumping off the deep end in here. It's all going to come crashing down really hard when over half of it may not even make it into the PowerMac by 2006. It's happened before. (Every Show)
I for one am bowing out of these drastic changes in the processor discussions until after SIGGRAPH. It's just way too early to start building something like this up. It's got catastrophe written all over it.
No PCI-X gets you a board for about $200. Adding PCI-X pushes you above $400 generally. Note this is for volume X86 chipsets. I'm sure Apples chipset are more spendy because they have to design them rather than just license them.
Your typical PC motherboard is pretty full featured at $149 roughly. Add another socket and that goes to $199-500 depending on whether you have SCSI or PCI-X etc.
Knowing that Apple likes to have their margins in the 20s I'm "assuming" the the Apple markup for adding another CPU socket would corrolate closer to $200 than the $50-100 premium seen on X86. I could be way off though and I realize that. I would be amazed however if Apple didn't save at least $100 bucks per unit stripping off the extra socket and using a smaller system controller.
Talks about multicore designs have been around since the early G4 days. Motorola was supposed to have worked to make that happen almost 3-4 years ago...but it never happened.
I'll believe it when I see it...no wait...when it's on my desk, running Doom 3.
Well, somebody already posted that link above, eh?
Hum.
A nice summary of all the MP 'rumours' that have been flying around. 'IBM refused to comment.'
Thankyou IBM, 'nuff said.
The Gadget Show (see macworld.co.uk) did a VERRRRRRRRRY superficial bench between Macs and PCs and guess which won on Games and Photoshop filter tests?
That's why Apple needs a chip that blows away the wintel cpus in non-apple lab' conditions.
The race for dual core is on. Hopefully Apple can offer a dual core chip running at 2.5-3.2 gig by next March. That will put Apple way ahead of AMD in 'gighz' and 'only' 800 mhz behind the 4 gig prescott if it ships.
Personally, I'm hoping IBM can reach 3.4 or even 3.6 gig with this even deeper pipelining. This instruction 'cracking' sounds like an interesting counter (ala 'micro opps' for x86?) to the deeper pipelining inefficiencies... What is interesting for me is the way different parts of the chip...integer, fpu, altivec will have different length pipelines...
This sounds like an awesome chip. Or at least the chip that will offer undisputed parity until the 'awesome' Power 5 derivative turns up. (A chip that truly promises to lay waist to the x86 opposition.)
Look at the positives of the MP.
1. More efficicent than dual processors. (with thread with instruction cracking...) So you get your dual processor effect more of the time. So a dual 2.5 should in MP speak by more like a 4 gig plus Pentium more of the time.
2. Higher clock speed. Putting AMD in the shade and running Intel down.
3. The MP should really help Apple's consumer desktops.
4. The MP can 'dual' the 'dual core'. That emphasises Apple's commitment to the workstation market they're clearly going after.
5. Dual Altivec. We know how powerful one altivec is. Dual that? Jaw drop!!!
6. Scalability of frontside bus, sleep modes etc. Speed throttling. This should allow a very interesting 'low clocked' dual core 1.6-2 gig G5MP into a Powerbook? Not gonna happen? We'll see. With 0.09 manufacturing behind them then the 'G4 in a Powerbook' won't happen camp may end up pleasantly surprised.
7. Big speed ups, FINALLY, for Apple's entire line from iBooks to Towers. When Apple gets to 3 gig plus it allows G5 adoption for everything...at various speed grades.
8. Dual fpu. Lightwave renders are going to be frighteningly fast. Not to mention Final Cut Pro', more real time rendering...Apple's workstation line optimised for MP is going to get even more pro people come back to Mac. More Maya adoption to the Mac workstation. At 25% and counting. Optimised for Maya Unlimited.
9. 'Snap' is back in the Mac. 'X' finally 'snaps'. yeah baby.
10. Casual benching eg 'Gadget Show' will show that Mac rocks on games. Macs will no longer be cpu bound on graphics cards like the Ati 9800 xt or the Nvid' 6800 Geforce. Important with next gen' games like Doom 3 and Half Life 2 already stressing 3.4 gig Pentiums to the limit at 1200 x 1024 resolutions even with latest 6800.
(11. Safari blows Interent ExSlower out the water in speed bench by Jobs and Phil. Hey, I want that MP vs Pentium 4 gig speed bake off...we all do, right? )
Lemon Bon Bon (Optimistially...though constipatedly waiting for next 'March', eh?)
PS. 12. Mix MP with the 'Tiger' effect. eg Core Image, Core Video and Photoshop benching absolutely destroys the opposition!!! (Adobe, stand up and be counted. We'll just where their loyalties 'lie' after 'Tiger'.) Final Cut with Core Image and Video. iPhoto with Core Image. iMovie with Core Video. 'Tiger' optimised with new Compiler tech'? The Mac feels faster than ever. 'FEELS' fast and fluid as opposed to pauseworth, stately and elegant. Put in new Quicktime codec.
'Tiger' with 970 MP. The Mac, 'X' and 'PPC' promise finally lands!?!?!?!?
[edit: Whoops! Didn't notice the extra pages on this thread. Sorry. ]
Quote:
Originally posted by Whisper
<waves hand around> OOOh OOOOOh I have a question!!</waves hand around>
Does this mean that instead of have 2 2.5GHz chips running on two separate 1.25GHz busses, there'll be two 3GHz cores running on a single shared 1GHz bus? Cause I think that would be a downgrade. If the new bus is 2.5 times slower -- lets say 2 times slower because there's only one bus worth of overhead now -- that means that the CPUs can only process half as much data. IIRC a significant chunk of the G5's speed comes from its super sweet FSB, which this new version would lack.
In practice, this is not that big a deal, because the system bus isn't a bottleneck in the overwhelming majority of cases: RAM is. There will be a few boundary cases where this implementation will hurt, but we are not looking at Son of MaxBus here. Hell, look at MaxBus: Even with two CPUs hanging off the same hoary old SDR bus, we got 160%-180% the speed of a single CPU for any task that didn't require streaming huge amounts of memory (which is to say, most tasks).
This is speculation, and it probably stems from the fact that the current 970-family CPUs only support whole-number bus multipliers. With a 3GHz core, you have a choice between a 1.5GHz bus and a 1GHz bus (I'm ignoring higher multipliers, since there is zero chance that Apple will use them). We know Elastic Bus can do 1GHz, but 1.5GHz would be a stretch. Not impossible, but of unknown difficulty. Not only is that an unprecedented frequency for a bus, but it requires that the companion chip (northbridge) run at 1.5GHz, and that's unprecedented as well. Remember, the heat pipe on the daughtercards in the original PowerMac G5s wasn't for the 970: It was for the northbridge.
As for what Apple will do with these: They hardly have to do anything to support dual dual-core 970s, because the interface to the northbridge is almost identical, and they've already designed a northbridge that supports two busses. That's a non-trivial design effort for a chip that's essentially brand new, and I can't really see them throwing it out. What they can do, and what they probably will do, is go back to something like the original PMG5 rollout, with the bottom two machines using a simpler motherboard with a single CPU daughtercard and one dual-core CPU, and a top-end model with a twin CPU daughtercards, each with a dual-core CPU. There's no reason on Earth for them to hold back from shipping a quad system when it's only slightly different — and no more complex, from their POV — than the duallie they're already shipping.
As for SMT, since that question's just upthread from this post: It's coming in the POWER5-derived processor, which will be the successor to the current 970 family. Dual core and SMT can work together very well. In fact, that describes the POWER5 itself.
In the very narrow niche that Apple products have succeeded in enterprise accounts (publishing and AV departments) the clamor for quads would be significant. With Oracle releasing it's database for OSX, might not a quad Powermac to run it make sense?
Trust me on this: Nobody (except perhaps for a few lucky developers) will deploy Oracle on a PowerMac[1].
Now, on the other hand, if you roll out a quad Xserve... Max that sucker's RAM, cluster it with a couple more Xserves, attach it to an Xserve RAID or 10, and you're speaking Oracle's language.
Quad PowerMacs will be attractive to a fairly small but crucial set of people for whom there simply isn't enough power available, period: 3D artists, high-end video and audio engineers and artists, engineers and scientists, etc. Basically, the same people whose wallets cowered and whimpered when Apple rolled out the 30" Cinema Display.
[1] Yeah, Personal Oracle. But that's not what I'm talking about. You can run that on a laptop now.
Comments
Originally posted by hmurchison
Yeah I'm sorry but I just don't see where an application would suffer at all with a dual core implementation.
The only issue with the dual cores is the FSB being shared. They've doubled the L2 cache and the increased speed in "chip to chip" communication should more than make up for any latencies caused by the sharing of the FSB.
The more you look at it the more dual cores make perfect sense architecturally and financially. I think the impact to developers will be quite transparent.
Right, multicore is the future : at least for motorola, Intel, AMD and IBM ...
Originally posted by BRussell
Wait a second. Dual-core chips are better than dual-processor chips. It would be just like having a more efficient dual-processor system. Why would you switch if they replaced something worse with something better?
read the rest of the thread and you'll get it.
I'm saying just dual core isn't enough. They need to be dual chips there as well.
Originally posted by onlooker
read the rest of the thread and you'll get it.
I'm saying just dual core isn't enough. They need to be dual chips there as well.
There will be. The though of an affordable 4 socket Quad system was a mere fantasy. I don't think there is a mother less than $1000 that has 4 CPU sockets.
Apple's delimma on whether to offer 2 socke DC Powermacs will hinge on IBM yields. The question at hand is:
1. Does a Quad system provide enough sales to match the profit that can be obtained by selling 2 whole DC Powermac systems?
What Apple may want to do is keep the fastest CPU as a DC Powermac system and then take the processor that yields the highest and create a Quad system. Then say that the 2.5Ghz DC chips yield exceptionally well.
So let's say they can hit 3Ghz DC Q2 2005.
They should ship a $2799 DC 3Ghz system
and then follow that up with a $3499 Quad 2.5Ghz.
For some users the DC 3Ghz would be more beneficial to their needs while others would enjoy a Quad system better in their workflow.
Originally posted by onlooker
read the rest of the thread and you'll get it.
I'm saying just dual core isn't enough. They need to be dual chips there as well.
OK. I'm sure they will make "quads" (two dual-core machines) eventually. But I also have to wonder how efficient they will be, given the state of today's software. How many threads can an app generate and at what point does adding processors/cores not make sense?
Originally posted by RolandG
Does all this talk about dual-cores effectively kill any possibility to see multi-threading implemented into a PPC970 chip?
No, multi-threading still might see the light of day on a PPC chip.
Originally posted by NovaVengeance
No, multi-threading still might see the light of day on a PPC chip.
Originally posted by dfryer
Could you give an example of something that would perform *worse* on a dual core chip than on two separate chips? Is it just the possibility that they wouldn't up the FSB bandwidth correspondingly that makes you say this?
Applications at this point of course not, we don't have the chips to run the software on! The number one problem as I see it is the high potential for FSB saturation on the current implementation.
The other issue, now that we are talking about future chips, is the impact on hard single thread applications. This comes from my current belief that you will not be able to realize the same top speed in a dual core unit simply because of the heat issue. That is the near 2X increase in power disapation, though this may not be as bad as it sounds as IBM has a long way to go with power management. Of course if a single core chip is never produced with the technology IBM applies against the dual core unit, there will never be a possibility for comparison. The other reality is that there is very little in this world that can't be multithreaded to some extent, so this argument only applies to specific cases.
One must remember that the technology to build this dual core unit could be applied to building a very robust single core unit. Given the same amount of transitors very high performance could be achieved. This would be via SMT, AltVec2, a huge cache and other improvements. Of course they might not even call such a chip a 970, that sort of jump is probally reserved for the 980.
I would not be surprised to find that the first dual core chips end up delivered at a lower clock rate than is available in single core processors at the time. AMD has pretty much indicated that their first dual core chips will be a bit slower than the single core chips out at the time. Slower in the sense of clock rate, obviously system performance will be better, though some applications may be hampered.
All that being said if the new iMac is delivered with a dual core chip I'd be very surprised and extremely happy. This is exactly what Apple needs to deliver in its iMac3 to spark the market a bit. Dual core is a great solution for the low end of the market.
I think Peter Glaskowsky hits the same vein we're discussing here
According to Glaskowsky, customers running imaging and scientific applications developed for the Mac platform will appreciate the multicore design. "A dual-core chip is more effective than a single-core chip on problems that stress the computational resources of the chip, more than the front-side bus bandwidth. Because the 970FX has a very fast, efficient front-side bus, most Mac applications will favor the dual-core configuration."
Also of note in the article are the comments on the "Code Cracking" functions that appear new. I can see how this function fits better with the superior low latency benefits of dual core systems.
Also there seems to be a bit more confirmatin that the pipelines have indeed been lengthened somewhat which should help clocking.
I think this is good timing for Apple. I don't see FSB saturation happening until memory moves beyond DDR2 667. That gives Apple 12-18 months to milk the current setup and prepare for a 65nm 980 system which I believe is going to have Ondie Mem Controllers, SMT and dual cores as well.
The proof will be in the pudding if Apple starts to bang the "multithread your apps!" drum again. WWDC 2005 might have more sessions on threading and supporting SMP well in preparation for what seems to be "SMP nirvana" in 2006.
Originally posted by hmurchison
Eweek article about DC with Peter G comments
I think Peter Glaskowsky hits the same vein we're discussing here
According to Glaskowsky, customers running imaging and scientific applications developed for the Mac platform will appreciate the multicore design. "A dual-core chip is more effective than a single-core chip on problems that stress the computational resources of the chip, more than the front-side bus bandwidth. Because the 970FX has a very fast, efficient front-side bus, most Mac applications will favor the dual-core configuration."
Also of note in the article are the comments on the "Code Cracking" functions that appear new. I can see how this function fits better with the superior low latency benefits of dual core systems.
Also there seems to be a bit more confirmatin that the pipelines have indeed been lengthened somewhat which should help clocking.
I think this is good timing for Apple. I don't see FSB saturation happening until memory moves beyond DDR2 667. That gives Apple 12-18 months to milk the current setup and prepare for a 65nm 980 system which I believe is going to have Ondie Mem Controllers, SMT and dual cores as well.
The proof will be in the pudding if Apple starts to bang the "multithread your apps!" drum again. WWDC 2005 might have more sessions on threading and supporting SMP well in preparation for what seems to be "SMP nirvana" in 2006.
When you say milk the current setup I hope your only referring to 90nm process, and not 2.5GHz for 12 - 18 months.
I think we have extended our wishful thinking in this forum well past the line of sight. I think we'll see things like PCIe, and maybe a show like CBIT (or what ever that microprocessor showcase is) before we should go jumping off the deep end in here. It's all going to come crashing down really hard when over half of it may not even make it into the PowerMac by 2006. It's happened before. (Every Show)
I for one am bowing out of these drastic changes in the processor discussions until after SIGGRAPH. It's just way too early to start building something like this up. It's got catastrophe written all over it.
When you say milk the current setup I hope your only referring to 90nm process, and not 2.5GHz for 12 - 18 months.
Yes I mean 90nm tech for IBM probably has another year maximum. The cost savings of the 970MP systems will be too good to pass up.
I bet moving to a 1 socket Dual Core Powermac system could easily shave off as much as $200 from the retail pricing while keeping the same margins.
Originally posted by hmurchison
Yes I mean 90nm tech for IBM probably has another year maximum. The cost savings of the 970MP systems will be too good to pass up.
I bet moving to a 1 socket Dual Core Powermac system could easily shave off as much as $200 from the retail pricing while keeping the same margins.
any evidence to back that up?
id be surprised if the entire motherboard had a cost of 200 dollars.
Originally posted by applenut
any evidence to back that up?
id be surprised if the entire motherboard had a cost of 200 dollars.
Dual Socket Opteron motherboards
No PCI-X gets you a board for about $200. Adding PCI-X pushes you above $400 generally. Note this is for volume X86 chipsets. I'm sure Apples chipset are more spendy because they have to design them rather than just license them.
Your typical PC motherboard is pretty full featured at $149 roughly. Add another socket and that goes to $199-500 depending on whether you have SCSI or PCI-X etc.
Knowing that Apple likes to have their margins in the 20s I'm "assuming" the the Apple markup for adding another CPU socket would corrolate closer to $200 than the $50-100 premium seen on X86. I could be way off though and I realize that. I would be amazed however if Apple didn't save at least $100 bucks per unit stripping off the extra socket and using a smaller system controller.
I'll believe it when I see it...no wait...when it's on my desk, running Doom 3.
Originally posted by JRC
Again, I am the longest-registered account on Appleinsider!
Ahem.
Well, somebody already posted that link above, eh?
Hum.
A nice summary of all the MP 'rumours' that have been flying around. 'IBM refused to comment.'
Thankyou IBM, 'nuff said.
The Gadget Show (see macworld.co.uk) did a VERRRRRRRRRY superficial bench between Macs and PCs and guess which won on Games and Photoshop filter tests?
That's why Apple needs a chip that blows away the wintel cpus in non-apple lab' conditions.
The race for dual core is on. Hopefully Apple can offer a dual core chip running at 2.5-3.2 gig by next March. That will put Apple way ahead of AMD in 'gighz' and 'only' 800 mhz behind the 4 gig prescott if it ships.
Personally, I'm hoping IBM can reach 3.4 or even 3.6 gig with this even deeper pipelining. This instruction 'cracking' sounds like an interesting counter (ala 'micro opps' for x86?) to the deeper pipelining inefficiencies... What is interesting for me is the way different parts of the chip...integer, fpu, altivec will have different length pipelines...
This sounds like an awesome chip. Or at least the chip that will offer undisputed parity until the 'awesome' Power 5 derivative turns up. (A chip that truly promises to lay waist to the x86 opposition.)
Look at the positives of the MP.
1. More efficicent than dual processors. (with thread with instruction cracking...) So you get your dual processor effect more of the time. So a dual 2.5 should in MP speak by more like a 4 gig plus Pentium more of the time.
2. Higher clock speed. Putting AMD in the shade and running Intel down.
3. The MP should really help Apple's consumer desktops.
4. The MP can 'dual' the 'dual core'. That emphasises Apple's commitment to the workstation market they're clearly going after.
5. Dual Altivec. We know how powerful one altivec is. Dual that? Jaw drop!!!
6. Scalability of frontside bus, sleep modes etc. Speed throttling. This should allow a very interesting 'low clocked' dual core 1.6-2 gig G5MP into a Powerbook? Not gonna happen? We'll see. With 0.09 manufacturing behind them then the 'G4 in a Powerbook' won't happen camp may end up pleasantly surprised.
7. Big speed ups, FINALLY, for Apple's entire line from iBooks to Towers. When Apple gets to 3 gig plus it allows G5 adoption for everything...at various speed grades.
8. Dual fpu. Lightwave renders are going to be frighteningly fast. Not to mention Final Cut Pro', more real time rendering...Apple's workstation line optimised for MP is going to get even more pro people come back to Mac. More Maya adoption to the Mac workstation. At 25% and counting. Optimised for Maya Unlimited.
9. 'Snap' is back in the Mac. 'X' finally 'snaps'. yeah baby.
10. Casual benching eg 'Gadget Show' will show that Mac rocks on games. Macs will no longer be cpu bound on graphics cards like the Ati 9800 xt or the Nvid' 6800 Geforce. Important with next gen' games like Doom 3 and Half Life 2 already stressing 3.4 gig Pentiums to the limit at 1200 x 1024 resolutions even with latest 6800.
(11. Safari blows Interent ExSlower out the water in speed bench by Jobs and Phil. Hey, I want that MP vs Pentium 4 gig speed bake off...we all do, right?
Lemon Bon Bon (Optimistially...though constipatedly waiting for next 'March', eh?)
PS. 12. Mix MP with the 'Tiger' effect. eg Core Image, Core Video and Photoshop benching absolutely destroys the opposition!!! (Adobe, stand up and be counted. We'll just where their loyalties 'lie' after 'Tiger'.) Final Cut with Core Image and Video. iPhoto with Core Image. iMovie with Core Video. 'Tiger' optimised with new Compiler tech'? The Mac feels faster than ever. 'FEELS' fast and fluid as opposed to pauseworth, stately and elegant. Put in new Quicktime codec.
'Tiger' with 970 MP. The Mac, 'X' and 'PPC' promise finally lands!?!?!?!?
Originally posted by CharlesS
Ahem.
Ehhh.. ADMIN ALERT::::::! AI has been accidentally wiped clean of all post counts, and registration dates at least 3 times sense then.
Originally posted by Whisper
<waves hand around> OOOh OOOOOh I have a question!!</waves hand around>
Does this mean that instead of have 2 2.5GHz chips running on two separate 1.25GHz busses, there'll be two 3GHz cores running on a single shared 1GHz bus? Cause I think that would be a downgrade. If the new bus is 2.5 times slower -- lets say 2 times slower because there's only one bus worth of overhead now -- that means that the CPUs can only process half as much data. IIRC a significant chunk of the G5's speed comes from its super sweet FSB, which this new version would lack.
In practice, this is not that big a deal, because the system bus isn't a bottleneck in the overwhelming majority of cases: RAM is. There will be a few boundary cases where this implementation will hurt, but we are not looking at Son of MaxBus here. Hell, look at MaxBus: Even with two CPUs hanging off the same hoary old SDR bus, we got 160%-180% the speed of a single CPU for any task that didn't require streaming huge amounts of memory (which is to say, most tasks).
This is speculation, and it probably stems from the fact that the current 970-family CPUs only support whole-number bus multipliers. With a 3GHz core, you have a choice between a 1.5GHz bus and a 1GHz bus (I'm ignoring higher multipliers, since there is zero chance that Apple will use them). We know Elastic Bus can do 1GHz, but 1.5GHz would be a stretch. Not impossible, but of unknown difficulty. Not only is that an unprecedented frequency for a bus, but it requires that the companion chip (northbridge) run at 1.5GHz, and that's unprecedented as well. Remember, the heat pipe on the daughtercards in the original PowerMac G5s wasn't for the 970: It was for the northbridge.
As for what Apple will do with these: They hardly have to do anything to support dual dual-core 970s, because the interface to the northbridge is almost identical, and they've already designed a northbridge that supports two busses. That's a non-trivial design effort for a chip that's essentially brand new, and I can't really see them throwing it out. What they can do, and what they probably will do, is go back to something like the original PMG5 rollout, with the bottom two machines using a simpler motherboard with a single CPU daughtercard and one dual-core CPU, and a top-end model with a twin CPU daughtercards, each with a dual-core CPU. There's no reason on Earth for them to hold back from shipping a quad system when it's only slightly different — and no more complex, from their POV — than the duallie they're already shipping.
As for SMT, since that question's just upthread from this post: It's coming in the POWER5-derived processor, which will be the successor to the current 970 family. Dual core and SMT can work together very well. In fact, that describes the POWER5 itself.
Originally posted by Aphelion
In the very narrow niche that Apple products have succeeded in enterprise accounts (publishing and AV departments) the clamor for quads would be significant. With Oracle releasing it's database for OSX, might not a quad Powermac to run it make sense?
Trust me on this: Nobody (except perhaps for a few lucky developers) will deploy Oracle on a PowerMac[1].
Now, on the other hand, if you roll out a quad Xserve... Max that sucker's RAM, cluster it with a couple more Xserves, attach it to an Xserve RAID or 10, and you're speaking Oracle's language.
Quad PowerMacs will be attractive to a fairly small but crucial set of people for whom there simply isn't enough power available, period: 3D artists, high-end video and audio engineers and artists, engineers and scientists, etc. Basically, the same people whose wallets cowered and whimpered when Apple rolled out the 30" Cinema Display.
[1] Yeah, Personal Oracle. But that's not what I'm talking about. You can run that on a laptop now.