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.
970 Motherboard and Chip Delivery?
- Thread starter astray
- Start date
- Sort by reaction score
AidenShaw
macrumors P6
Re: About those 8 slots...
Windows 2000 (32bit) can handle up to 32 GiB of memory, Windows Server 2003 (32bit) supports up to 64GiB.... Both the G4 and the P4 support up to 64GiB of RAM, you don't need a 64-bit operating system to make good use of the extra RAM.
If max memory support is the only issue, it would make more sense to simply restrict the quantity of installed memory, not the number of slots. One could use 8 of the (often cheaper) 512 MiB DIMMs to get to 4GiB.
If the 8 slot story is true, I suspect that the final system will have interleaved (dual channel) memory, where DIMMs have to be used in pairs. You'd need one DIMM in the lower bank, and an identical DIMM in the matching slot in the upper bank.
If MacB really did see a proto mobo with the slots blocked off, I would be more inclined to believe that the proto unit didn't support the dual channel memory controller, and therefore the upper bank was blocked.
Windows 2000 (32bit) can handle up to 32 GiB of memory, Windows Server 2003 (32bit) supports up to 64GiB.... Both the G4 and the P4 support up to 64GiB of RAM, you don't need a 64-bit operating system to make good use of the extra RAM.
If max memory support is the only issue, it would make more sense to simply restrict the quantity of installed memory, not the number of slots. One could use 8 of the (often cheaper) 512 MiB DIMMs to get to 4GiB.
If the 8 slot story is true, I suspect that the final system will have interleaved (dual channel) memory, where DIMMs have to be used in pairs. You'd need one DIMM in the lower bank, and an identical DIMM in the matching slot in the upper bank.
If MacB really did see a proto mobo with the slots blocked off, I would be more inclined to believe that the proto unit didn't support the dual channel memory controller, and therefore the upper bank was blocked.
Re: Re: About those 8 slots...
Depends on what you mean as "good use" of the 64GB of memory. 64GB systems work great for servers which require the extra memory because they have multiple separate processes running, but not so well for, say, a video editting app which might want to allocate 5GB of memory all for itself (IIRC, the P4 36-bit memory addressing only allows an individual program to access a 32-bit window of memory, so any one app cannot have access to more than 4GB memory).
On the other hand, no Apple system today is bumping against the 4GB memory barrier either (2GB is the limit). Is there something in the PPC or Mac architecture that forbids use of the 32nd bit of pointers? Granted, 2GB is a practical limit for physical memory on a consumer system, but you'd think that if it were possible for Apple to use 4GB of memory on the XServe they would have ...
Depends on what you mean as "good use" of the 64GB of memory. 64GB systems work great for servers which require the extra memory because they have multiple separate processes running, but not so well for, say, a video editting app which might want to allocate 5GB of memory all for itself (IIRC, the P4 36-bit memory addressing only allows an individual program to access a 32-bit window of memory, so any one app cannot have access to more than 4GB memory).
On the other hand, no Apple system today is bumping against the 4GB memory barrier either (2GB is the limit). Is there something in the PPC or Mac architecture that forbids use of the 32nd bit of pointers? Granted, 2GB is a practical limit for physical memory on a consumer system, but you'd think that if it were possible for Apple to use 4GB of memory on the XServe they would have ...
Rincewind42
macrumors 6502a
Re: Re: Re: About those 8 slots...
IIRC the Memory Controller that Apple currently uses doesn't support more than 2GB (I refuse to use GiB, sorry, just too old =p) Because it is no problem for any single process in MacOS X to use a full 4GB of memory (and in fact, many system routines are loaded into a process' memory space above the 2GB line).
IIRC the Memory Controller that Apple currently uses doesn't support more than 2GB (I refuse to use GiB, sorry, just too old =p) Because it is no problem for any single process in MacOS X to use a full 4GB of memory (and in fact, many system routines are loaded into a process' memory space above the 2GB line).
Oh, thats quite simple. You have to pay on a per processor basis - make one look like two - pay double.
Makes sense right?
Actually, Windows XP Home handles either a single physical processor or dual logical processors (ie, P4 3.0GHz with HT) with no extra cost. Windows XP Pro, required for dual physical processors but which can't handle 3+ processors, handles 4 logical processors (2 HT-enabled P4s) at no extra cost.
On the other hand, applications like Oracle et al might have different licensing policies, but that has nothing to do with Windows reporting multiple logical processors.
Well, i was reading Macworld Uk today and came across this article from yesterday (15/5/03), i havent seen any references posted to it b4, but it looks mighty tantalising:
""IBM last night made additional details regarding its PowerPC 970 public.
The company released formerly limited circulation excerpts from October 2002's Microprocessor Review to selected media outlets, including Macworld UK.
The report, written by senior editor Tom Halfhill, confirms some details about the chip that have been circulating through the Internet.
Halfhill said he found it 'hard to imagine' that Apple would use anything else in its desktop Macs and servers."
The report begins: "Nobody at IBM would confirm rumours that a leading customer for the PowerPC 970 is Apple - and Apple is even more tight-lipped. Nevertheless, the 970 is such an obvious improvement over today's Motorola G4 family PowerPC chips that it's hard to imagine Apple using anything else in its desktop Macs and servers."
Claiming the chip to be "tailor-made" for professional publishing and media-processing applications, the report confirms: "It has 64-bit datapaths and memory addressing, yet it's natively compatible with 32-bit PowerPC software."
Additional features confirmed within the report include: a deeper data pipeline; bus rates of 900MHz; data bandwidth of 6.4GB/s; AltiVec. The PowerPC has 52 million transistors and is a stripped down version of IBM's Power4 processor, used in large-scale server installations.
"The deeper pipeline is especially welcome because it could allow the 970 to reduce the growing clock-frequency gap between today's PowerPC chips and the speedy x86 competition."
Calling Motorola's G4 and G4+ data pipelines "stunted" in comparison to Pentium 4, Halfhill suggests that initial estimates of 970 processor speeds are "conservative". In October it appeared the chips would offer speeds of 1.2-1.8GHz.
The report reveals the 970 to offer a 16-stage pipeline - 9 more than the G4+ and 12 more than the G4 and G3 processors. The processor offers two built-in floating point units, AltiVec support, a 64K L1 cache with a 512K L2 cache and a front side data bandwidth (FSB) of 2 x 3.2GB/s. The G4+ offers a 1GB FSB.
The FSB is the part of the processor that allows data to be moved from memory into the processor to be manipulated, and back again.
"The PowerPC 970 is clearly in a different class than existing G4+, G4 and G3 PowerPC chips. Its deeper pipelines and much faster FSB fix the most serious shortcomings of today's PowerPCs."
Looking at the history of the AIM (Apple, IBM, Motorola) alliance, the report says: "In fact, a 64-bit PowerPC was planned right from the start, when IBM, Motorola and Apple began creating the PowerPC architecture in the early 90's, the alliance promised to deliver three 32-bit processors - the 601, 603 and 604 and one 64-bit implementation, the 620."
The latter chip was never used on a Mac, but the inherent architecture of the PowerPC has always supported 64-bit memory addressing.
Execution is significantly faster - the 970 can fetch up to eight instructions per clock cycle, can dispatch up to five instructions per cycle to the function units, issue up to eight instructions per cycle and retire results for up to five instructions per cycle. "The 970 can juggle an unusually large number of instructions in its piopelines," Halfhill writes.
The FSB implementation is crucial. The processor drives this at up to 900MHz. The Pentium 4 offers 533MHz and the Athlon XP gets 333MHz.
The report offers several comparisons with competing processors, and the 970 seemingly surpasses or competes with x86-based competition. Its adoption would help Apple remain competitive in the market.
Halfhill concludes: "It's a good bet the 970 will also end up in a Mac - unless Apple's thinking is even more different than advertised."
Online, Mac rumour sites have begun to claim 970 processors are already arriving with manufacturer's in the Far East - who are working for a client called Apple.
Apple does not comment on rumour or unreleased products.""
The bits i liked best were:
"The report offers several comparisons with competing processors, and the 970 seemingly surpasses or competes with x86-based competition. Its adoption would help Apple remain competitive in the market."
and "Halfhill concludes: "It's a good bet the 970 will also end up in a Mac - unless Apple's thinking is even more different than advertised.""
""IBM last night made additional details regarding its PowerPC 970 public.
The company released formerly limited circulation excerpts from October 2002's Microprocessor Review to selected media outlets, including Macworld UK.
The report, written by senior editor Tom Halfhill, confirms some details about the chip that have been circulating through the Internet.
Halfhill said he found it 'hard to imagine' that Apple would use anything else in its desktop Macs and servers."
The report begins: "Nobody at IBM would confirm rumours that a leading customer for the PowerPC 970 is Apple - and Apple is even more tight-lipped. Nevertheless, the 970 is such an obvious improvement over today's Motorola G4 family PowerPC chips that it's hard to imagine Apple using anything else in its desktop Macs and servers."
Claiming the chip to be "tailor-made" for professional publishing and media-processing applications, the report confirms: "It has 64-bit datapaths and memory addressing, yet it's natively compatible with 32-bit PowerPC software."
Additional features confirmed within the report include: a deeper data pipeline; bus rates of 900MHz; data bandwidth of 6.4GB/s; AltiVec. The PowerPC has 52 million transistors and is a stripped down version of IBM's Power4 processor, used in large-scale server installations.
"The deeper pipeline is especially welcome because it could allow the 970 to reduce the growing clock-frequency gap between today's PowerPC chips and the speedy x86 competition."
Calling Motorola's G4 and G4+ data pipelines "stunted" in comparison to Pentium 4, Halfhill suggests that initial estimates of 970 processor speeds are "conservative". In October it appeared the chips would offer speeds of 1.2-1.8GHz.
The report reveals the 970 to offer a 16-stage pipeline - 9 more than the G4+ and 12 more than the G4 and G3 processors. The processor offers two built-in floating point units, AltiVec support, a 64K L1 cache with a 512K L2 cache and a front side data bandwidth (FSB) of 2 x 3.2GB/s. The G4+ offers a 1GB FSB.
The FSB is the part of the processor that allows data to be moved from memory into the processor to be manipulated, and back again.
"The PowerPC 970 is clearly in a different class than existing G4+, G4 and G3 PowerPC chips. Its deeper pipelines and much faster FSB fix the most serious shortcomings of today's PowerPCs."
Looking at the history of the AIM (Apple, IBM, Motorola) alliance, the report says: "In fact, a 64-bit PowerPC was planned right from the start, when IBM, Motorola and Apple began creating the PowerPC architecture in the early 90's, the alliance promised to deliver three 32-bit processors - the 601, 603 and 604 and one 64-bit implementation, the 620."
The latter chip was never used on a Mac, but the inherent architecture of the PowerPC has always supported 64-bit memory addressing.
Execution is significantly faster - the 970 can fetch up to eight instructions per clock cycle, can dispatch up to five instructions per cycle to the function units, issue up to eight instructions per cycle and retire results for up to five instructions per cycle. "The 970 can juggle an unusually large number of instructions in its piopelines," Halfhill writes.
The FSB implementation is crucial. The processor drives this at up to 900MHz. The Pentium 4 offers 533MHz and the Athlon XP gets 333MHz.
The report offers several comparisons with competing processors, and the 970 seemingly surpasses or competes with x86-based competition. Its adoption would help Apple remain competitive in the market.
Halfhill concludes: "It's a good bet the 970 will also end up in a Mac - unless Apple's thinking is even more different than advertised."
Online, Mac rumour sites have begun to claim 970 processors are already arriving with manufacturer's in the Far East - who are working for a client called Apple.
Apple does not comment on rumour or unreleased products.""
The bits i liked best were:
"The report offers several comparisons with competing processors, and the 970 seemingly surpasses or competes with x86-based competition. Its adoption would help Apple remain competitive in the market."
and "Halfhill concludes: "It's a good bet the 970 will also end up in a Mac - unless Apple's thinking is even more different than advertised.""
Rincewind42
macrumors 6502a
Sigh
I wish that article had never been released at this point - about the only thing it offers that everything else that has been published up to this point is Speculation on Apple from a non-Mac rumors source 😀. Not trying to flame you, just wishing there really was new information out there...
skunk
macrumors G4
NEW FROM MACBIDOUILLE
Just in: they surmise that the absence of releases of Darwin code since February except for official updates is because there's too much PPC970 stuff in there these days to hide. Same goes for Developer Tools: no visible progress since December. Another thing: confirmation that 970s at 1.7GHz are due out in July, manufactured using SiLK (low-K silicon) at 0.13microns, which will support an eventual processor speed of 2.3GHz or more. When they move to .09micron wafers at the start of 2004, the material will be something else again, enabling speeds of 2.5GHz, as originally announced. At the end of 2004, IBM will add "strained silicon", enabling the Power 5+ to reach 3GHz, and the PPC970/980 4.5GHz. If all goes according to plan.....
Just in: they surmise that the absence of releases of Darwin code since February except for official updates is because there's too much PPC970 stuff in there these days to hide. Same goes for Developer Tools: no visible progress since December. Another thing: confirmation that 970s at 1.7GHz are due out in July, manufactured using SiLK (low-K silicon) at 0.13microns, which will support an eventual processor speed of 2.3GHz or more. When they move to .09micron wafers at the start of 2004, the material will be something else again, enabling speeds of 2.5GHz, as originally announced. At the end of 2004, IBM will add "strained silicon", enabling the Power 5+ to reach 3GHz, and the PPC970/980 4.5GHz. If all goes according to plan.....
BaghdadBob
macrumors 6502a
Snowy_River
macrumors 68030
Alas, so much for that theory, skunk:
The Darwin team is pleased to announce the immediate availability of
the Darwin 6.6 and Darwin 6.5 source code corresponding to Mac OS X
10.2.6 and Mac OS X 10.2.5.
http://developer.apple.com/darwin/
The Darwin team is pleased to announce the immediate availability of
the Darwin 6.6 and Darwin 6.5 source code corresponding to Mac OS X
10.2.6 and Mac OS X 10.2.5.
http://developer.apple.com/darwin/
regarding the memory issue
i thought that i read somewhere that OSX can only allocate 1gb of memory to any application
so if you have 1.5gb and run photoshop . the max osx can give photoshop is 1gig..
is this accurate?
I'm sure that's OS 9, you were limited to 999,999K as the maximum allocation.
OS X has dynamic memory allocation, no limitations at all apart from available disk space for paging and the amount of physical ram that can be addressed by the cpu or installed in the mac itself. The G4 cpu itself can handle 4Gb of physical RAM and 64Gb (possibly wrong here) of virtual memory.
i thought that i read somewhere that OSX can only allocate 1gb of memory to any application
so if you have 1.5gb and run photoshop . the max osx can give photoshop is 1gig..
is this accurate?
I'm sure that's OS 9, you were limited to 999,999K as the maximum allocation.
OS X has dynamic memory allocation, no limitations at all apart from available disk space for paging and the amount of physical ram that can be addressed by the cpu or installed in the mac itself. The G4 cpu itself can handle 4Gb of physical RAM and 64Gb (possibly wrong here) of virtual memory.
The G4 has 36 lines to select an address on its external bus, cache logic could also work in 36 bits mode. The point is Mac OS X will probably never use the memory managment hacks to take avantage of those 4 extra bits, since by the time people will ask for more than 4 GiB all the Macs will probably ship with 64 bits CPUs.
The PowerPC 970 can physically address 2^42 bytes of memory that is 4398046511104 bytes, 4096 GiB.
AidenShaw
macrumors P6
You've got that backwards.
The G4 can handle 64GiB of physical RAM (those 36-bits of address lines). No Apple mobo has supported more than 2GiB, however.
An individual process gets 4GiB of virtual memory (the 32-bit virtual address). This memory is often divided into system and process spaces, often 2GiB for process and the other 2GiB for system. (Windows has an option to move the split to 3GiB process and 1GiB system, with Linux you can modify the kernel if you want to move the split.)
AidenShaw
macrumors P6
This is exactly the way to describe it - the 42-bit physical limit is a design choice for the PPC970, not a PPC architectural limit. Individual chip designs can stop at practical limits.
The designers probably thought that 4TiB of RAM was enough headroom for the life of any PPC-970 derived chip. There's no point in wasting transistors doing a full 64-bit physical addressing implementation, when something far narrower will never be a bottleneck.
pimentoLoaf
macrumors 68010
Snowy_River
macrumors 68030
Just to be nit-picky, that should be '4TeB', or '4TB', as it's four tera-bytes.
But, really, I doubt it will be long before 4TB seems like a completely reasonable amount of memory for our computers to have. After all, I can remember all too well when 1MB of RAM seemed like a substantial amount...
Rincewind42
macrumors 6502a
MacOS X makes no such seperation, but rather each process gets the entier 4GB address space, and then system libraries loaded into that space and mearly take the amount that they require. In theory, the OS uses 512MB of address space for frameworks, 256MB for the stack, and little bits here and there for minor services (like page 0 for NULL page fault checks). But if system frameworks only need 128MB then that space can be allocated to the process itself.
BaghdadBob
macrumors 6502a
Depends on what your definition of "long" is. Right now you are considered to have a buttload of RAM if you have 2 GB, and a buttload of HD space if you have 500 GB (250x2GB). It's going to have to become common for a TB of HD space to be in your computer long before even 1 TB of RAM is considered normal.
How much space will your Photoshop folder take up then, eh? 5 GB? 😱
I was just reminiscing how much the CD-ROM changed computing by changing the realistic sizes you could make applications before someone would go through a 20 disk installation process....ughhh....remember floppies, anyone?
Of course, we have all these revolutions to thank for application bloat (along with Windowsfication).
Anyone here know off the top of their head what a DVD-ROM holds?
We're both right/wrong
The G4 has 36 lines to select an address on its external bus, cache logic could also work in 36 bits mode. The point is Mac OS X will probably never use the memory managment hacks to take avantage of those 4 extra bits, since by the time people will ask for more than 4 GiB all the Macs will probably ship with 64 bits CPUs.
The PowerPC 970 can physically address 2^42 bytes of memory that is 4398046511104 bytes, 4096 GiB.
and
quote:
------------------------------------------------------------------------
Originally posted by barkmonster
The G4 cpu itself can handle 4Gb of physical RAM and 64Gb (possibly wrong here) of virtual memory.
------------------------------------------------------------------------
You've got that backwards.
The G4 can handle 64GiB of physical RAM (those 36-bits of address lines). No Apple mobo has supported more than 2GiB, however.
I'm well aware of this but I've also read that it's awkward to implement a 36bit address space with a 32bit cpu. (read it on here not on a technical site like arstechnica or anything like that).
Also, I know that not all G4s support the 36bit address space, only the G4+ cpus can. Even then, only the real (phsyical ?) address space is 36bit, the virtual address space is still only 32bit. The G4 as a whole (as in ANY G4 past or present) isn't capable of using more than a 32bit address space. I guess this doesn't matter so much for models with the 745x series cpus but it would be wrong to claim the G4 has a 36bit address space across the board.
There was a link on here recently to an IBM pdf of the presentation last year on the PPC970 with this info in a table for comparison with G3, Power4 and PPC970 cpus. I can't be bothered searching the motorola site for more info because of those awful javascript menus they use 😀
The G4 has 36 lines to select an address on its external bus, cache logic could also work in 36 bits mode. The point is Mac OS X will probably never use the memory managment hacks to take avantage of those 4 extra bits, since by the time people will ask for more than 4 GiB all the Macs will probably ship with 64 bits CPUs.
The PowerPC 970 can physically address 2^42 bytes of memory that is 4398046511104 bytes, 4096 GiB.
and
quote:
------------------------------------------------------------------------
Originally posted by barkmonster
The G4 cpu itself can handle 4Gb of physical RAM and 64Gb (possibly wrong here) of virtual memory.
------------------------------------------------------------------------
You've got that backwards.
The G4 can handle 64GiB of physical RAM (those 36-bits of address lines). No Apple mobo has supported more than 2GiB, however.
I'm well aware of this but I've also read that it's awkward to implement a 36bit address space with a 32bit cpu. (read it on here not on a technical site like arstechnica or anything like that).
Also, I know that not all G4s support the 36bit address space, only the G4+ cpus can. Even then, only the real (phsyical ?) address space is 36bit, the virtual address space is still only 32bit. The G4 as a whole (as in ANY G4 past or present) isn't capable of using more than a 32bit address space. I guess this doesn't matter so much for models with the 745x series cpus but it would be wrong to claim the G4 has a 36bit address space across the board.
There was a link on here recently to an IBM pdf of the presentation last year on the PPC970 with this info in a table for comparison with G3, Power4 and PPC970 cpus. I can't be bothered searching the motorola site for more info because of those awful javascript menus they use 😀