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.
New Motherboard Details?
- Thread starter MacRumors
- Start date
- Sort by reaction score
macB rumours
It's quite obvious to me that macB. is being fed this information by Apple, the shear volume and that fact that Apple has not denied any of it point to this, after all if there were no rumours at all one would have to probably assume that Apple has nothing up it's sleeve, and even Apple can't afford that, but at the same time it has to distance itself from the "source", hence the french aspect.
And we are all busy speading the word, albeit unofficially.
It's quite obvious to me that macB. is being fed this information by Apple, the shear volume and that fact that Apple has not denied any of it point to this, after all if there were no rumours at all one would have to probably assume that Apple has nothing up it's sleeve, and even Apple can't afford that, but at the same time it has to distance itself from the "source", hence the french aspect.
And we are all busy speading the word, albeit unofficially.
Re: Re: Stop and think for a second...
Very true, but what rumor site has been 100% accurate. MacBidouille like any other site has had its hits and misses. A lot depends on their sources. In the end, only time will tell. Rumors are usually a little bit of fact enhanced by speculation. The farther a rumor gets from its source, the greater the speculation and the less the fact. Each person who relays it, alters it in some manner according to their own perception and knowledge.
To prove this point. line up 10 people 10 feet apart. Give the first person a specific message of at least 10 words, and have them wait 2 minutes then walk over to the next person and repeat the message, and so on down the line. By the time it reaches the 10th person, there is a better then 90% probability it will be dramatically changed and possibly not even recognizable.
Very true, but what rumor site has been 100% accurate. MacBidouille like any other site has had its hits and misses. A lot depends on their sources. In the end, only time will tell. Rumors are usually a little bit of fact enhanced by speculation. The farther a rumor gets from its source, the greater the speculation and the less the fact. Each person who relays it, alters it in some manner according to their own perception and knowledge.
To prove this point. line up 10 people 10 feet apart. Give the first person a specific message of at least 10 words, and have them wait 2 minutes then walk over to the next person and repeat the message, and so on down the line. By the time it reaches the 10th person, there is a better then 90% probability it will be dramatically changed and possibly not even recognizable.
Memory architecture of PPC 970 motherboard from MacBidouille
Translated from French by your's truly (French is my first language)...
"OK, take this with a major dose of salt, as usual.
The dual-processor motherboards should have 8 memory slots in all. There seems to be 4 per processor. These slots are currently compatible with 4 GiB memory modules. Therefore we should have a maximum memory capacity of 32 GiB !!
We have no clue how the Mac will manage it's RAM. We don't know if each processor will have it's own dedicated Twin Bank memory allocation, or if the two groups of 4 slots will be combined to form one large Twin Bank.
The question now remains: How much will the ultimate PowerMac 970 configuration with 32 GiB of RAM cost?"
My comments on this:
1) I find it strange that they use the words "should have 8 memory slots". But that is the exact (translated) wording which their article uses. I speak French fluently. Does that mean that only the dual-processor motherboards will have 8 slots? That would be surprising, as that would mean different motherboards for single and dual processors. Of course that would explain a former post on their site which mentioned that only 4 of the slots on the board their contact saw were working (had only 1 processor?). Or single-processor PowerMac 970s would only have access to 4 of the slots? More likely, but still a strange and suprising restriction.
2) This article is the first place I've ever seen the term "Twin Bank" used. I therefore don't know what this technology is. My educated guess is that it takes two memory busses, and multiplexes them together into a single bus running at twice the effective clock rate. In other words: a RAM RAID. That would be fabulous, as it would allow using two 200 Mhz DDR (that is supposedly what these motherboards use) banks to acheive an effective speed of 400 MHz DDR. The PPC 970's FSB can handle 450 MHz DDR, so Apple would be using the FSB close to it's maximum capacity. Goodbye data starvation! 😀 If my guess is right, this thing would kick a G4's butt, even running at the same clock rate. Can anyone shed some light on this?
Translated from French by your's truly (French is my first language)...
"OK, take this with a major dose of salt, as usual.
The dual-processor motherboards should have 8 memory slots in all. There seems to be 4 per processor. These slots are currently compatible with 4 GiB memory modules. Therefore we should have a maximum memory capacity of 32 GiB !!
We have no clue how the Mac will manage it's RAM. We don't know if each processor will have it's own dedicated Twin Bank memory allocation, or if the two groups of 4 slots will be combined to form one large Twin Bank.
The question now remains: How much will the ultimate PowerMac 970 configuration with 32 GiB of RAM cost?"
My comments on this:
1) I find it strange that they use the words "should have 8 memory slots". But that is the exact (translated) wording which their article uses. I speak French fluently. Does that mean that only the dual-processor motherboards will have 8 slots? That would be surprising, as that would mean different motherboards for single and dual processors. Of course that would explain a former post on their site which mentioned that only 4 of the slots on the board their contact saw were working (had only 1 processor?). Or single-processor PowerMac 970s would only have access to 4 of the slots? More likely, but still a strange and suprising restriction.
2) This article is the first place I've ever seen the term "Twin Bank" used. I therefore don't know what this technology is. My educated guess is that it takes two memory busses, and multiplexes them together into a single bus running at twice the effective clock rate. In other words: a RAM RAID. That would be fabulous, as it would allow using two 200 Mhz DDR (that is supposedly what these motherboards use) banks to acheive an effective speed of 400 MHz DDR. The PPC 970's FSB can handle 450 MHz DDR, so Apple would be using the FSB close to it's maximum capacity. Goodbye data starvation! 😀 If my guess is right, this thing would kick a G4's butt, even running at the same clock rate. Can anyone shed some light on this?
Re: Memory architecture of PPC 970 motherboard from MacBidouille
Even in a cc-NUMA architecture any processor should be able to access remote memory that is local to a missing processor, this is not true with Opteron systems since the memory controller is part of the CPU.
Before you ask:
http://lse.sourceforge.net/numa/faq/
The PowerPC 970 uses a point to point connexion not a bus, each processor has its own dedicated wires between its socket and the chipset, whereas both G4e were on the same small daughtercard. Early rumors said you could distinguish the single CPU motherboard from de the dual one, this could be the reason.
Even in a cc-NUMA architecture any processor should be able to access remote memory that is local to a missing processor, this is not true with Opteron systems since the memory controller is part of the CPU.
Before you ask:
http://lse.sourceforge.net/numa/faq/
Re: Re: Memory architecture of PPC 970 motherboard from MacBidouille
Hey thanks. That was quite an eductional read. Basically, NUMA is a micro-network on a motherboard. Sounds like an interesting architecture technology. Though I have trouble seeing this being used in a PowerMac, as (because of physical limits) I doubt there will be more than 4 processors in a machine. This NUMA thing sounds like it would require some pretty complex support circuitry.
Hey thanks. That was quite an eductional read. Basically, NUMA is a micro-network on a motherboard. Sounds like an interesting architecture technology. Though I have trouble seeing this being used in a PowerMac, as (because of physical limits) I doubt there will be more than 4 processors in a machine. This NUMA thing sounds like it would require some pretty complex support circuitry.
BenRoethig
macrumors 68030
Re: Memory architecture of PPC 970 motherboard from MacBidouille
Well, now that would be something worth thinking in detail. It definately makes a lot of sense on HD's and it would make even more sense on RAM as a RAm module failure wouldn't cause more permanent damage on a Raid as it would on a normal bank.
I guess the problem up to now was the slow bus, efficentlky preventing data from getting to the CPU, and eventually the CPU beeing not fast enough.
Anyway, the only problem I see now is that RAM modules would need to be a perfect match.
I checked SDRAM registers about a a year ago, and no two modules had the same setup.
So it's good bye cheap addonram, and ups that RAM is oh so expensive.
Well, now that would be something worth thinking in detail. It definately makes a lot of sense on HD's and it would make even more sense on RAM as a RAm module failure wouldn't cause more permanent damage on a Raid as it would on a normal bank.
I guess the problem up to now was the slow bus, efficentlky preventing data from getting to the CPU, and eventually the CPU beeing not fast enough.
Anyway, the only problem I see now is that RAM modules would need to be a perfect match.
I checked SDRAM registers about a a year ago, and no two modules had the same setup.
So it's good bye cheap addonram, and ups that RAM is oh so expensive.
I think Apple might leak this information intentionally, and leak it to a french site because it's in the 'all so evil outlaw France' that people who ask themselves why Apple doesn't sue the site, will eventually think 'oh it's one of those french sites - Apple couldn't do anything if they wanted to.'
Knowing Apple employs quite a lot of native french speakers (just watch ADC TV) does help developing such thoughts 😉
Why would Apple leak so many details? Well, simple, they need a pull factor to the WWDC. They figure hardware is the most interesting issue apart from 'future mac platform' and 'get a preview version of panther'
Also they don't want so many people to switch - to a PC - in the time beeing.
Re: Memory architecture of PPC 970 motherboard from MacBidouille
Does no-one remeber twin-bank memory? My Powermac 7500 supported it. If you paired identical speed and capacity SIMMS then the data transferred at twice the rate. Gave about a 5% speed boost from memory. Apple should do this to feed the 970 monster!
Does no-one remeber twin-bank memory? My Powermac 7500 supported it. If you paired identical speed and capacity SIMMS then the data transferred at twice the rate. Gave about a 5% speed boost from memory. Apple should do this to feed the 970 monster!
AidenShaw
macrumors P6
Nothing new, interleaved and RAID memory on Intel servers very common
Interleaved memory (using 2 banks of memory in unison for twice the bandwidth) is common on Intel desktops - most of the RDRAM and DDR boards use dual-channel memory. (Hint - if the board requires that DIMMs be inserted in matched pairs, almost certainly it is dual channel or interleaved.)
The higher end Intel servers most often have 4-way interleaved memory - 4 DIMMs clocked in unison for 4 times the bandwidth.
As far as "RAID" (Redundant Arrays of Independent DIMMs) memory, HPaq servers have hot swappable RAID RAM, spare DIMM support, and other techniques:
http://h71025.www7.hp.com/support/r...=5730&source=TC020905TB.xml&dt=21&docid=11063
This technology brief describes the Hot Plug RAID Memory technology developed by Compaq to give enterprise-class servers the level of memory fault tolerance todays 7x24 applications demand. It explains memory operation, provides background information on memory reliability, reviews current error detection and correction techniques....
[It] explains why the likelihood of memory errors grows with increased memory capacity. It discusses Hot Plug RAID Memory in depth and provides information on less robust, alternative fault-tolerant memory solutions.
IBM also has advanced memory error handling:
http://www.pc.ibm.com/us/eserver/xseries/x440.html
Active Memory
- Introduces memory mirroring capabilities to dramatically reduce unscheduled downtime for memory failures through a RAID-like mirroring of memory data - a first among industry-standard systems
- Provides first-ever Memory ProteXion similar to hot-spare on DIMM that uses spare bits for redundancy that re-route data around a failed chip on a DIMM in the event of a failure
- Offers third-generation Chipkill memory using off-the-shelf industry-standard DIMMs as Chipkill algorithm is now integrated into the XA-32 chipset for best performance and lowest cost
Of course, today Apple doesn't even offer simple error detection by supporting common ECC memory. When a Mac gets a memory error (and "when" is the right word, not "if") you get a crash, or corrupted data, or something unusual happening.
Interleaved memory (using 2 banks of memory in unison for twice the bandwidth) is common on Intel desktops - most of the RDRAM and DDR boards use dual-channel memory. (Hint - if the board requires that DIMMs be inserted in matched pairs, almost certainly it is dual channel or interleaved.)
The higher end Intel servers most often have 4-way interleaved memory - 4 DIMMs clocked in unison for 4 times the bandwidth.
As far as "RAID" (Redundant Arrays of Independent DIMMs) memory, HPaq servers have hot swappable RAID RAM, spare DIMM support, and other techniques:
http://h71025.www7.hp.com/support/r...=5730&source=TC020905TB.xml&dt=21&docid=11063
This technology brief describes the Hot Plug RAID Memory technology developed by Compaq to give enterprise-class servers the level of memory fault tolerance todays 7x24 applications demand. It explains memory operation, provides background information on memory reliability, reviews current error detection and correction techniques....
[It] explains why the likelihood of memory errors grows with increased memory capacity. It discusses Hot Plug RAID Memory in depth and provides information on less robust, alternative fault-tolerant memory solutions.
IBM also has advanced memory error handling:
http://www.pc.ibm.com/us/eserver/xseries/x440.html
Active Memory
- Introduces memory mirroring capabilities to dramatically reduce unscheduled downtime for memory failures through a RAID-like mirroring of memory data - a first among industry-standard systems
- Provides first-ever Memory ProteXion similar to hot-spare on DIMM that uses spare bits for redundancy that re-route data around a failed chip on a DIMM in the event of a failure
- Offers third-generation Chipkill memory using off-the-shelf industry-standard DIMMs as Chipkill algorithm is now integrated into the XA-32 chipset for best performance and lowest cost
Of course, today Apple doesn't even offer simple error detection by supporting common ECC memory. When a Mac gets a memory error (and "when" is the right word, not "if") you get a crash, or corrupted data, or something unusual happening.
go to the 970 odd photos thread...
hey guys... i posted in the 970 odd rumors thread yesterday a link to an ibm pdf file about the 970 that basically talked about apple using the 970 in its new machines.. however today that link has been made redundent by ibm.. another member has taken a screenhsot of the last page which is quite interesting if you want to take a look.
hey guys... i posted in the 970 odd rumors thread yesterday a link to an ibm pdf file about the 970 that basically talked about apple using the 970 in its new machines.. however today that link has been made redundent by ibm.. another member has taken a screenhsot of the last page which is quite interesting if you want to take a look.
silvergunuk
macrumors regular
biscuits are total ace!
why dont you just paste the image in here? well all take copies and start ranting even more
why dont you just paste the image in here? well all take copies and start ranting even more
i have always been confused how a place in france could get info before people in the states... maybe thats sayign something...
Re: Nothing new, interleaved and RAID memory on Intel servers very common
Yep, used to be on macs too. Sort of a pain in the ass if you ask me. Mobo has to have twice as many slots for the same number of ram upgrades, dimms (nee simms) have to be matched, installation is more confusing for novices etc (no not that slot, that one, there, no, to the left, up one...) but its worth it if it makes it run faster...
Yep, used to be on macs too. Sort of a pain in the ass if you ask me. Mobo has to have twice as many slots for the same number of ram upgrades, dimms (nee simms) have to be matched, installation is more confusing for novices etc (no not that slot, that one, there, no, to the left, up one...) but its worth it if it makes it run faster...