Confirmation of Westmere Memory Config

[xgman]

eidt: the 6 core is on a workstation board it appears and not a server board. The server boards may or may not be dual channel capable but the 4 slot boards are 3 + 1, so disregard the rest of this post. Sorry about the misinfo. If only Apple had spelled this out for us. . . . .


I spoke with the Micron engineer who personally tested the 4, 6 and 12 core 2010 mac pros on August 9th with regard to memory compatibility.
After a long conversation and some additional research he confirmed that in fact the six core Westmere Mac Pro is a dual channel memory system (integrated in the chipset) and NOT a triple channel system like the 2009 Nehalem chipset Mac Pro systems were. Therefore the best way to populate the 4 slots is to populate them all with identical memory which will run in effect in dual channel mode.

In addition. He stated that he in fact tested 8GB sticks of 1333 ram in the six core system and it did recognize and run the 8GB sticks. In fact Crucial has these 8GB sticks (albeit very expensive) listed now on their site as compatible with the 2010 Mac Pros.

I certainly wish I had this info when I placed my memory order with Transintl. Nevertheless, it seems to indicate that the max amount of ram for the six core is looking much better indeed. By the way, the board is a derivative of the Intel 5520 for anyone who is interested.

"Unlike Core i7, however, mainstream desktop/mobile Westmere chips will not feature a QPI bus or a triple-channel memory controller. Rather, we'll revert to a dual-channel controller and have no QPI clock to worry about"

[reel2reel]

Nice, thanks for posting this. We should get tests from diggloyd, too. :)

[xgman]

Nice, thanks for posting this. We should get tests from diggloyd, too. :)

I forwarded this to him. he is going to test this as soon as a system arrives.

[skiffx]

I forwarded this to him. he is going to test this as soon as a system arrives.

I just cancelled my order of 4x4 and replaced it with 3x4gb thinking it was triple, so this is 100% confirmed that it runs as dual channel and 4x4 would run faster?!

[xgman]

I just cancelled my order of 4x4 and replaced it with 3x4gb thinking it was triple, so this is 100% confirmed that it runs as dual channel and 4x4 would run faster?!

Yes. Everyone wrongly assumed it would be as the 2009. (well most everyone )

[lssmit02]

I'm not sure this is right. According to Apple (http://www.apple.com/macpro/specs.html#overlay-compare), the chip in the six core is the W3680. According to Intel's site, # of Memory Channels 3. Link (http://ark.intel.com/Product.aspx?id=47917&processor=W3680&spec-codes=SLBV2). I assume this means it's triple channel memory. Note also, Intel's site also says the maximum amount of memory is 24GB.

[xgman]

I'm not sure this is right. According to Apple (http://www.apple.com/macpro/specs.html#overlay-compare), the chip in the six core is the W3680. According to Intel's site, . Link (http://ark.intel.com/Product.aspx?id=47917&processor=W3680&spec-codes=SLBV2). I assume this means it's triple channel memory. Note also, Intel's site also says the maximum amount of memory is 24GB.


Westmere is based on dual channel. Not sure what that is referring to above.

Westmere’s memory controller can drive two DIMMs per channel at the full 1.33GT/s. Current Nehalem systems (and most AMD systems) run two DIMMs per channel at reduced bandwidth; Nehalem’s memory controller operates two DIMMs per channel at 1.06GT/s, sacrificing about 20% of the bandwidth.

the Nehalem core supported three channels of DDR3 memory, but you could only acheive maximum speed (1333mhz) with a single DDR3 DIMM per channel. With Westmere, Intel has tweaked the memory controller for performance, so now you can still get to that 1333mhz number with two DIMMs installed per channel. An end user can now buy cheaper, lower density DIMMs without fear of performance degradation. The new and improved memory controller also supports new low-voltage DDR3 memory modules.

[Nautigar]

Westmere is based on dual channel. Not sure what that is referring to above.


Do you have a link or source to support this claim?

I am rather sure this is wrong ...

[deconstruct60]

After a long conversation and some additional research he confirmed that in fact the six core Westmere Mac Pro is a dual channel memory system (integrated in the chipset) and NOT a triple channel system like the 2009 Nehalem chipset Mac Pro systems were.


Err the Nehalem 2009 also is dual interleave mode if fill 4 slots. If these experiements were in filling 4 slots then this is no suprise, nor an confirmation that it solely runs in dual interleave mode in all contexts. ( Very unlikely that this interleave mode does not change as the configurations are changed. )




Therefore the best way to populate the 4 slots is to populate them all with identical memory which will run in effect in dual channel mode.


The problem with 8GB DIMMs and that approach is that you end up with 32GB. The spec sheets on the 3680 say that it tops out at 24GB.

http://ark.intel.com/Product.aspx?id=47917&processor=W3680&spec-codes=SLBV2

[xgman]

To be honest we need to know exactly which model Intel motherboard this is based on and is it in fact the W3680 or the X5680. The resulting answer is quite different. The W3680 is a workstation processor that fits on theWX58EP Intel board and does have triple channel 3 slots plus on extra slot.

The X5680 fits on a server board, like the Intel 5520 and they use dual channel.

So which is it?

[trankdart]

So the 2010 Nehalem-based quads use a different chipset? Different mobo? They're still Nehalems but they have fewer memory controllers than the Nehalems in the 2009 version? Or does the Westmere use two channels while the 2010 Nehalems still use 3?

I would bet my house that the Micron engineer is mistaken. I believe Apple is using the same motherboard with the same or nearly the same chipset as last year, and that the Westmere has the 3 memory channels that Intel most unambiguously says it does, just like the Nehalem. I think the Westmere goes into the same socket on the same board with the same chipset as the Nehalems (it certainly does in my Linux server), and that all of them talk directly to the main memory subsystem with their 3 memory controllers independently of the chipset anyway.

Unless there is a technical document from Apple or Intel confirming what the Micron employee said, I see no reason to believe it. It contradicts all the other available evidence.

TD

[lssmit02]

According to Apple (http://www.apple.com/macpro/specs.html#overlay-compare) it's the W3680 for the six core Mac Pro.

[Umbongo]

To be honest we need to know exactly which model Intel motherboard this is based on and is it in fact the W3680 or the X5680. The resulting answer is quite different. The W3680 is a workstation processor that fits on theWX58EP Intel board and does have triple channel 3 slots plus on extra slot.

The X5680 fits on a server board, like the Intel 5520 and they use dual channel.

So which is it?

5520 is triple channel with two sockets each with their own memory.

[VirtualRain]

I spoke with the Micron engineer who personally tested the 4, 6 and 12 core 2010 mac pros on August 9th with regard to memory compatibility.
After a long conversation and some additional research he confirmed that in fact the six core Westmere Mac Pro is a dual channel memory system (integrated in the chipset)

Ok, that makes no sense at all. The Memory Controller for Nehalem/Westmere is on the CPU die, not on the chipset... so you've lost me here.

Also, most certainly, the memory controller on Westmere is the same as on Nehalem... that is... a tri-channel controller.

I suppose it might be possible, that the 2010 CPU daughter card is only using 128 bits of the 192 bit memory bus with two DIMM sockets wired to each of two 64 bit channels, but that would have required a completely unnecessary redesign of the CPU daughter card to provide down-graded performance. Very unlikely Apple did this.


In addition. He stated that he in fact tested 8GB sticks of 1333 ram in the six core system and it did recognize and run the 8GB sticks. In fact Crucial has these 8GB sticks (albeit very expensive) listed now on their site as compatible with the 2010 Mac Pros.

I certainly wish I had this info when I placed my memory order with Transintl. Nevertheless, it seems to indicate that the max amount of ram for the six core is looking much better indeed. By the way, the board is a derivative of the Intel 5520 for anyone who is interested.

"Unlike Core i7, however, mainstream desktop/mobile Westmere chips will not feature a QPI bus or a triple-channel memory controller. Rather, we'll revert to a dual-channel controller and have no QPI clock to worry about"

This is complete nonsense... where did this quote come from? EDIT: I see this came from the Techgage page which is in reference to Arrandale and Clarkdale.

Of course all Nehalem/Westmere CPU's have a QPI bus to communicate with the X58 chipset.

[xgman]

According to Apple (http://www.apple.com/macpro/specs.html#overlay-compare) it's the W3680 for the six core Mac Pro.

Should have merged these threads:

the W3680 is a workstation processor, not a server processor and would go into a Intel WX58EP motherboard. Now, I can't confirm that is the board the Apple is based on, but no server board is listed as compatible with the W3680. That means our original thoughts on the 3 channels are correct. I am going to test 8gb x 3 as soon as I can. Sorry for any confusion. Apple could have given us a memory config lesson on this but .. .

[deconstruct60]

Westmere is based on dual channel. Not sure what that is referring to above.
....
the Nehalem core supported three channels of DDR3 memory, but you could only acheive maximum speed (1333mhz) with a single DDR3 DIMM per channel. With Westmere, Intel has tweaked the memory controller for performance, so now you can still get to that 1333mhz number with two DIMMs installed per channel.

Self contradictory statements. In the first there are only two. Later Westmere is Nehalem with an update so that can run two banks of DIMMs at the higher speed. The latter is true. The former... have no idea where it comes from.

Running both bank 1 and 2 at 1333MHz doesn't necessarily make Westmere "dual channel". That would be one way of implementing it but likely kept the physical channel the same, just improvements on processing the signal quality. Each channel can support up to 3 DIMMs. Most designs conserve board space though and leave the 3rd off. (It additionally drops the memory transfer speed even more for both new and tweaked versions. )

[xgman]

There actually are dual channel memory based server boards and triple channel based workstation boards both taking different westmere processors and therein lies the confusion. Bottom line is Apple should have spelled all this out in detail. How else are we to decide how to configure the optimal size, number of sticks and speed of the ram? Anyway, sorry again for any confusion I may have caused.

[sirnh]

There actually are dual channel memory based server boards and triple channel based workstation boards both taking different westmere processors and therein lies the confusion. Bottom line is Apple should have spelled all this out in detail. How else are we to decide how to configure the optimal size, number of sticks and speed of the ram? Anyway, sorry again for any confusion I may have caused.

The memory controller is in the the processor package. I think a board vendor would have to go out of their way to limit a Westmere to just dual channel, mainly by providing just 2 slots. The controller will default to triple channel if 3 modules are installed, and dual channel if 2 or 4 modules are installed.

[xgman]

The memory controller is in the the processor package. I think a board vendor would have to go out of their way to limit a Westmere to just dual channel, mainly by providing just 2 slots. The controller will default to triple channel if 3 modules are installed, and dual channel if 2 or 4 modules are installed.


I think they are using a board like this:

http://www.intel.com/products/workstation/motherboards/wx58bp/wx58bp-overview.htm

[trankdart]

I think they are using a board like this:

http://www.intel.com/products/workstation/motherboards/wx58bp/wx58bp-overview.htm

The technical specs on that board say "four sockets in three channels".

http://www.intel.com/Products/Workstation/Motherboards/WX58BP/WX58BP-specifications.htm

[glassbathroom]

Now I am really confused. I am guessing we are not going to finally resolve this 100% until someone has a new MP and has tested it. Is this correct?

[trankdart]

I agree there remain unanswered questions, and I don't think anybody here is going to do the test that resolves them. And I don't think a Kingston or Crucial employee who gets happier the more memory sticks get installed is a reliable source for answers (except maybe whether 8GB sticks work at all).

There are some questions only Apple can answer. Intel can't even answer them. Example: on a DP machine, how does the Nehalem NUMA architecture play into the whole memory channel latency story? 2 channels, 3 channels or no channels, how much latency is added if one CPU has to go over the interprocessor QPI link to get data in the memory attached to the other one? That depends to some extent on how (and whether) the operating system deals with the tricky fact of NUMA.

I think it's going to take anandtech or somebody like that to come up with some meaningful numbers. It would be nice if Intel would just answer, completely and in plain language, the simpler questions, like PRECISELY what happens when you populate that 4th slot?

[dhro1]

I I don't think a Kingston or Crucial employee who gets happier the more memory sticks get installed is a reliable source for answers

I fully agree with this. I enquired at crucial UK about the possibility of a compatible 3x4GB set for the six core and the response was:

Thanks for your e-mail.

This system can support up to 16GB over the four slots (4GB per slot) and the system uses Dual Channel functionality so we would recommend installing matching pairs.

Please also bear in mind that a 32-bit Operating System will only see and utilise around 3GB when more is installed.

You can order the compatible 4GB memory modules at the following link:

http://www.crucial.com/uk/store/partspecs.aspx?imodule=CT51264BA1339

Or you can order the 2GB memory modules here:

http://www.crucial.com/uk/store/partspecs.aspx?imodule=CT25664BA1067

Or you can order the 12GB kit 3x4GB modules here:

http://www.crucial.com/uk/store/partspecs.aspx?IMODULE=CT3KIT51264BA1339

Please let us know if we can offer additional assistance or advice.

This directly contradicts their claim of a maximum of 32GB and the 8GB sticks they sell in pairs only on their compatibility page. While the memory they recommend is non-ECC and I'm sure OS X SL is predominantly 64 bit and has no problems addressing more than 3GB.

[xgman]

I spoke with a high level apple rep last night and they agreed that apple should have posted a full description. Of the memory channel setup for each version of the new Mac pro. They failed to understand the "pro" computer needs "pro" explanations critical to how buyers will configure their orders for optimal performance. Hopefully they will update the spec chart. Sorry for the bad info previously posted on this. It has been an eye opening lesson. Hopefully I will be able to test some 8gb modules on this next week, but there is no guaranty they will work.

[mattmower]

I spoke with a high level apple rep last night and they agreed that apple should have posted a full description. Of the memory channel setup for each version of the new Mac pro. They failed to understand the "pro" computer needs "pro" explanations critical to how buyers will configure their orders for optimal performance. Hopefully they will update the spec chart. Sorry for the bad info previously posted on this. It has been an eye opening lesson. Hopefully I will be able to test some 8gb modules on this next week, but there is no guaranty they will work.

A small amount of extra information, I emailed Crucial in the UK about their 8GB part and whether they had tested on pre-release units and they responded:

>Thanks for your email. I can confirm that the 2010 6 Core Mac Pro will
>support 8GB per slot for a 32GB capacity. We have indeed tested the 8GB
>modules to work on pre release units of this model and we guarantee them
>to be compatible.

Matt

[VirtualRain]

A small amount of extra information, I emailed Crucial in the UK about their 8GB part and whether they had tested on pre-release units and they responded:

>Thanks for your email. I can confirm that the 2010 6 Core Mac Pro will
>support 8GB per slot for a 32GB capacity. We have indeed tested the 8GB
>modules to work on pre release units of this model and we guarantee them
>to be compatible.

Matt

That's good news... what would 3x8GB cost? :o

[fearoftigers]

I'm really confused.

I would love to put 3x8gb or 4x8gb sticks in.

I wonder when we will get a definitive answer.

[xgman]

We will know when one of us puts them in and tries it. 3 makes more sense than 4 because it would stay in the base triple channel mode per the design of the board. I hope to test some of these late next week.

[eponym]

3 x 8 GB I can believe working.

But I doubt 32 GB worked. The CPU's specs say its limit is 24. Maybe the system recognized 32 being installed, but couldn't actually use it all?

[mattmower]

3 x 8 GB I can believe working.

But I doubt 32 GB worked. The CPU's specs say its limit is 24. Maybe the system recognized 32 being installed, but couldn't actually use it all?

From a CPU perspective this makes no sense to me.

It's the same chip in the dual CPU model and the memory isn't partitioned with each CPU only seeing 24MB.

M.

[eponym]

It's not the same CPU in the dual.

The 3.33 6-core W3680 (http://ark.intel.com/Product.aspx?id=47917) can't be used in a dual config.

The specs themselves say:

Max Memory Size
(dependent on memory type): 24 GB

^ the vague part is "memory type". We don't know what that really means.

[mattmower]

It's not the same CPU in the dual.

The 3.33 6-core W3680 (http://ark.intel.com/Product.aspx?id=47917) can't be used in a dual config.

The specs themselves say:

Max Memory Size
(dependent on memory type): 24 GB

^ the vague part is "memory type". We don't know what that really means.

Ah yes, that was sloppy of me. The W3680 with 24GB vs the X5660 with its, rather enviable, 288GB memory limit! That's quite a steep drop off.

W

[xgman]

I'm praying for 3x8 working. 12GB won't cut it for long.

[Concorde Rules]

I think you lot are getting all over yourselfs for no reason at all.

The chipset is the same as the 2009 MP, all has changed are the CPUs slotted into it.

Im 99% confident that 3 slots vs 4 slots will still yield the best memory bandwidth.

We shall see when barefeats/other places get the 2010 MP.

[trankdart]

...

Im 99% confident that 3 slots vs 4 slots will still yield the best memory bandwidth.

I think there's no doubt about that. But in no way does that mean that everyone should conclude that 3 sticks are better than 4 in their own case.

Let's say that populating the 4th slot reduces the bandwidth by around 25% because it drops from 3-way to 2-way interleave when your main-memory subsystem is running flat-out. So?

How often is your memory subsystem running flat out, given the unique, new, and relatively enormous cache system of the Nehalem/Westmere architecture? How much is that drop reflected in your real overall system performance? How much is it offset in your particular system by having an extra 4GB of RAM? I assure you that for every synthetic benchmark you can come up with showing a slowdown of something that basically never happens in practice (your super-high-speed memory system getting saturated), Kingston and Crucial can come up with 5 more showing that the extra 4GB in the fourth slot improves your performance on XYZ applications even though it cuts down your interleave factor.

It is true that your maxed out memory scenario runs best with a multiple of 3. And IF it is also true that you're fine with 12GB instead of 16 with a single CPU, then maybe you should only populate 3, that's what I'm doing.

But, the more I read about this stuff (give this (http://whitepapers.opensystemsmedia.com/u/pdfs/WhitePaper.MaximizingMemoryResources.pdf) a long slow glance if you want some detail), the more I realize what the truth is:

For 99.98% of users the 3-sticks vs. 4 sticks controversy simply doesn't matter in real-life practice. It's a non-issue that's easy to mistake for an issue. 'Do what thou wilt' shall be the whole of the law. If you want 3, get 3. If you want 4, get 4. Ignore the many over-hyped benchmarks aimed at arousing the paranoia wired into our lizard brains. The difference is a few percent either way in real life situations on the memory access, and you'll be helped by more memory more often than you'll be hurt by less interleave.

Bottom line: we are indeed getting all over ourselves for no reason at all. The right thing to do: forget the whole issue.

EDIT: Nothing about the above mitigates the indisputable fact that it's kind of a shame to design a motherboard for these CPU's that has anything other than a multiple of 3 memory sockets.

[Concorde Rules]

I think there's no doubt about that. But in no way does that mean that everyone should conclude that 3 sticks are better than 4 in their own case.

Let's say that populating the 4th slot reduces the bandwidth by around 25% because it drops from 3-way to 2-way interleave when your main-memory subsystem is running flat-out. So?

How often is your memory subsystem running flat out, given the unique, new, and relatively enormous cache system of the Nehalem/Westmere architecture? How much is that drop reflected in your real overall system performance? How much is it offset in your particular system by having an extra 4GB of RAM? I assure you that for every synthetic benchmark you can come up with showing a slowdown of something that basically never happens in practice (your super-high-speed memory system getting saturated), Kingston and Crucial can come up with 5 more showing that the extra 4GB in the fourth slot improves your performance on XYZ applications even though it cuts down your interleave factor.

It is true that your maxed out memory scenario runs best with a multiple of 3. And if it is also true that you're fine with 12GB instead of 16 with a single CPU, then maybe you should only populate 3, that's what I'm doing.

But the truth is, the more I read about this stuff (give this (http://whitepapers.opensystemsmedia.com/u/pdfs/WhitePaper.MaximizingMemoryResources.pdf) a long slow glance if you want some detail), the more I settle on what the truth is:

For 99.98% of users the 3-sticks vs. 4 sticks controversy simply doesn't matter in real-life practice. It's a non-issue that's easy to mistake for an issue. 'Do what thou wilt' shall be the whole of the law. If you want 3, get 3. If you want 4, get 4. Ignore the many over-hyped benchmarks aimed at arousing the paranoia wired into our lizard brains. The difference is a few percent either way in real life situations on the memory access, and you'll be helped by more memory more often than you'll be hurt by less interleave.

Bottom line: we are indeed getting all over ourselves for no reason at all. The right thing to do: forget the whole issue.

Absolutely (I didn't include the real-life results of 4 vs 3 sticks)

I personally went for 6Gb however my 2011 Mac Pro (Sandy Bridge) will have 12Gb minimum :D

[skiffx]

I think there's no doubt about that. But in no way does that mean that everyone should conclude that 3 sticks are better than 4 in their own case.

Let's say that populating the 4th slot reduces the bandwidth by around 25% because it drops from 3-way to 2-way interleave when your main-memory subsystem is running flat-out. So?

How often is your memory subsystem running flat out, given the unique, new, and relatively enormous cache system of the Nehalem/Westmere architecture? How much is that drop reflected in your real overall system performance? How much is it offset in your particular system by having an extra 4GB of RAM? I assure you that for every synthetic benchmark you can come up with showing a slowdown of something that basically never happens in practice (your super-high-speed memory system getting saturated), Kingston and Crucial can come up with 5 more showing that the extra 4GB in the fourth slot improves your performance on XYZ applications even though it cuts down your interleave factor.

It is true that your maxed out memory scenario runs best with a multiple of 3. And IF it is also true that you're fine with 12GB instead of 16 with a single CPU, then maybe you should only populate 3, that's what I'm doing.

But, the more I read about this stuff (give this (http://whitepapers.opensystemsmedia.com/u/pdfs/WhitePaper.MaximizingMemoryResources.pdf) a long slow glance if you want some detail), the more I realize what the truth is:

For 99.98% of users the 3-sticks vs. 4 sticks controversy simply doesn't matter in real-life practice. It's a non-issue that's easy to mistake for an issue. 'Do what thou wilt' shall be the whole of the law. If you want 3, get 3. If you want 4, get 4. Ignore the many over-hyped benchmarks aimed at arousing the paranoia wired into our lizard brains. The difference is a few percent either way in real life situations on the memory access, and you'll be helped by more memory more often than you'll be hurt by less interleave.

Bottom line: we are indeed getting all over ourselves for no reason at all. The right thing to do: forget the whole issue.

EDIT: Nothing about the above mitigates the indisputable fact that it's kind of a shame to design a motherboard for these CPU's that has anything other than a multiple of 3 memory sockets.

Well put sir.

[VirtualRain]

I think there's no doubt about that. But in no way does that mean that everyone should conclude that 3 sticks are better than 4 in their own case.

Let's say that populating the 4th slot reduces the bandwidth by around 25% because it drops from 3-way to 2-way interleave when your main-memory subsystem is running flat-out. So?

How often is your memory subsystem running flat out, given the unique, new, and relatively enormous cache system of the Nehalem/Westmere architecture? How much is that drop reflected in your real overall system performance? How much is it offset in your particular system by having an extra 4GB of RAM? I assure you that for every synthetic benchmark you can come up with showing a slowdown of something that basically never happens in practice (your super-high-speed memory system getting saturated), Kingston and Crucial can come up with 5 more showing that the extra 4GB in the fourth slot improves your performance on XYZ applications even though it cuts down your interleave factor.

It is true that your maxed out memory scenario runs best with a multiple of 3. And IF it is also true that you're fine with 12GB instead of 16 with a single CPU, then maybe you should only populate 3, that's what I'm doing.

But, the more I read about this stuff (give this (http://whitepapers.opensystemsmedia.com/u/pdfs/WhitePaper.MaximizingMemoryResources.pdf) a long slow glance if you want some detail), the more I realize what the truth is:

For 99.98% of users the 3-sticks vs. 4 sticks controversy simply doesn't matter in real-life practice. It's a non-issue that's easy to mistake for an issue. 'Do what thou wilt' shall be the whole of the law. If you want 3, get 3. If you want 4, get 4. Ignore the many over-hyped benchmarks aimed at arousing the paranoia wired into our lizard brains. The difference is a few percent either way in real life situations on the memory access, and you'll be helped by more memory more often than you'll be hurt by less interleave.

Bottom line: we are indeed getting all over ourselves for no reason at all. The right thing to do: forget the whole issue.

EDIT: Nothing about the above mitigates the indisputable fact that it's kind of a shame to design a motherboard for these CPU's that has anything other than a multiple of 3 memory sockets.

Exactly!... put simply: Configure your memory for quantity first and speed second... Having enough memory is more important than faster memory. :)

[barefeats]

I agree with "trankdart." This is what I learned on the 2009 Nehalem in my testing as recent as yesterday:

1. Though DLT stress test showed the 8-core Nehalem with 6 memory modules having much higher bandwidth than with 8 memory modules, when I ran the bandwidth test so that you only use the first 12G of the 16G config, the bandwidth is the same for both memory configs.

2. Real world apps (currently) don't saturate the memory bandwidth so it does not make a difference in the real world. I can back that up with test data showing the Nehalem with 16GB of RAM equaling or beating the time it took the 12GB config to complete After Effects CS5 renders, Compressor 3.5 renders, Photoshop CS5 action files, etc.

3. In the AE CS5 render, 14G of the 16GB of memory was in use -- which brings me to my third point. It's more often better to have sufficient memory than to have the perfect theoretical memory config.

[barefeats]

Assuming the 6-core Westmere handles 8G modules from Crucial (or whomever), with the 24G limit and dual channel design, seems to me the best way to achieve 24G is to install 2x8G + 2x4G = 24G.

[xgman]

Assuming the 6-core Westmere handles 8G modules from Crucial (or whomever), with the 24G limit and dual channel design, seems to me the best way to achieve 24G is to install 2x8G + 2x4G = 24G.

They recommend using the exact same sizes and throughout.

[Concorde Rules]

They recommend using the exact same sizes and throughout.

I've just thought what about this setup:

Channel 1 - 2x 4Gb
Channel 2 - 8Gb
Channel 3 - 8Gb

Then the addressable space is the same for each channel and in theory could possibly allow three channel operation at max speed?

Although I suspect the lack of ability to address either of the 4Gb sticks at the same time might block that idea. :(

[eponym]

I doesn't work that way. And you can't mix 8GB modules with other sizes.

If you go with 8GB DIMMs, you can only use 8GB DIMMs.

[xgman]

Very interesting board layout from a 3.33 6 core box just received by one of our members regarding the 4 slots:

"Yes -- they are all black "
"The originally installed 3 x 1 GB chips were installed in the forward 3 slots (towards the open side) so it is a good guess that those are the tri-channel slots."

I would have expected a triple channel design like the Intel WX58BP to have one a different color. probably means nothing, but still odd.

http://forums.macrumors.com/showthread.php?p=10875147&posted=1#post10875147

http://att.macrumors.com/attachment.php?attachmentid=247203&d=1282232023

also read page 49 of the manual":
http://manuals.info.apple.com/en_US/Mac_Pro_Mid2010_UG_5686.pdf

and Intel spec:
http://www.intel.com/support/motherb...965.htm#triple

and see page 35 of this MB manual which is the only Intel retail board to support this cpu:
http://download.intel.com/support/motherboards/server/wx58bp/sb/e69058001_wx58bp_productguide_001.pdf

[xgman]

The 2010 manual says if you don't populate the slots to provide optimum performance, a "Memory Slot Utility" appears on the screen and recommends a better configuration.

[Icaras]

They recommend using the exact same sizes and throughout.

Who recommends that? I checked the manual you linked to, and I don't seem to read anywhere that equal sized DIMMs are recommended?

I doesn't work that way. And you can't mix 8GB modules with other sizes.

If you go with 8GB DIMMs, you can only use 8GB DIMMs.

According to the manual, page 49, it's possible.

[Concorde Rules]

I doesn't work that way. And you can't mix 8GB modules with other sizes.

If you go with 8GB DIMMs, you can only use 8GB DIMMs.

Could two 4Gb sticks and two 2Gb sticks work in tri-channel still?

Want to add a few more Gb to my Mac Pro 2009.