Is Tri Channel functionality maintained when 4th memory stick used?

[fruitatcost]

Greetings all. First time poster. I have been pouring over these threads for the past few days and am having a hard time interpreting some of the posts on this subject.

I have just purchased a Mac Pro Quad Core Nehalem. I purchased an 8gb kit from OWC.

The question: It is unclear to me if installing all 8gb will neutralize the tri channel memory functionality and therefore revert all memory into dual channel. I know this has been covered before and there are even diagrams posted in this forum but, to be honest, it is still not clear to me. Can someone humor me and spell this out for me?

The background is that I will be using this box for audio production (pro tools and Logic). My goal is to be able to use as many plugins as possible in order to mix-in-the-box. So, if I can still have tri channel with the extra 2gb stick in the fourth slot without compromising tri channel, that's what I will do. However, if it "neturalizes" tri channel and bumps everything down to dual channel, can I still obtain my goal or should I just leave the 4th ram slot free?

Thanks and apologies for any redundancy this thread may have caused

 

[Dr.Pants]

Unfortunatly not, as far as I am aware four DIMMs will trigger dual-channel mode.

 

[fruitatcost]

Unfortunatly not, as far as I am aware four DIMMs will trigger dual-channel mode.

Thanks for the input Doc

 

[UltraNEO*]

have a read of this..

"3 Things You Should Know About Nehalem Mac Pro Memory"

http://www.barefeats.com/nehal02.html

 

[fruitatcost]

have a read of this..

"3 Things You Should Know About Nehalem Mac Pro Memory"

http://www.barefeats.com/nehal02.html

Thanks for the link Ultra. Interesting that the article states that memory throughput is 72% less on a quad versus an 8 core. The article also touches on the less-is-more for the 8 core but doesnt actually come out and say whether the same is true for the quad. But, I will just assume that it is and just plug in the 6gb on the 3 slots and leave the last one empty. Thanks again.

 

[Loa]

Hello,

I sent an email to Rob-ART from Barefeats with this exact question (for a quad). His answer is that, theoretically, if apps could completely saturate the RAM bandwidth, 6GB (3*2GB) would be faster than 8GB (4*2GB).

But going down to dual-channel memory is more than compensated by going up 2GB for apps like PS. For music editing I don't know: do you use the HD (as scratch?) often like PS does?

Essentially, if your app has to go to virtual memory (on the HD), you're better off with 8GB.

But: nothing beats testing it out yourself!

Loa

 

[VirtualRain]

The question: It is unclear to me if installing all 8gb will neutralize the tri channel memory functionality and therefore revert all memory into dual channel. I know this has been covered before and there are even diagrams posted in this forum but, to be honest, it is still not clear to me. Can someone humor me and spell this out for me?

There are two possibilities, and unfortunately, I don't think anyone's been able to verify this one way or the other and the Intel documentation is also vague...

1. As the good Dr. suggests, the memory controller will default to dual-channel mode with 4 DIMMS
2. It's possible, but perhaps unlikely, that the memory controller will interleave data across the first three DIMMS and simply access the fourth DIMM in single channel mode.

Neither is ideal, but I also agree that dual-channel vs. tri-channel is less important a consideration than having enough memory.

 

[fruitatcost]

Thanks for the input Loa and V.Rain. All of the vagueness on this topic only confirms that I wasnt really that empty headed

I will conduct tests with both however I suspect that the improvements and/or degradation will be negligbile and perhaps immeasurable to the naked eye. Still, it would be good know

Thanks all for your input. Much appreciated.

 

[Loa]

I will conduct tests with both however I suspect that the improvements and/or degradation will be negligbile and perhaps immeasurable to the naked eye. Still, it would be good know

I'll do the same as I'm planning to order four 2GB sticks from OWC in the next few weeks.

I'll report the finding here.

Loa

 

[Umbongo]

There are two possibilities, and unfortunately, I don't think anyone's been able to verify this one way or the other and the Intel documentation is also vague...

1. As the good Dr. suggests, the memory controller will default to dual-channel mode with 4 DIMMS
2. It's possible, but perhaps unlikely, that the memory controller will interleave data across the first three DIMMS and simply access the fourth DIMM in single channel mode.

Neither is ideal, but I also agree that dual-channel vs. tri-channel is less important a consideration than having enough memory.

Not only is it possible, it's what happens. The connections don't physically change because you insert another module. The slower speeds you see from benchmarks when using 4 or 8 DIMMs rather than 3 or 6 is because there is no management to only use the extra DIMMs only when the others are fully utilized.

I'd reccomend anyone unsure on memory usage to try with 3 or 6 modules and see how it goes, then buy an extra 1 or 2 if you are finding yourself often out of memory.

 

[Tesselator]

Hello,

I sent an email to Rob-ART from Barefeats with this exact question (for a quad). His answer is that, theoretically, if apps could completely saturate the RAM bandwidth, 6GB (3*2GB) would be faster than 8GB (4*2GB).

But going down to dual-channel memory is more than compensated by going up 2GB for apps like PS. For music editing I don't know: do you use the HD (as scratch?) often like PS does?

Essentially, if your app has to go to virtual memory (on the HD), you're better off with 8GB.

I disagree with this but my argument is much stronger at 12 vs. 16 GB instead of 6 vs. 8. The two are different. VERY different! And really can't be compared or offered as substitute alternatives. RAM Amount adds performance only in high memory usage situations - which are rare if you tune your usage to your machine spec. as we all should be doing anyway. Triple-channel accesses are faster in ALL situations and if the benefits are as significant as dual over single channel access is on the 2006 ~ 2008 then it's NOT something you want to do without. Suggesting that 8GB is better and will make up the difference is wrong and uninformed. It won't!

But: nothing beats testing it out yourself!

Loa

Yupperz!

 

[Loa]

Suggesting that 8GB is better and will make up the difference is wrong and uninformed. It won't!

I'm eagerly awaiting any test that will show me that 6GB + scratch will be faster than 8GB without scratch, but sincerely I doubt it will happen. Unless you're using a fast-writing SSD or big RAID0. Even then I'd be curious.

Diglloyd also made tests with quads and octo Nehalem, and his conclusion is the same: more RAM > using triple-channel.

(http://macperformanceguide.com/Reviews-MacProNehalem-Tests-Memory.html)

His webpage is, sadly, very low on detailed graphs.

Can't wait to do the tests on my own.

Loa

 

[Umbongo]

I'm eagerly awaiting any test that will show me that 6GB + scratch will be faster than 8GB without scratch, but sincerely I doubt it will happen. Unless you're using a fast-writing SSD or big RAID0. Even then I'd be curious.

Diglloyd also made tests with quads and octo Nehalem, and his conclusion is the same: more RAM > using triple-channel.

(http://macperformanceguide.com/Reviews-MacProNehalem-Tests-Memory.html)

His webpage is, sadly, very low on detailed graphs.

Can't wait to do the tests on my own.

Loa

I think the confusion comes from there being two situations under discussion. If you need more than 6GB of memory then 8GB is way better than any scratch disk, single channel access is still 8.5GB/s. If you need fast memory performance and aren't using more than 6GB then having 8GB installed will offer slower performance. Though likely not noticable to the user.

 

[Loa]

I think the confusion comes from there being two situations under discussion. If you need more than 6GB of memory then 8GB is way better than any scratch disk, single channel access is still 8.5GB/s. If you need fast memory performance and aren't using more than 6GB then having 8GB installed will offer slower performance. Though likely not noticable to the user.

That's my position, but Tess seems to disagree.

Waiting on his reasons.

Loa

 

[Tesselator]

No, that's the same thing I'm saying and also what DL said at the link you posted:

"A garden hose delivers a trickle compared to a fire hose. It’s the same idea for memory: you want a “fire hose” delivering data to CPUs that need it."​

More RAM in this metaphor would be what, a lake?

I'm just contending two distinct differences. One is that the speed difference is indeed noticeable at least under my mouse and the benchmarks back it up, and two, that we even as professional hard-core users, spend far far more time using our machines under conditions where a cache is not utilized at all. At least not the kind where the difference between 6 and 8 GB is going to make any difference. This is absurdly true if we're comparing 12GB to 16GB instead of 6 to 8 where it's only mostly true.

So the user question for me becomes: Which would I rather have, an 8% ~ 10% increase in snappy GUI interaction ALL THE TIME, or an 8% ~ 10% speed increase in operations that way more than likely take up less than 5% of my day?

I think DL said it right:

"Whenever it’s a case of disk access versus memory, more memory is always as good or better (see Optimizing Photoshop). This applies not just to Photoshop, but to the ability of the system to use unused memory for caching, which speeds up all programs.

Bottom line: forget about 8 modules vs 6 and get what you need eg 16GB vs 12GB."​

I have 12GB right now in my system. It's not a 2009 so it's in 8 DIMMs and no triple channel is possible, but with 12GB I rarely see any applications grinding away at their cache. Also page-outs are displayed in my menu-bar, and I almost never see any at all. I just upgraded a few months ago from 4GB and even with low amount the times I ever even noticed cache usage were very few and very far between.

Anyway, that's my logic and reasons on it.

 

[Dr.Pants]

I apologize if I am mistaken on the dual-channel configuration

 

[Loa]

an 8% ~ 10% increase in snappy GUI interaction

Snappy is good, but it's also hard to measure.

Could this be a solution to measure it: we set-up an AppleScript to do a long series of usual actions (across many apps) and see how long it takes to complete it.

Would that work?

Loa

P.S. My (8GB) RAM is on its way now!

 

[Tesselator]

Snappy is good, but it's also hard to measure.

Could this be a solution to measure it: we set-up an AppleScript to do a long series of usual actions (across many apps) and see how long it takes to complete it.

Would that work?

Loa

P.S. My (8GB) RAM is on its way now!

Maybe. Sounds good to me.


I notice it in icon refreshes with pages already in the RAM based system cache and the speed at which 3D models initialize just shortly after disk loading. When working on wireframe models (pushing & pulling points ) the difference becomes noticeable too. For fullscreen refreshes of detailed CAD models of around a million or two vertices it's also apparent. A 3D card accelerates cam, world, and object rotations (among other things) but in an editor the data has to be updated in local RAM as well. XBench and Geekbench poorly written tho they are, may be able to measure the differences too (?).

 

[VirtualRain]

Here's some evidence that the Nehalem memory controller defaults to dual-channel mode with 4 dimms populated (this is using Intel's desktop reference board which has 4 DIMM slots just like Apple's 2009 quad).

Also of note is the fact that adding in that 4th DIMM also hits the memory sub-system performance really dramatically. When we have 8GB of DDR3-1333 installed the system in fact runs at nearly the same speed as the dual-channel 4GB configuration. Memory in sets of three seems to be the requirement for optimal Core i7 systems.

Link

 

[Umbongo]

Here's some evidence that the Nehalem memory controller defaults to dual-channel mode with 4 dimms populated (this is using Intel's desktop reference board which has 4 DIMM slots just like Apple's 2009 quad).

It doesn't default to dual-channel mode. It isn't physically wired like that. It's just accessing at a mix of triple channel and single channel bringing the overall speed down.

 

[netkas]

cpu can't just switch memory stick from one controller to another

run cpu-z in windows installed on mac to get real info about channels used with 4 dimms setup, rest is suggestions

 

[Tesselator]

---

 

[VirtualRain]

It doesn't default to dual-channel mode. It isn't physically wired like that. It's just accessing at a mix of triple channel and single channel bringing the overall speed down.

I agree that physically, there are three data busses connecting the three channels of RAM to the memory controller. That does not preclude the memory controller from interleaving data across channels differently depending on how the DIMMS are populated. It could very easily treat the 2 DIMMS on channel 0 as one channel (obviously) and the other two DIMMS as one channel.

I think if it was operating as you suggest, the average bandwidth would be around 19GB/s (3GBx23GB/s and 1GBx9GB/s) not identical to dual-channel performance as their test results show.

At any rate, I would stick to three DIMMS to maximize memory performance. If you can't make do with 6GB then go to 12GB.

 

[Umbongo]

I agree that physically, there are three data busses connecting the three channels of RAM to the memory controller. That does not preclude the memory controller from interleaving data across channels differently depending on how the DIMMS are populated. It could very easily treat the 2 DIMMS on channel 0 as one channel (obviously) and the other two DIMMS as one channel.

The memory controller doesn't work like that. It wouldn't make sense for it to either, you'd get poorer performance overall.

 

[Loa]

At any rate, I would stick to three DIMMS to maximize memory performance. If you can't make do with 6GB then go to 12GB.

I'd gladly go from 6GB to 12GB, but the price (at OWC) on a quad goes from 118$ to 540$... For now I'll take the theoretical memory speed hit with 8GB, and prevent more VM usage...

Also, OWC results are interesting... Unless I'm misreading them, they actually show that for 2 apps, the theoretical "hit" in RAM speed by going from 6GB to 8GB on a quad, and 3*4GB to 4*4GB on an octo, is inexistent in real applications. And in After Effects, that 12 -> jump actually increases performance...

Loa

P.S. Tesselator: can you tell me that name of the app/plug-in that allows you to see the page in-out in your menu bar?