I have the 2.66 Quad configuration and I want to upgrade to 6GB of RAM. Is it ok to just order 2x2GB modules from OWC and then use two 1GB sticks to fill out the remaining two slots? I mean I know it will work, but is it really detrimental to not have it all paired?
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.
MacPro Memory Question
- Thread starter jfriedman8
- Start date
- Sort by reaction score
You will use the 2x2gb on the top and the 2x1gb on the bottom.
It will work if he configures it correctly.
(note: image for mem config good on early 2008 models. I have no idea if it's good on any other models)
![MemConfig.jpg]( "MemConfig.jpg")
It will work if he configures it correctly.
(note: image for mem config good on early 2008 models. I have no idea if it's good on any other models)
Those are paired.
You mean all matching? No big deal there. What's in there now tho? Are you sure it's 1GB sticks? They originally sold the MP 2006 models with four 512MB sticks.Why does it matter where they are seated? You can get technical, I'll probably be able to figure it out.
I have the 2009 version btw. It came with 3X1GB sticks.
I have the 2009 version btw. It came with 3X1GB sticks.
He said he had a 2.66 Quad and wanted to put in 2x2GB and the 2x1GB to 'fill out the remaining two slots'. That and the fact that his history shows he posted he has a 2009 2.66 Quad means he has 4 memory slots. Ignore the picture Jessica posted, that's for a previous generation Mac Pro with 8 FB-DIMM slots.
Yes, you can do it, but you will lose the triple channel feature, which means you may have slower memory access. If you want to maintain the triple channel, order 3x2GB DIMMS.
Yes, you can do it, but you will lose the triple channel feature, which means you may have slower memory access. If you want to maintain the triple channel, order 3x2GB DIMMS.
Ok thats what I was looking for. What kind of performance decrease am I looking at if I lose that feature.
Hi. I figured it would be best to post my questions in this thread since they're pretty similar and they probably don't warrant their own thread.
I just ordered my first Mac this weekend. Just coming out of college, I wasn't financially in a position to go crazy with the Mac Pro, so I ended up getting the default quad core Mac Pro, with the 3gb of RAM.
Before I ordered the Mac, when it was on sale on newegg I picked up two additional 2gb sticks for it. Now from what I understand from the thread, having the 2 x 2gb and the 2 x 1gb in all four slots will switch it from triple channel to dual channel.
So basically my question is, performance wise, until I can get another 2gb stick, would it be better to run the 3gb on triple channel or the 6gb on dual? I'm unfamiliar with the performance differences between dual and triple channel.
Thanks.
I just ordered my first Mac this weekend. Just coming out of college, I wasn't financially in a position to go crazy with the Mac Pro, so I ended up getting the default quad core Mac Pro, with the 3gb of RAM.
Before I ordered the Mac, when it was on sale on newegg I picked up two additional 2gb sticks for it. Now from what I understand from the thread, having the 2 x 2gb and the 2 x 1gb in all four slots will switch it from triple channel to dual channel.
So basically my question is, performance wise, until I can get another 2gb stick, would it be better to run the 3gb on triple channel or the 6gb on dual? I'm unfamiliar with the performance differences between dual and triple channel.
Thanks.
This will depend on what your specific needs are. By installing all of the memory you've on hand ATM, you'd see dual channel. But if you're only using the system for browsing for example, it's not going to matter. Capacity may be the greater issue, or throughput.
More details would be needed to give any further information.
If we just consider memory speed and execution without the "buffering" affects that come with more RAM then the difference between dual channel and triple are noticeable and the differences between single channel and triple channel are very noticeable. Each core has access to a 8MB L3 cache. Pretty much anything the processor does to/with data that data needs to be loaded into and usually back out of, that 8MB cache. This is level 3 so in a four core chip all 4 cores share this 8MB area. Generally speaking it's the speed between your RAM and this cache that is increased 1x, 2x, or 3x as a result of your memory configuration. If your processor "loads" 2 MB of data and then performs a trillion calculations on it before sending it on it's way then the triple speed was only realized twice during the operation. If that operation took 20 seconds to complete then the overall advantages of the 3x were fairly negligible.
If on the other hand, each core loads a 1MB chunk and preforms 3 operations on it before sending it on it's way where each operation is only nano-seconds in time well, then, the faster you can get data in and out of that L3 cache the better and 3x transmit speeds are going to be very noticeable indeed - yeah about 3x faster.
Unfortunately this doesn't affect the speed at which any other data pools or streams move. Only RAM to Processing unit - but that's the most crucial in most execution models. I would be interested to know how this affects OpenCL on a hardware level if at all but very generally speaking (sans lots of details and facts!) this is how RAM channel-plexing affects processing speed.
The "buffering" affect we neglected to consider earlier is very pronounced in low system-memory configurations. Meaning that the difference between 2GB and 6GB will generally be much greater than the differences between 16GB and 32GB - for the average user on a Mac Pro.
Where these two performance curves intersect is the subject of considerable discussion here at MR and on other sites. Bare feats attempted (and IMO miserably failed) to show the relationship between these two curves. Like nano-froggy says it all just depends on the usage profile (doing what with which apps - how often and in what combinations?). The best way is to cover all your bases at once by having as much triple channel RAM installed as possible. It's just too bad Apple was so damn cheap with the RAM slots! The 5500 series supports no less than 18 slots of DDR3 for a total of 144 GB of RAM! Apple gave us 4 (or 8 on the octad systems).
.
If on the other hand, each core loads a 1MB chunk and preforms 3 operations on it before sending it on it's way where each operation is only nano-seconds in time well, then, the faster you can get data in and out of that L3 cache the better and 3x transmit speeds are going to be very noticeable indeed - yeah about 3x faster.
Unfortunately this doesn't affect the speed at which any other data pools or streams move. Only RAM to Processing unit - but that's the most crucial in most execution models. I would be interested to know how this affects OpenCL on a hardware level if at all but very generally speaking (sans lots of details and facts!) this is how RAM channel-plexing affects processing speed.
The "buffering" affect we neglected to consider earlier is very pronounced in low system-memory configurations. Meaning that the difference between 2GB and 6GB will generally be much greater than the differences between 16GB and 32GB - for the average user on a Mac Pro.
Where these two performance curves intersect is the subject of considerable discussion here at MR and on other sites. Bare feats attempted (and IMO miserably failed) to show the relationship between these two curves. Like nano-froggy says it all just depends on the usage profile (doing what with which apps - how often and in what combinations?). The best way is to cover all your bases at once by having as much triple channel RAM installed as possible. It's just too bad Apple was so damn cheap with the RAM slots! The 5500 series supports no less than 18 slots of DDR3 for a total of 144 GB of RAM! Apple gave us 4 (or 8 on the octad systems).
.
You'll lose zero real-world performance.
Take a read in this thread if you want more info (https://forums.macrumors.com/threads/729603/) , but the botom line is this: in most (all?) real-world applications, you're going to be better off with more ram, regardless of the number of channels actually used.
That is, of course, if you need that RAM.
Don't be impressed by raw bandwidth graphs: apps cannot saturate the ram bandwidth in any case.
In the end, if your machine has to write out to cache (page out) or if apps like PS need to use the scratch disk, add more ram and don't even think about the number of channels used.
Two interesting reads:
http://macperformanceguide.com/Reviews-MacProNehalem-Tests-Memory.html
http://www.anandtech.com/memory/showdoc.aspx?i=3589&p=1
Loa
P.S. When you're reading about performance, be very careful to seperate the theoretical and the actual. Switching from triple to dual channel will indeed cause a theoretical slowdown in ram speed. If all you're interested in is the speed between the ram and the buss, then that slowdown will be very real. On the other hand, if you're interested in the performance of your apps (in other words the entire machine), then look for actual app speed tests.
What people in this thread are saying is very true: ram performance will decrease significantly by going from triple to dual channel. How will it affect your PS performance? That's an entirely different question.
I'm enough of a nerd to be interested in theoretical speed differences, but a lot more interested in actual, real-world application speeds!!!
Gee, I wonder why Intel spent millions on developing it then?
So when an "app" tells the CPU to "do something" to an image for example, and the processor executes that - you're saying that the processor doesn't do it at it's fastest? What? It says: "Oh, it's only PS, a mere app, I can lag around and take my time. I don't actually need to move the data at the speed of my clock or bus, I'm enjoying a nice cup of tea right now anyway."?
Only true if we're talking about very low quantities of RAM tho. Typically more than 6GB of RAM will not prevent PS from paging out. PS is 32 Bit and all it can see is 3 GB. Even if you have 64 GB PS will still page-out under the same conditions as it did in 6 GB.
The proper practice and application of the theoretical is the actual - in the terms we're dealing with here.
The answer to which depends on a a wide variety of variables! What tools or filters you're using in PS, the size of the image, the amount and speed of RAM, etc. etc. Generally yes, it is noticeable. And in a testbed environment (given testers who know what they're doing) where you're singling out this specific aspect the differences are dramatic.
.
Wow. Thanks for all the replies. As for my use with the Mac Pro, I mainly do a lot of HD video editing (mainly got the Mac for FCP) and animating with After Effects.
While before I was looking at that 3rd 2gb stick as a "maybe", after going through this thread I think I'll definitely be picking it up.
I'm curious though. If the memory performs better in sets of three what with triple channel and all, why does Apple only over 4 or 8 slots? Wouldn't it make more sense to offer 3, 6 or 9?
While before I was looking at that 3rd 2gb stick as a "maybe", after going through this thread I think I'll definitely be picking it up.
I'm curious though. If the memory performs better in sets of three what with triple channel and all, why does Apple only over 4 or 8 slots? Wouldn't it make more sense to offer 3, 6 or 9?
Hello,
Tess, I usually enjoy and learn a lot from your comments, but we have to keep raw machine-level bandwidth and actual application performance seperated. Otherwise we're befuddling ourselves.
http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3448&p=7
A single page read will give you the answer as to why it's not better than DDR2 right now. Especially with the "slow" DDR3 on our 2009MP.
---
Two series of real world tests should make my point painfully obvious.
1) Anandtech compared DDR3 ram with varying speeds (from our own 1066MHz in our 2009 MP all the way to a 1866MHz + 2000MHz overclocked on other tests).
![phcs4a.jpg]()
That's certainly not the same thing as going from triple channel to dual channel, but keep on reading.
(Note: rest of the article is here, with LOTS of similar tests with other apps: http://www.anandtech.com/memory/showdoc.aspx?i=3589&p=1)
2) Anandtech again, using Nehalem architechture: differences between DDR2 and DDR3. Full article here (http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3448&p=1)
Ineteresting bit here:
![anandtech.jpg]()
WOW! A whopping 0% difference between dual and triple channel in real world useage.
So according to Intel themselves, our 2009MP memory is less than HALF the speed it needs to be to take real advantage of DDR3 (as stated in the article I linked to reply to your "Gee").
It's a bit like putting a huge water pump behind a small hose: theoretically the power (DDR3 bandwidth) is there, but that small hose (1066MHz) just won't let it through. (If that's not a mixed metaphor, I don't know what is!)
So unless I've badly misread some of those articles, I think the question is settled: in our machines, more ram > faster ram.
Loa
Tess, I usually enjoy and learn a lot from your comments, but we have to keep raw machine-level bandwidth and actual application performance seperated. Otherwise we're befuddling ourselves.
http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3448&p=7
A single page read will give you the answer as to why it's not better than DDR2 right now. Especially with the "slow" DDR3 on our 2009MP.
---
Two series of real world tests should make my point painfully obvious.
1) Anandtech compared DDR3 ram with varying speeds (from our own 1066MHz in our 2009 MP all the way to a 1866MHz + 2000MHz overclocked on other tests).
That's certainly not the same thing as going from triple channel to dual channel, but keep on reading.
(Note: rest of the article is here, with LOTS of similar tests with other apps: http://www.anandtech.com/memory/showdoc.aspx?i=3589&p=1)
2) Anandtech again, using Nehalem architechture: differences between DDR2 and DDR3. Full article here (http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3448&p=1)
Ineteresting bit here:
WOW! A whopping 0% difference between dual and triple channel in real world useage.
So according to Intel themselves, our 2009MP memory is less than HALF the speed it needs to be to take real advantage of DDR3 (as stated in the article I linked to reply to your "Gee").
It's a bit like putting a huge water pump behind a small hose: theoretically the power (DDR3 bandwidth) is there, but that small hose (1066MHz) just won't let it through. (If that's not a mixed metaphor, I don't know what is!)
So unless I've badly misread some of those articles, I think the question is settled: in our machines, more ram > faster ram.
Loa
Point 1: That entire article is about ram speed not channel access modes.
Point 2: Those tests are being run in an ENTIRELY different environment. A windows PC running a single Core i7 isn't a very good test bench for dual processor OS X or even OS X in general.
So with point one, the article is completely off-topic anyway so why reference it? The graph you included is just silly (as far as WHAT they tested) for testing the difference between dual and triple channel modes. You can maybe look at the three top-most rows of stats and safely disregard the rest completely.
So what you're saying is that Intel's comment about there being no speed gain with DDR3 until you go over 2GHz doesn't apply to Macs? How lucky we are!!! We have magical DDR3 ram! Woohoo!
Also, give me some Mac OS X test results that you won't dismiss out of hand (like Diglloyd's).
Loa
Does Apple sell a DDR3 capable machine at under 2.0 GHz? But no matter about the facts what does that have to do with the difference between triple, dual, and single channel access modes which is the topic of THIS discussion?
All the ones I've seen on-line are pretty much done by retards! But soon you will have 8GB in a 2009 machine right? Get a stop watch and do a series of tests - repeat and average them 10x or more. Then setup tripple channel @ 6GB and repeat. Then you'll know. I've done this with dual vrs. single channel modes.
Do we have 2GHz ram chips to put in our MPs? Will (I really don't know the answer to this one) my MP adapt to the correct voltages required for DDR3's best results?
Also, it's another facet to the bigger point I'm trying to make: theoretically better pieces in a machine doesn't mean that the entire machine will work faster.
But soon you will have 8GB in a 2009 machine right? Get a stop watch and do a series of tests - repeat and average them 10x or more. Then setup tripple channel @ 6GB and repeat. Then you'll know. I've done this with dual vrs. single channel modes.
I have it now. Do you have some tests to recommend? I've used the typical RetouchArtists test in the past. Is this still current?
Loa
The 2008 Mac Pro's memory bandwidth is limited by FSB speeds.
The 2009 Mac Pro's memory bandwidth is significantly improved due to the on-die memory controller and tri-channel capability (as well as faster DDR3 speeds), but there are two mitigating factors that reduce the impact this has on every-day application performance:
1. Even 4 cores and 8 threads have difficulty consuming data at a rate of 25GB/s and so the CPU is now the bottleneck.
2. The enormous cache sizes employed on both Harpertown (12MB L2) and Nehalem (8MB L3) ensure it's extremely rare for an application to experience a cache miss, reducing the relevance of memory performance.
Over time (subsequent generations), we may see Intel shift more transistor count away from cache to core logic (more cores)... only then will the capabilities of tri-channel DDR3 be realized.
Edit: here's another great review that concludes the same... http://www.anandtech.com/memory/showdoc.aspx?i=3589
A 7-8% gain for a doubling in memory bandwidth underscores the points above.
The 2009 Mac Pro's memory bandwidth is significantly improved due to the on-die memory controller and tri-channel capability (as well as faster DDR3 speeds), but there are two mitigating factors that reduce the impact this has on every-day application performance:
1. Even 4 cores and 8 threads have difficulty consuming data at a rate of 25GB/s and so the CPU is now the bottleneck.
2. The enormous cache sizes employed on both Harpertown (12MB L2) and Nehalem (8MB L3) ensure it's extremely rare for an application to experience a cache miss, reducing the relevance of memory performance.
Over time (subsequent generations), we may see Intel shift more transistor count away from cache to core logic (more cores)... only then will the capabilities of tri-channel DDR3 be realized.
Edit: here's another great review that concludes the same... http://www.anandtech.com/memory/showdoc.aspx?i=3589
A 7-8% gain for a doubling in memory bandwidth underscores the points above.
"Reduce the impact" yes. Not eliminate it!
And I don't honestly think you can say that article underscores any of the important points in this particular discussion.
Every graph in that document is measuring the differences in RAM speed (clock) and CL rating (waits). Very little or no mention (certainly no intelligent mention!) of channel multiplexing/interleaving.
.
And I don't honestly think you can say that article underscores any of the important points in this particular discussion.
Every graph in that document is measuring the differences in RAM speed (clock) and CL rating (waits). Very little or no mention (certainly no intelligent mention!) of channel multiplexing/interleaving.
.
True, but the net effect of interleaving memory across multiple channels is an increase in memory bandwidth... which can also be achieved by increasing the module/bus speed, or reducing latency.
Using your own analogy, it's the same as improving storage performance... you can improve HD performance by RAIDing more disks, decreasing access times, or improving the transfer rate of individual disks (or ideally all three). The overall effect is improved STR in any case (and likely a proportionally small impact on overall system or application performance).
For memory, the fact is (due to the two points I mentioned above) increasing memory bandwidth beyond about 10-12GB/s does not materially impact application performance significantly... small amounts for very specialized apps... yes, but it's still more important to have enough memory, than faster memory... else you end up relying on orders of magnitude slower storage.
EDIT: Even more evidence to support the fact that memory performance plays a small role in overall application performance is the fact that many applications benchmark similarly on either 2008 (Harpertown) Mac Pro's or 2009 (Nehalem) Mac Pro's even though Nehalem has at least a 2-3 times better performing memory architecture.