Mac Pro 3,1 - 16GB 667mhz faster than 20GB 800mhz RAM??

[lewis_cooper]

Here's a conundrum for you all…

I have a 2008 Mac Pro 3,1 2 x 2.8 GHz running El Capitan 10.11.5.

I had 16GB of RAM made up of a mixture of 800mhz and 667mhz modules.The presence of the 667mhz causes even the 800mhz modules to run at 667mhz so I decided to purchase 4 x 4GB modules so that I could do away with the 667mhz modules and increase my total memory at the same time.

I ran Geekbench tests along the way and I had surprising results…

First up is the pre-upgrade test. So 16GB total memory made up of 8 x 2GB modules, and a mixture of 667mhz and 800mhz.

Next, we have my 800mhz modules only total of 20GB total, made up of 4 x 4GB and 2 x 2GB. Oddly, more RAM and purely 800mhz yielded a lower score.

So, after coming to the conclusion that 667mhz wasn't so bad after all I filled my last remaining slots with 2 x 2GB 667mhz sticks to take me up to 24GB running but running at 667mhz. More RAM but lower than 16GB running at 667mhz. Go figure.

Does anyone have the foggiest idea why this could be??

 

[ActionableMango]

Does anyone have the foggiest idea why this could be??

I believe you have to do an SMC or PRAM reset to update the ram capacity and speed. Many people have complained about changing the RAM but the speed was stuck.

Did you do the resets and then verify the actual speed and capacity the RAM was running at? If not, this may explain the results. Speed and capacity can be seen in "About this Mac":

 

[666sheep]

Nothing surprising. This is the impact of 256 bit data path when risers are populated with 4 or 8 equal capacity modules.
As you can see, it gives more than 133 MHz bus difference can give.

 

[lewis_cooper]

I believe you have to do an SMC or PRAM reset to update the ram capacity and speed. Many people have complained about changing the RAM but the speed was stuck.

Did you do the resets and then verify the actual speed and capacity the RAM was running at? If not, this may explain the results. Speed and capacity can be seen in "About this Mac":

I didn't perform any resets but I don't think my system had any problem recognizing each new configuration. At every step I opened up System Profiler in order to check the modules were recognised and at which frequency they were all running at.

Is it possible that the System Profiler would display the correct info but that the system would continue to run at 667mhz without the reset?
[doublepost=1464166543][/doublepost]

Nothing surprising. This is the impact of 256 bit data path when risers are populated with 4 or 8 equal capacity modules.
As you can see, it gives more than 133 MHz bus difference can give.

Interesting, but I'm not sure I understand (my shortcoming not yours I suspect). Could you please elaborate on the 256 bit data path part?

So are you saying that 16GB RAM at 667mhz should score higher than 20GB at 800mhz (and 24GB at 667mhz)?

 

[nigelbb]

Nothing surprising. This is the impact of 256 bit data path when risers are populated with 4 or 8 equal capacity modules.
As you can see, it gives more than 133 MHz bus difference can give.

This. You need matching size RAM modules e.g. 8 x 2GB or 8 x 4GB but 667MHz vs 800 MHz makes very little difference (only 4% according to BareFeats benchmarks). The 667MHz parts are so cheap as used pulls from Dell & HP Xeon servers that it makes sense to put 32GB (8 x 4GB) in every 2008 Mac Pro 3,1.

 

[lewis_cooper]

This. You need matching size RAM modules e.g. 8 x 2GB or 8 x 4GB but 667MHz vs 800 MHz makes very little difference (only 4% according to BareFeats benchmarks). The 667MHz parts are so cheap as used pulls from Dell & HP Xeon servers that it makes sense to put 32GB (8 x 4GB) in every 2008 Mac Pro 3,1.

Okay, I think I'm starting to understand.

Basically… I wasted my money purchasing 4 new 4GB 800 MHz modules. And should have instead aimed for 8 x 4GB 667 MHz so I could fill all the slots with matching memory?

Perhaps I won't be able to return the new sticks. So should I buy 4 more but save money by going 667 MHz and run 32GB made up of 800s and 667s but all 4GB?

What I still don't understand is; was I really better off with 8GB less RAM just because they were all 2GBs?

 

[h9826790]

Okay, I think I'm starting to understand.

Basically… I wasted my money purchasing 4 new 4GB 800 MHz modules. And should have instead aimed for 8 x 4GB 667 MHz so I could fill all the slots with matching memory?

Perhaps I won't be able to return the new sticks. So should I buy 4 more but save money by going 667 MHz and run 32GB made up of 800s and 667s but all 4GB?

What I still don't understand is; was I really better off with 8GB less RAM just because they were all 2GBs?

If 24G is enough for you, I can't see why you have to spend money on it (apart for fun).

Only one simple question. Can you tell the difference without the Benchmark software? If NO, then why bother? (Unless money is not an issue for you. In this case, buy what even you want. And the highest speed RAM stick with all same size and spec should give you the best performance)

Of course, 32G RAM is better then 24G. I personally will go for it. And if you won't care about what system profile said, 677MHz should have better cost to performance ratio. However, if you care about that, may be you should pay more to keep everything at 800MHz. That's the way to keep yourself happy, and happiness is priceless.

 

[666sheep]

Could you please elaborate on the 256 bit data path part?

It's all about memory throughput. Optimal risers population gives maximum possible throughput.
From old barefeats article, but you'll get the idea:

DUAL CHANNEL vs QUAD CHANNEL
In order to get the full benefit of the Mac Pro's 256 bit memory data path, you'll want to populate both memory riser cards, each with at least one matched pair. If you put your memory on only one riser, you are dropping from quad channel to dual channel mode. See Apple's Mac Pro memory notes for more this.


Does this translate into faster real world speed? Not always. Though the Xbench memory fill rate test showed a 34% gain, it doesn't necessarily translate to faster application speeds. We ran some typical tests from our suite of real world tests (iMovie render effect, Final Cut Pro render clip, Cinebench CPU render, Motion render RAM preview, iMaginator Core Image morph). None of them showed any gains from Quad Channel mode.

Anandtech did some testing with real world apps using dual and quad channel memory configurations, too. Of their 15 real world tests, only 2 showed any gains from Quad channel configuration.

It's always better to have a little too much a bit "slower" RAM than not enough of "faster" one.

 

[ActionableMango]

Is it possible that the System Profiler would display the correct info but that the system would continue to run at 667mhz without the reset?

Not as far as I know.

 

[nigelbb]

Okay, I think I'm starting to understand.

Basically… I wasted my money purchasing 4 new 4GB 800 MHz modules. And should have instead aimed for 8 x 4GB 667 MHz so I could fill all the slots with matching memory?

Perhaps I won't be able to return the new sticks. So should I buy 4 more but save money by going 667 MHz and run 32GB made up of 800s and 667s but all 4GB?

What I still don't understand is; was I really better off with 8GB less RAM just because they were all 2GBs?

OS X will use any spare RAM for file caching so more RAM is always good. You can buy 32GB (8x4GB) of used 667MHz FB-DIMMs is now only about $40 to $50. I'm guessing the 4 x 4GB 800MHz parts cost considerably more. If you can't get a refund you can at least recoup much of the cost by selling them on eBay. Actually I see from a quick search of eBay that 800MHz FB-DIMMs are not half as expensive as they used to be. Maybe sellers find that they are now overstocked as AFAIK the only machine to use these is the 2008 Mac Pro 3,1 & the numbers of these systems must be diminishing daily. If I were you I would just spend $80 for another 4x4GB 800MHz FB-DIMMs & be done with it. You may be able to get a little money for all the other units that you have. http://www.ebay.com/itm/Apple-Mac-P...291985?hash=item4d17e3e0d1:g:7wwAAOSwxp9W6GXh

BTW For that same $80 you can purchase 4x8GB 667MHz FB-DIMMs at cheap prices but due to some quirk in the Mac Pro 3,1 if you install more than 56GB the performance of any PCI SSDs is halved.

 

[lewis_cooper]

OS X will use any spare RAM for file caching so more RAM is always good. You can buy 32GB (8x4GB) of used 667MHz FB-DIMMs is now only about $40 to $50. I'm guessing the 4 x 4GB 800MHz parts cost considerably more. If you can't get a refund you can at least recoup much of the cost by selling them on eBay. Actually I see from a quick search of eBay that 800MHz FB-DIMMs are not half as expensive as they used to be. Maybe sellers find that they are now overstocked as AFAIK the only machine to use these is the 2008 Mac Pro 3,1 & the numbers of these systems must be diminishing daily. If I were you I would just spend $80 for another 4x4GB 800MHz FB-DIMMs & be done with it. You may be able to get a little money for all the other units that you have. http://www.ebay.com/itm/Apple-Mac-P...291985?hash=item4d17e3e0d1:g:7wwAAOSwxp9W6GXh

Thanks nigelbb. Yes, I figured as much. Shame I learned all this after having ordered the 4x 800MHz DIMMs. But saying that, at least I only paid $97.87 (£66.99) for them.

BTW For that same $80 you can purchase 4x8GB 667MHz FB-DIMMs at cheap prices but due to some quirk in the Mac Pro 3,1 if you install more than 56GB the performance of any PCI SSDs is halved.

Special thanks for this note. That's something I really am pleased you pointed out and probably wouldn't have realised otherwise.

I'm not sure I have a need to take my 3,1 any higher than the 24GB RAM mark anyway. So far I've not come anywhere close to utilizing what I have. The memory pressure stays well below 30% so seems there is lots of room to breathe.

The only thing holding me back now is my choice of a non mac GTX 960 4GB graphics card… but I'll save that for my other thread!

Thanks again,