Resolved Mac Pro memory upgrade

[wiski15b]

Hey guys,
Quick question. Does the Mac Pro 4,1 / 5,1 get better performance by using fewer/larger memory modules? I read somewhere that filling all the slots with smaller modules would not give the same performance increase as going with fewer larger modules with the same amount of GB's.

Thanks,
Matt

 

[owbp]

Yes, AFAIK with 4,1 and 5,1 you want to have 3 modules per processor for best performance (3 for single and 6 for dual prosector setup).

 

[wiski15b]

Thanks for the input. Anyone else can confirm this?

 

[bookemdano]

Yes, the processors the 4,1 and 5,1 use are triple-channel, basically meaning they can access three modules at the same time. The bandwidth for the fourth slot (and eighth in dual-proc machines) is shared with the third slot.

So performance will take a hit if you populate the fourth/eighth slots. However, if you're actually running out of RAM and paging things back to disk frequently then you'll see better performance even if you use the fourth/eighth due to the increase in RAM amount.

Larger modules are better only in that they allow you to get higher RAM amounts using less slots. Ideally you want a matched set of three (or six for DP) modules all of the same brand and specs.

 

[wiski15b]

Yes, the processors the 4,1 and 5,1 use are triple-channel, meaning they can access three modules at the same time. The bandwidth for the fourth (and eighth in dual-proc machines) is shared with the third slot.

So performance will take a hit if you populate the fourth/eighth slots. However, if you're actually running out of RAM and paging things back to disk frequently then you'll see better performance even if you use the fourth/eighth due to the increase in RAM amount.

Larger modules are better only in that they allow you to get higher RAM amounts using less slots. Ideally you want a matched set of three (or six for DP) modules all of the same brand and specs.

Very informative, thank you.

 

[AidenShaw]

Yes, the processors the 4,1 and 5,1 use are triple-channel, basically meaning they can access three modules at the same time. The bandwidth for the fourth slot (and eighth in dual-proc machines) is shared with the third slot.

So performance will take a hit if you populate the fourth/eighth slots. However, if you're actually running out of RAM and paging things back to disk frequently then you'll see better performance even if you use the fourth/eighth due to the increase in RAM amount.

Larger modules are better only in that they allow you to get higher RAM amounts using less slots. Ideally you want a matched set of three (or six for DP) modules all of the same brand and specs.

Good post, but the sentence in bold cannot be over-stressed. The actual performance hits for the extra memory slots are quite small, and require carefully controlled benchmarks to see the effect. These benchmarks are usually written to effectively disable the CPU caches - typical applications that don't disable the caches see very little degradation.

Also, take a good look at how your system reports memory usage. For example, my home workstation shows:

Right now it's not doing much, so most tools report about 44 GiB available. A closer look, however, shows that only about 13 GiB is unused, and about 30 GiB is used for filesystem caches. Sometimes I'll see just a few hundred MiB as "free".

Depending on what you're doing, having dozens of free GiB for the caches can make Safari much snappier.

 

[wiski15b]

Depending on what you're doing, having dozens of free GiB for the caches can make Safari much snappier

 

[ibarnett]

Yes, the processors the 4,1 and 5,1 use are triple-channel, basically meaning they can access three modules at the same time. The bandwidth for the fourth slot (and eighth in dual-proc machines) is shared with the third slot.

So performance will take a hit if you populate the fourth/eighth slots. However, if you're actually running out of RAM and paging things back to disk frequently then you'll see better performance even if you use the fourth/eighth due to the increase in RAM amount.

Larger modules are better only in that they allow you to get higher RAM amounts using less slots. Ideally you want a matched set of three (or six for DP) modules all of the same brand and specs.

I recently blindly bought 64gb, 8 x 8.
After reading this I removed 2 sticks, can't believe the performance improvement (from Geekbench - not feel!) plus faster boot to chime time!!
Thanks for sharing.

 

[orph]

AidenShaw/bookemdano is spot on, so relay depends on what you are doing.
if you have to little ram then it's going to be a lot slower than any potential speed gain from 3 sticks of ram.

once you have the ram you need for what you do your not going to get a massive speed boost from trying to optmise ram placement, your better off spending on software/SSD's or cpu upgrades

at the mo i have 18gb ram most the time im fine but if i try to use AE it eats it all and i have way to little.

 

[h9826790]

I recently blindly bought 64gb, 8 x 8.
After reading this I removed 2 sticks, can't believe the performance improvement plus faster boot to chime time!!
Thanks for sharing.

Faster boot time because less RAM to initialise. If you leave only one stick at there, the boot time will much faster, but that doesn't mean your computer has better performance.

It's hard to believe anyone can feel the difference without benchmark, or not by running long time memory bandwidth intensive software. Unless, one of the RAM stick you removed is faulty.

However, if 48G is enough for you and the system can't even use the remaining RAM as cache, then 6x8 is better than 8x8 indeed.

 

[wiski15b]

Oddly enough, I noticed this by accident while doing a cpu upgrade on my 4,1/5,1 cMP. I had left in the original memory (only 2 sticks per cpu, totaling 4) after the cpu swap. I ran a geek bench and noticed my score had a difference of 2000 between the new memory (8 sticks) and old memory (only 4 sticks). The difference was solely in memory performance. FYI both sets were 1333.

 

[AidenShaw]

Oddly enough, I noticed this by accident while doing a cpu upgrade on my 4,1/5,1 cMP. I had left in the original memory (only 2 sticks per cpu, totaling 4) after the cpu swap. I ran a geek bench and noticed my score had a difference of 2000 between the new memory (8 sticks) and old memory (only 4 sticks). The difference was solely in memory performance. FYI both sets were 1333.

As I said a few posts back: "These benchmarks are usually written to effectively disable the CPU caches..."

If only the memory test in Geekbench was lower - you're not seeing lower performance, you're seeing a lower score on an artificial test.

Look at this test of the MP6,1 using one to four memory channels: https://forums.macrumors.com/thread...th-various-mem-configs.1704700/#post-18745317

Most components of Geekbench had nearly the same performance regardless of the number of channels.

 

[wiski15b]

As I said a few posts back: "These benchmarks are usually written to effectively disable the CPU caches..."

If only the memory test in Geekbench was lower - you're not seeing lower performance, you're seeing a lower score on an artificial test.

I understand completely. Thanks again for the input.

 

[ibarnett]

Faster boot time because less RAM to initialise. If you leave only one stick at there, the boot time will much faster, but that doesn't mean your computer has better performance.

It's hard to believe anyone can feel the difference without benchmark, or not by running long time memory bandwidth intensive software. Unless, one of the RAM stick you removed is faulty.

However, if 48G is enough for you and the system can't even use the remaining RAM as cache, then 6x8 is better than 8x8 indeed.

Edited my post - performance boost from Geekbench - not feel (don't believe I said feel originally anyway).