MOTUnation.com

So I had a little mishap with my MacPro4,1 8-core to 5,1 12-core upgrade. Don't know if I damaged CPUs during de-lidding or what, but I'll sort that out.

Anyway, I'm back to using my MacPro4,1 (single processor) that I upgraded to a 5,1 6-core and did some experimenting with RAM and ran GeekBench4 with 3 8GB sticks in slots 1-3, then I ran it again with all four slots filled with 8GB sticks. These 2009 Nehalem models are apparently designed such that slots 3&4 on the single processor models and slots 3&4 and 7&8 on the dual processor models share a memory channel and supposedly this slows it down. Well I'm here to share that it definitely did in my case.

I did this with Geekbench 4 which is a bit of a downer because its scores are lower than with Geekbench 3, but oh well. Here are the scores:

With 24GB RAM (8GB in Slots 1, 2, & 3):
Singe-Core Score: 3083 Multi-Core Score: 14813

With 32GB RAM (8GB in Slots 1, 2, 3, & 4):
Single-Core Score: 3008 Multi-Core Score: 13946

You can see the Single-Core score takes only a modest hit with RAM in Slot 4. It drops 75 points from 3083 to 3008, for about a 2.4% performance hit.

However, on the Multi-Core score it drops 867 points from 14813 to 13946, for a more significant 5.85% performance hit. That's nearly 6% lower score for the extra RAM in Slot 4.

I don't know what that really translates to in real life, but my gut is saying I'd rather run this machine with 24GB RAM than 32GB. Seems with the Nehalem machines, you're better off getting larger individual RAM sticks and increasing RAM that way instead of using slot 4 (or slots 4 & 8 in the dual processor machines).

Anyway... passing that along as I'm about to pull the 8GB from Slot 4 right now on my hexcore!

My understanding is that, as long as you don't use all your RAM (and disk caching is thus not put in play by the OS), the three sticks of RAM approach makes best use of the Mac Pro Xeon CPU's triple-channel memory path. However, performance will take a very substantial hit if RAM is filled and disk caching occurs.

As such, Geekbench scores may or may not reflect each user's real-world situation.

HCMarkus wrote:My understanding is that, as long as you don't use all your RAM (and disk caching is thus not put in play by the OS), the three sticks of RAM approach makes best use of the Mac Pro Xeon CPU's triple-channel memory path. However, performance will take a very substantial hit if RAM is filled and disk caching occurs.

As such, Geekbench scores may or may not reflect each user's real-world situation.

Does that mean that if I have 24GB using three sticks, but disk caching ends up occurring because I need more than that 24GB, then I'm better off filling Slot 4? I wonder how to check if I'm triggering disk caching? Probably something in activity monitor. That performance hit in Geekbench really threw me though. Always believed more RAM is better, but seems like avoiding slot 4 (or 4 & 8 in dual CPU machines) is sometimes a good idea.

I guess maybe a better experiment might be to load a large DP project when there's 24GB installed and then again when there's 32GB installed. I could use DP's performance meter but seems inaccurate (wish they had a numerical value along with the bar graph that would show you current and highest percent), or Activity Monitor.

Of course if money were no object I'd just buy 3 16GB sticks and make it 48GB. But what am I saying? That hexcore is going to be my office computer once I FINALLY sort out my 4,1>5,1 dual upgrade situation.

Yes, Apple's Activity Monitor utility's "memory" tab should give you the answer you seek.

HCMarkus wrote:Yes, Apple's Activity Monitor utility's "memory" tab should give you the answer you seek.

Sounds like that's a good test: open largest DP project, start playback in a very busy section of the project and make note of activity meter regarding percent of CPU DP is using, then add the 4th RAM stick and do the same thing. I may or may not do that though... depending on how soon I get the 8-core to 12-core dilemma solved. Knock on wood!

I read recently that this applies to “true” 5,1s also.

[img]Screen%20Shot%202017-10-02%20at%209.27.34%20AM[/img]

Screenshot from my 'About this Mac'/ Memory.

Knock on wood, I rarely, rarely, ever have any kinds of problems with DP and my 4,1 setup with the RAM setup like this.

I've brought the machine to it's 'knees' once in awhile if I'm using a bunch of intensive sample libraries, but I'm not loaded up with huge templates or anything like that.

A big project for me might use 4 to 6 different orchestral V.I. instances loaded with only a few choice instruments each (just the ones I like that are easy to use and always sound good that don't need tons of RAM or tracks (ie: using libraries with Kontakt, PLAY, Mach 5, Garritan Aria), maybe some soft-synths (easy on the CPU, ie: Korg Legacy series, SynthMaster), some real audio tracks, a drum program of some kind (ie: EZ Drummer, Stylus RMX), and some plugins (Wave, Nomad Factory plugs, DP native plugs) using Aux tracks.

I'd call myself an average, medium 'intense' user (audio, sample libraries).

If I'm using my 'Omnisphere', I'm much more careful. (But, then, when you use Omnisphere for a pad or something else, you don't need much else happening in your music or I find you end up with a sonic mess....because Omnisphere seems to create it's own 'sonic world' and it will dictate if you really need any other sound design or music happening in addition to it's sounds(...which are, actually, pretty fantastic to have access to.)

Unfortunately I can't see your graphic. It has to be hosted on a server somewhere with a URL like:
... in order to be visible.

Well here's something interesting.

So I recently, and FINALLY, successfully upgraded my 4,1 8-core to a 5,1 12-core. I ran Geekbench 4 with 48GB of RAM installed (8GB sticks in slots 1,2,3 and 5,6,7). Here were the scores:

Single-Core Score: 2,864 Multi-Core Score: 24,343

So today I installed the 16GB additional RAM I had ordered by putting 2 8-GB sticks in slots 4 and 8. Look what happened to the Geekbench 4 scores:

Single-Core Score: 2814 Multi-Core Score: 20,149

Breaking it down that's 50 points or a 1.7% performance hit on the Single-Core score which is negligible, but on the Multi-Core score it dropped 4,194 points or a 17% performance hit! I guess it being almost 10 times worse is perhaps the accumulation of the single-core performance hit across the 12 cores.

I haven't tested too much real world, but this bothers me to the point that I think I'm going to pull the two 8GB sticks and perhaps sell them. If I need to go bigger I can swap pairs of 8GB for 16GB in the future. Looking at the memory tab in Activity monitor I'm assuming the important stat is "Swap Used" and I have not see that be anything other than zero on the DP projects I've loaded thus far so I'm assuming that with 48GB, I'm not having to page memory to boot drive (SSD).

So... unhooking cables and pulling out the machine and yanking the two 8GB sticks for now.