2018 Mini: 8GB vs 20GB benchmark

padams35

macrumors regular
Original poster
Nov 10, 2016
218
102
I just added a single 16GB module to my i5 2018 mini and thought I'd share the before/after Geekbench results for anyone interested in the single channel vs dual channel performance penalties.

Before: 4+4: https://browser.geekbench.com/v4/cpu/10925733
After: 4+16: https://browser.geekbench.com/v4/cpu/10949017

Overall the size mismatch created a 4% lower Geekbench 4 score.

However when viewed by category this becomes a 2% integer/floating-point penalty, 9% lower crypto score, and an 18% lower memory score.

An even closer look shows the larger single-channel memory has 5% longer latency and a staggering 33% lower memory bandwidth.

Maybe those numbers mean something to somebody with a more professional workload, but since the new Mini's memory in single-channel mode is still objectively better than the 2011 iMac's memory it is replacing I'm personally only seeing improvement.

Edit: About the iGPU... 3D rendering tests with Unigine Valley show the Intel UHD 630 is about 5% slower with 4+16 vs matched 4+4. eGPU tests showed no difference with the RAM change.
 
Last edited:

tpivette89

macrumors 6502
Jan 1, 2018
411
212
Newark, DE
Interesting results. I had intended on doing this (pairing a 16GB stick with one of the factory 4GB sticks) for now, and purchasing a second 16GB stick at a later date. Doesnt seem as though this is a good option, however, and I suppose I will just stick with the stock 8GB until I can afford both 16GB sticks.
 

Spectrum

macrumors 65816
Mar 23, 2005
1,347
703
Never quite sure
I just added a single 16GB module to my i5 2018 mini and thought I'd share the before/after Geekbench results for anyone interested in the single channel vs dual channel performance penalties.

Before: 4+4: https://browser.geekbench.com/v4/cpu/10925733
After: 4+16: https://browser.geekbench.com/v4/cpu/10949017

Overall the size mismatch created a 4% lower Geekbench 4 score.

However when viewed by category this becomes a 2% integer/floating-point penalty, 9% lower crypto score, and an 18% lower memory score.

An even closer look shows the larger single-channel memory has 5% longer latency and a staggering 33% lower memory bandwidth.

Maybe those numbers mean something to somebody with a more professional workload, but since the new Mini's memory in single-channel mode is still objectively better than the 2011 iMac's memory it is replacing I'm personally only seeing improvement.
Great test! DO you have any objective way to tell if it affects GPU performance? People have reported somewhat sluggish UI with large 4K resolution screens, that goes away with 2x16GB GB. I'm just wondering if that is also true with a mismatched 4+16Gb configuration.
 

macdos

macrumors regular
Oct 15, 2017
214
291
Great test! DO you have any objective way to tell if it affects GPU performance? People have reported somewhat sluggish UI with large 4K resolution screens, that goes away with 2x16GB GB. I'm just wondering if that is also true with a mismatched 4+16Gb configuration.
It would not have anything to do with that. If RAM is low, VRAM is also low.
 

rmdeluca

macrumors 6502
Oct 30, 2018
250
391
It would not have anything to do with that. If RAM is low, VRAM is also low.
It kind of would though. That “VRAM” is just shared system RAM. If you cripple its bandwidth by running single channel, it’s going to hurt the iGPU’s performance.

What some of us are curious about is how much the cache can hide that reduced bandwidth from real-world performance.
 

padams35

macrumors regular
Original poster
Nov 10, 2016
218
102
Interesting results. I had intended on doing this (pairing a 16GB stick with one of the factory 4GB sticks) for now, and purchasing a second 16GB stick at a later date. Doesnt seem as though this is a good option, however, and I suppose I will just stick with the stock 8GB until I can afford both 16GB sticks.
I chose the 4+16 because
a) It seemed likely I'd eventually outgrow 16
b) I'd rather start with single-channel now and upgrade to dual 2x16 if necessary than start with 2x8 now and be forced to choose between completely replacing the upgrade I'd paid for or downgrade to single after adjusting to dual.

Great test! DO you have any objective way to tell if it affects GPU performance? People have reported somewhat sluggish UI with large 4K resolution screens, that goes away with 2x16GB GB. I'm just wondering if that is also true with a mismatched 4+16Gb configuration.
Why... yes. I did run Unigine Valley benchmarks before the upgrade and can rerun those tests again after. Unfortunately I just jotted the 8GB average FPS results on note paper when testing iGPU vs eGPU and I won't have access to said paper again until next Monday.
 

Spectrum

macrumors 65816
Mar 23, 2005
1,347
703
Never quite sure
I did run Unigine Valley benchmarks before the upgrade and can rerun those tests again after. Unfortunately I just jotted the 8GB average FPS results on note paper when testing iGPU vs eGPU and I won't have access to said paper again until next Monday.
A 3D test is fine, but I was mostly just wondering whether 20GB seems smoother than 8 GB for 2D macOS user interface animations. Window resizing, expose, fullscreen etc.

I was also contemplating adding a 16GB stick now. And then in the future either another 16GB, or maybe even a 32GB. But you've got me wondering whether to just go for 32GB straight away. I can't see I would ever need 64GB...but I guess...one never knows...if the CPU lasts 6+ years again...
 

padams35

macrumors regular
Original poster
Nov 10, 2016
218
102
A 3D test is fine, but I was mostly just wondering whether 20GB seems smoother than 8 GB for 2D macOS user interface animations. Window resizing, expose, fullscreen etc.
Oh, for the MacOS user interface both were perfectly smooth on the 1080p test display I've been using. No problems noticed either before or after.
 
  • Like
Reactions: Spectrum

See this post: Officially giving up on the Mac Mini 2018 + 4K monitor dream

It looks like dual channel and 2x8GB is sweet spot.
Thanks, that was interesting. But, I still wonder how dual channel with 8 + 8 would compare to dual channel with 8 + 16, whether there would be any benefit to the latter arrangement, or if it would actually be worse. I have 8 + 8 now and I'm happy with it, but I'm just wondering if I ever wanted to go up to 32, would it make any sense to start by adding one 16 stick paired with one of the old 8's, and go all the way up later.
 

Spectrum

macrumors 65816
Mar 23, 2005
1,347
703
Never quite sure
Someone posted here that memory bandwidth is reduced (to the level of single channel) whenever the RAM sticks are mismatched. 4+16 has lower bandwidth than 4+4 in a Geekbench test (first post in this thread), which I presume is largely the difference between single and dual channel.
 
Someone posted here that memory bandwidth is reduced (to the level of single channel) whenever the RAM sticks are mismatched. 4+16 has lower bandwidth than 4+4 in a Geekbench test (first post in this thread), which I presume is largely the difference between single and dual channel.
I read that post. That's what got me wondering about other combinations, and whether anyone had any actual data to share.
 

ixxx69

macrumors 65816
Jul 31, 2009
1,119
635
United States
Dual channel only works with matched pairs. For general system performance, dual channel doesn't have much real-world benefits. For a dGPU, there wouldn't be any noticeable difference in gaming performance (like maybe a few percentage points).

For gaming performance with an iGPU, the dual channel does make a real-world difference because of the shared memory... I've seen benchmarks in the 25% range. But that's for 3D gaming.

For 2D/GUI, dual channel should not make any difference... that kind of memory bandwidth isn't necessary for compositing the desktop. But I haven't seen any benchmarks to confirm that speculation, so don't take that as fact.
 

Spectrum

macrumors 65816
Mar 23, 2005
1,347
703
Never quite sure
Dual channel only works with matched pairs. For general system performance, dual channel doesn't have much real-world benefits. For a dGPU, there wouldn't be any noticeable difference in gaming performance (like maybe a few percentage points).

For gaming performance with an iGPU, the dual channel does make a real-world difference because of the shared memory... I've seen benchmarks in the 25% range. But that's for 3D gaming.

For 2D/GUI, dual channel should not make any difference... that kind of memory bandwidth isn't necessary for compositing the desktop. But I haven't seen any benchmarks to confirm that speculation, so don't take that as fact.
Perhaps...but this post states that a single 16GB RAM stick still resulted in sluggish 2D GUI...
Officially giving up on the Mac Mini 2018 + 4K monitor dream
So either 16GB isn't enough RAM, or the single channel isn't enough. (Or it was user error.)
 

ixxx69

macrumors 65816
Jul 31, 2009
1,119
635
United States
Perhaps...but this post states that a single 16GB RAM stick still resulted in sluggish 2D GUI...
Officially giving up on the Mac Mini 2018 + 4K monitor dream
So either 16GB isn't enough RAM, or the single channel isn't enough. (Or it was user error.)
Yeah, you got to consider the "source" before using that as a "source". I put zero stock in it. Not saying it couldn't turn out be that way, but needs to be properly studied before being taken seriously. I've been trouble-shooting computers for decades, and 9/10 times it's user error.
 
  • Like
Reactions: DesertSurfer

padams35

macrumors regular
Original poster
Nov 10, 2016
218
102
Update

I reran Unigine Valley for 3D graphic comparison.

The iGPU was ~5% slower with mismatched 4+16 vs matched 4+4.
(5.9 -> 5.6 fps on Extreme preset, 21.7->20.7 fps on Basic preset.)

eGPU graphical scores were essentially unchanged.
 

padams35

macrumors regular
Original poster
Nov 10, 2016
218
102
Update #2

Unboxed my shiny new LG 24" 4K monitor and ran half-scale vs custom-scale tests. Sadly I don't have any 'before' numbers for 8GB on this one, but I noticed a couple of 4K questions. Maybe someone else with matched ram and a 4K display can supply those numbers.

Anyway, Unigine Valley Basic Preset with UHD 630 and 20GB Ram on...
1080p monitor w/ 8GB: 21.7 fps (
1080p monitor: 20.7 fps (100%)
4K monitor, half scale: 18.8 fps (~91%)
4K monitor, custom scale: 18.1 fps (~87.4%)

Clicking the green Fullscreen Window maximize button had some noticeable stutter with a custom scaled resolution on a 4K monitor (vs simply a little rough on default half scale) but I couldn't find any other UI animation issues.

My eGPU (RX 460-2GB) benchmarked ~5% slower with custom scaling vs default scaling, but otherwise didn't have any issues.