Sidebar Sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

2018 Mac mini ± Hyperthreading

GrumpyCoder said:
While it is the user that interprets the results, it's the creators of benchmark tools that suggest you can just download, let it run and read some number.
Click to expand...
You're saying:
  • The results were misinterpreted.
  • It is the user that interprets the results.
  • The misinterpretation is not the users fault.
That doesn't follow. Or am I missing something?

I am no fan of Geekbench and we're more in alignment than we are apart on this issue. But the above doesn't make sense.
 
I looked at the breakdown of the +/- HT tests.
It seems that some of the individual benchmarks within Geekbench do show a decent HT boost, but others show a lower score. I am guessing that this averages out at almost no calculated net benefit across the range of tests included. Hence no change in GB score +HT.
 
pl1984 said:
That doesn't follow. Or am I missing something?
Click to expand...
What I'm saying is that the average user is made to believe he has to push a button and the benchmark tool spits out a meaningful number and that's it. That's deceiving and that fault is on the creators of such tool for spreading such nonsense information. The average user on the other hand doesn't know any better. Maybe you could blame the user for being naive... but I'd rather blame the tool creators for their nonesense.


zandorf said:
So "it depends".
Click to expand...
It does. The second link shows a lot of games. If gaming is your preference, stay away from HT as far as possible. It's just not worth it for games and in some cases, makes things worse.
However, for a general mix of applications you can't go wrong with HT. In some cases there's a benefit, in others there isn't. So if price isn't an issue, always pick HT (at least until Apple shuts off HT in macOS... OpenBSD did this, not sure if Apple will).
 
GrumpyCoder said:
What I'm saying is that the average user is made to believe he has to push a button and the benchmark tool spits out a meaningful number and that's it. That's deceiving and that fault is on the creators of such tool for spreading such nonsense information. The average user on the other hand doesn't know any better. Maybe you could blame the user for being naive... but I'd rather blame the tool creators for their nonesense.
Click to expand...
I am not trying to defend Geekbench (I am not a fan of it) nor blame the user. However you're saying "...the average user is made to believe...". Who is making them believe that? To the creators of GB they do provide some information on how the tests are performed under the "Understanding Scores" section from the download page.
 
Ploki said:
only 22k? Isn't the average 26k
Click to expand...
Yeah. I don't really understand that either. My single threaded score is higher than average, but the MC score is 10% below average. Something to do with only 8GB RAM? Or maybe I just have a duff chip haha!
 
Last edited:
pl1984 said:
It means Geekbench is a poor way to benchmark a system.
Click to expand...

No it means adding RAM increased his score for those particular tests. There is merit to Geekbench scores even if it doesn't paint the whole picture. It could indicate that there was a memory shortage in some task. Completely dismissing geekbench without learning from the results would be a poor choice to take.
 
Cheapassmac said:
No it means adding RAM increased his score for those particular tests. There is merit to Geekbench scores even if it doesn't paint the whole picture. It could indicate that there was a memory shortage in some task. Completely dismissing geekbench without learning from the results would be a poor choice to take.
Click to expand...

I find Geekbench useful, because it does roughly correspond with how Logic Pro (my main app) performs.
i.e.; if GeekBench multicore score is 35% better, I usually get 35% more tracks on logic benchmarks, or at least in that vicinity. (this has been the case pretty much since 2008)
 
  • Like
Reactions: Hessel89 and Cheapassmac
Spectrum said:
Update: I recently added 2x16GB to my mac mini. Geekbench scores increased:
SC: 5739 (8GB) > 5896 (32GB) +3%
DC: 22516 (8GB) > 25872 (32GB) +15%
Whatever (if anything) that means...
Click to expand...
The change in MC score probably means your earlier score was an outlier. On a freshly booted system, without any background stuff running (e.g. virus scanner), the GB CPU test shouldn't run any faster on 8GB vs 32GB (all else being equal). The test barely uses any RAM (you can monitor it in Activity Monitor while running the test).

I've seen other threads where the poster wonders why their scores went down after adding more RAM, so this question can run both ways.

The GB CPU test does run some CPU <-> RAM throughput testing, which could be affected by faster RAM (if the mini supports it - I don't know that it does), but that would largely be considered a synthetic benchmark with not much real-world difference in performance, and likely a couple percentage points at most in any case (nothing close to 15% for overall CPU performance).

Here's a quote from the Primate Labs support staff answering a similar question...
Primate Labs Support Staff said:
The effects of changing your RAM configuration on your Geekbench score can be difficult to predict. Geekbench 4 uses a fixed amount of memory, and scores are partly correlated to the speed at which your processor can access memory used for the workloads. Adding a substantial amount of RAM in excess of the amount that Geekbench uses is not expected to increase your score.

Your Geekbench score may increase if you add RAM to make more complete use of, for example, a quad-channel architecture. Adding RAM can also decrease your score if it is misconfigured, causing slow memory accesses. These kinds of special cases make it difficult to determine how adding RAM will affect your scores, if at all.
Click to expand...
 
ixxx69 said:
The change in MC score probably means your earlier score was an outlier. On a freshly booted system, without any background stuff running (e.g. virus scanner), the GB CPU test shouldn't run any faster on 8GB vs 32GB (all else being equal). The test barely uses any RAM (you can monitor it in Activity Monitor while running the test).

I've seen other threads where the poster wonders why their scores went down after adding more RAM, so this question can run both ways.

The GB CPU test does run some CPU <-> RAM throughput testing, which could be affected by faster RAM (if the mini supports it - I don't know that it does), but that would largely be considered a synthetic benchmark with not much real-world difference in performance, and likely a couple percentage points at most in any case (nothing close to 15% for overall CPU performance).

Here's a quote from the Primate Labs support staff answering a similar question...
Click to expand...

The mini supports dual channel memory support as well as lower CAS timing. You can't increase the mhz without additional modifications though.

RAM speed and bandwidth can and will affect CPU scores. It is common knowledge that a CPU can suffer 10% or more by going single channel instead of dual channel.
 
ixxx69 said:
The change in MC score probably means your earlier score was an outlier. On a freshly booted system, without any background stuff running (e.g. virus scanner), the GB CPU test shouldn't run any faster on 8GB vs 32GB (all else being equal). The test barely uses any RAM (you can monitor it in Activity Monitor while running the test).

I've seen other threads where the poster wonders why their scores went down after adding more RAM, so this question can run both ways.

The GB CPU test does run some CPU <-> RAM throughput testing, which could be affected by faster RAM (if the mini supports it - I don't know that it does), but that would largely be considered a synthetic benchmark with not much real-world difference in performance, and likely a couple percentage points at most in any case (nothing close to 15% for overall CPU performance).

Here's a quote from the Primate Labs support staff answering a similar question...
Click to expand...
I tested the Geekbench multiple times with 8GB RAM, over a course of time and many reboots. Always got scores around the same as posted above, even when being careful to quite all other active processes.

By contrast, the first (and subsequent) score after 32GB gives me consistently higher, even if I am sloppy and have software active in background.

I'm not arguing that it means anything (but it might). Certainly macOS GUI performance increases with more RAM, so maybe this also affects other things in some way we don't yet understand.

I'd be interested to see some other 8GB i7 scores.
[doublepost=1547890087][/doublepost]
Cheapassmac said:
The mini supports dual channel memory support as well as lower CAS timing. You can't increase the mhz without additional modifications though.

RAM speed and bandwidth can and will affect CPU scores. It is common knowledge that a CPU can suffer 10% or more by going single channel instead of dual channel.
Click to expand...
Both 8GB and 32GB are dual channel config.
 
Spectrum said:
I tested the Geekbench multiple times with 8GB RAM, over a course of time and many reboots. Always got scores around the same as posted above, even when being careful to quite all other active processes.

By contrast, the first (and subsequent) score after 32GB gives me consistently higher, even if I am sloppy and have software active in background.

I'm not arguing that it means anything (but it might). Certainly macOS GUI performance increases with more RAM, so maybe this also affects other things in some way we don't yet understand.

I'd be interested to see some other 8GB i7 scores.
[doublepost=1547890087][/doublepost]
Both 8GB and 32GB are dual channel config.
Click to expand...

What is the CAS timing of the 8gb and 32gb config, since we know they were in dual chan config? This can provide a different CPU score as well, though we're talking 2-5 percent at best here.
 
Cheapassmac said:
What is the CAS timing of the 8gb and 32gb config, since we know they were in dual chan config? This can provide a different CPU score as well, though we're talking 2-5 percent at best here.
Click to expand...
I replaced with the recommended Crucial pair. https://uk.crucial.com/gbr/en/mac-mini-(2018)/CT13492047
Timings are CL19, I really doubt this is the reason...I suspect that with 8GB, something was limiting somewhere.

Perhaps more importantly: Cinebench tests are identical with 8GB and 32GB RAM.
 
I was kind of in the same boat as you. The first geekbench score on my i7 with 32 gb mini was around 26900 and all scores after that were lower. Geekbench scores seem to be very volatile in that sense. I already knew that, but unconciously it still made me doubt my decicion in the i7.

Yesterday I tried novabench, which focusses more on raw CPU performance and these scores seem alot more consistant. Even after 10 sessions I get CPU scores around 1940, which puts it in the top of high end CPUs today. Now my mind is put at ease.
 
Hessel89 said:
I was kind of in the same boat as you. The first geekbench score on my i7 with 32 gb mini was around 26900 and all scores after that were lower. Geekbench scores seem to be very volatile in that sense. I already knew that, but unconciously it still made me doubt my decicion in the i7.

Yesterday I tried novabench, which focusses more on raw CPU performance and these scores seem alot more consistant. Even after 10 sessions I get CPU scores around 1940, which puts it in the top of high end CPUs today. Now my mind is put at ease.
Click to expand...
https://novabench.com/systems/mac
less than 15" tho? :confused: It turbos higher (due to thermals) and has higher base clock
 
Spectrum said:
Is no-one else surprised that turning off hyperthreading has no effect on the Geekbench score?
Click to expand...

Not surprised at all. Because that's how hyperthreading works.

Hyperthreading allows a second thread to operate on the same core. The only way it can do MORE work is if the first thread is blocked (typically on storage or networking I/O).

If a benchmark shows a difference between hyperthreading and non-hyperthreading, by definition it would be measuring a I/O-bound transaction workload. Which is not a CPU-bound workload.

---
Edited later: By more work, I mean that the idle CPU can work on that second thread if the first thread is blocked. If neither is blocked on I/O, then both share the CPU... TANSTAAFL.
 
  • Like
Reactions: Hessel89 and Ploki
bplein said:
Not surprised at all. Because that's how hyperthreading works.

Hyperthreading allows a second thread to operate on the same core. The only way it can do MORE work is if the first thread is blocked (typically on storage or networking I/O).

If a benchmark shows a difference between hyperthreading and non-hyperthreading, by definition it would be measuring a I/O-bound transaction workload. Which is not a CPU-bound workload.

---
Edited later: By more work, I mean that the idle CPU can work on that second thread if the first thread is blocked. If neither is blocked on I/O, then both share the CPU... TANSTAAFL.
Click to expand...
Sure...but then why does Cinebench benefit so much from Hyperthreading? I would not imagine it is I/O bound.
 
Spectrum said:
Sure...but then why does Cinebench benefit so much from Hyperthreading? I would not imagine it is I/O bound.
Click to expand...
Hyperthreading is about scheduling resources whereas cores are about doing work. If the cores are being fully utilized then hyperthreading will not be able to schedule anything on them. There are any number of reasons a core might not be fully utilized. I/O is one such thing. Memory access is another. Waiting on a dependency from another piece of code can be another.
 
  • Like
Reactions: Ploki
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back