Cinebench and Memory
it's actually quite true that Cinebench does benefit from more RAM bandwidth. Check this:
http://www.bit-tech.net/hardware/memory/2011/01/11/the-best-memory-for-sandy-bridge/8
Both Nehalem and Lynnfield can get a few hundreds more points depending on how much memory bandwidth they get. It seems small when considering that it's 12,000 points total, but... try that with about 2000 or 3000 and the scaling of a few hundreds will mean something.
i don't know if we're looking at the same numbers; i see 17000's for Nehalem and Lynnfield, with the speed mattering most to the Lynnfield. the results for the Sandy Bridge are strange, as the highest score happens with the slowest memory, and bounces in between.
in any case, the biggest difference is for the Lynnfield, which increases from 17189 @ 1.333MHz to 18098 @ 2.133MHz.
18098/17189 = 1.053, so we're talking about a 5% difference.
if we 'scale', we do so precisely by assuming this percentage stays the same, of course. so the relative difference is the same. the absolute difference will just get smaller with smaller scores though. (5% of 1,000 is much bigger than 5% of 200.)
anyways, it seems memory is doing something, but not too much.
My GarageBand Test
I'm interested in making sure I understand this test and the results that I and others get. So far, I'm very confused by your results. You've also raised questions about my results, or at least their interpretation. I'd like to settle this if possible.
i. Measuring CPU usage:
Well, now that I see your test methodology, I think
it's worth pointing out that under Activity Monitor, there is a line that reads "User%", which is more accurate than taking the CPU % inside and divide by 200 or 400.
Why? Because there is no way CPU scaling is that linear. 120% may actually mean that one core runs at 80% while the other runs at 40%, and that's perfectly normal. That doesn't mean only 60% of the CPU is used. We're dealing with 2 cores, not a single processor split in two.
As I pointed out above, though I did not use it originally, in my recent second run of the 2.26GHz Mini, i used the istat widget to monitor things like temperature and fan speed. this also gives a CPU reading, which coincides exactly with User%/2 (for the total two cores of the Mini).
if this doesn't satisfy you, please explain why.
ii. Throttling?
Also, this test just shows to me that your Mac Mini is indeed throttling itself. ... It can throttle because running at maximum frequency bumps up voltage, and thus... heat.
As I pointed out above, istat is telling us where my temp and fans were at, and these, as well as CPU readings, were quite stable during the test.
I think this is evidence that the Mini was, in fact, NOT throttling, no?
iii. Your Results
My Macbook Air 13" [2010 1.86GHz] gets 55% or 110% average during that "test". At 1.60GHz [running Coolbook].
So somehow my Macbook Air with a slower processor is actually faster than your Mac Mini? That doesn't seem right.
Also I think it's worth pointing out that I repeated the test with Coolbook on a Macbook Pro as well.
At 1.33GHz, GarageBand reports that there are too many effects to be played in real time, so it cuts off right there.
1.6GHz, 1.86GHz, 1.99GHz, 2.16GHz, 2.4GHz, and 2.66GHz all yield around 110% or 55% user % in both the Macbook Air and Macbook Pro. There is practically no difference whatsoever. The only difference is that at 1.6GHz on the Air and 1.99GHz on the Pro, there is a lot less heat. 10C less, to be more precise. The Air hits 55C while the Pro hits 45C.
Again, I don't understand your results.
First of all, it's a fact that audio effects like amp simulators are primarily CPU-intensive creatures.
My results are all different for different CPUs; the 2011 MBAs we tested all have SSDs and 4GB RAM. (I doubt disk has anything to do with this test, and Samsung vs Toshiba even less.)
Your results seem not to change when underclocking your CPUs (until you go too far). But also, your 2010 1.86GHz MBA does better than my Mini.
I think you've tried to explain this with throttling, but again, I don't see the Mini throttling. I think you tried to explain that your underclocked test being the same means GarageBand was working relative to the available resources, but then what about my results on the 2011 MBAs?
