iMac Core Duo 17" 1.83 GHz Dual Channel vs. Single Channel RAM

[jepjepjep]

The dual channel vs. single channel performance debate was really intriguing me, so I headed down to the local Fry’s and picked up their last two 1 GB sticks of RAM (Mushkin #4700769, $99).

I ran my own little benchmark using DVD2OneX.

I encoded a DVD movie (previously decrypted and ripped to my hard drive Mac the Ripper) with DVD2OneX and measured the time it took to go from 0 to 50% complete.

With 1.5 GB RAM (Single-Channel) it took 15 minutes and 21 seconds to encode from 0-50%.

With 2 GB of RAM (Dual-Channel) it took 15 minutes and 16 seconds to perform the same task.

More details at my blog: http://papac.homeip.net/joe/blog/index.html



I hope this helps some of you in choosing memory upgrades.

 

[powerbook911]

A better comparison might be comparing 2x512 sticks to 1 1GB stick.

Nevertheless, thanks for sharing your results. I also wish I had a Fry's by me. That is a great price on the memory!!!! I had to pay $68 shipped, for a 512 stick.

 

[reberto]

I would test it with Xbench. What you did does not use very much memory and mostly used the DVD and Hard drive, Xbench, tests everything.

 

[jared_kipe]

but xbench is unreliable and a poor test. My 20" with 2gb ram and 256mb VRAM gets anywhere between 55 and 63 or so in xbench. The UI test is a real killer, scores around 20. What a pitiful test, I can tell the UI is as snappy if not more so than my PM dual 1.8GHz, so why doesn't it show it?

 

[reberto]

but xbench is unreliable and a poor test. My 20" with 2gb ram and 256mb VRAM gets anywhere between 55 and 63 or so in xbench. The UI test is a real killer, scores around 20. What a pitiful test, I can tell the UI is as snappy if not more so than my PM dual 1.8GHz, so why doesn't it show it?

remember, it compares it to a Dual 2.7 ghz G5

 

[jepjepjep]

I would test it with Xbench. What you did does not use very much memory and mostly used the DVD and Hard drive, Xbench, tests everything.

The DVD contents were decrypted onto my hard drive prior to the test. The encoding was done from the hard drive and written to a different folder on the hard drive.

Any other suggestions for a better test? I shy away from XBench because it has a rather poor reputation.

 

[majorp]

i've said it before and i'll say it again, running dual channel is of no benefit unless the memory of single channel memory is less than that of the system bus, it's 667/667MHz thus single channel saturates the fsb.

https://forums.macrumors.com/threads/173944/

It seems if your using 667mhz Ram dual channel doesn't make any difference (in theory)

 

[DarkAdept]

Improved test suggestions ...

The DVD contents were decrypted onto my hard drive prior to the test. The encoding was done from the hard drive and written to a different folder on the hard drive.

Any other suggestions for a better test? I shy away from XBench because it has a rather poor reputation.

XBench may not give you a good overall picture, but the memory and threading tests have shown a real advantage on the new machines and may give you some insight. It would be worth a try.

The two problems I see with your current test are:

1) Lots of disk I/O. Even though you've ripped the DVD content and are just encoding it, you still spend a lot of time reading/writing from your hard drive where memory bandwidth is a non-issue. This will likely reduce the effectiveness of the test.

2) Likely single-threaded operation. The dual-channel memory is more likely to come in handy when trying to keep both cores fully occupied. You could try something trivial like running two encodings in parallel or look for other tests that peg CPU usage at 200%.

 

[mkrishnan]

The two problems I see with your current test are:

Thanks (to the OP) for sharing, but yes, I also tend to think this is not the most memory dependent test... So what are you going to do, in any event? Return the memory, or keep it?

 

[jared_kipe]

remember, it compares it to a Dual 2.7 ghz G5

Nope it compares it to a Dual 2GHz G5 PM. I wouldn't have a problem if it was compared to a dual 2.7 sheesh.

 

[jared_kipe]

Keep in mind without the UI test I can score 96 or so. And there are some mini tests that really bring the score down even if they are somewhat erellivant. Like the vecLibFFT I got 43. They really need to optimize the code for the machine, then it wouldn't surprize me if we see 120 or better for the high end imacs. Oh and its a thread test monster, scoring an average 200.

 

[jepjepjep]

Thanks (to the OP) for sharing, but yes, I also tend to think this is not the most memory dependent test... So what are you going to do, in any event? Return the memory, or keep it?

I will return one of the 1 GB chips and keep one along with the default 512. 1.5 GB seems to be the sweet spot interms of price/performance ratio.

 

[gnasher729]

With 1.5 GB RAM (Single-Channel) it took 15 minutes and 21 seconds to encode from 0-50%.

With 2 GB of RAM (Dual-Channel) it took 15 minutes and 16 seconds to perform the same task.

Excellent information. Seems the best value for money would be to add a 1 GB chip bought from Crucial or someone else for a total of 1.5 GB. Don't know about prices in USA, but in Britain the best buys are: 512+512 MB: Buy from Apple for £70. 512+1GB: Buy from Crucial for £105. 2GB+2GB: Buy iMac with 1GB from Apple for £70, another 1GB from Crucial for £105.

 

[gnasher729]

but xbench is unreliable and a poor test. My 20" with 2gb ram and 256mb VRAM gets anywhere between 55 and 63 or so in xbench. The UI test is a real killer, scores around 20. What a pitiful test, I can tell the UI is as snappy if not more so than my PM dual 1.8GHz, so why doesn't it show it?

The user interface will not display any faster than the framerate of your monitor, that is 60 frames per second.

Let's say you write a program that changes the title of a button as fast as it can. The hardware may be fast enough to do that thousand times per second. But the operating system decides that this is entirely pointless and a waste of processor power, because your monitor has a frame refresh rate of 60 frames per second, and if the button title is changed more often than that, you will never, ever see it!

So this is just an example of XBench getting it completely wrong. What they have to do is change the test; giving it so much work that even a fast Mac can't do all the work in 1/60th of a second; that way they will measure how fast the Macintosh is.

 

[mkrishnan]

I will return one of the 1 GB chips and keep one along with the default 512. 1.5 GB seems to be the sweet spot interms of price/performance ratio.

That's what I have in my iMac G5. Glad to see the bench doesn't seem to have moved up that much because of Rosetta...

 

[bsn]

Xbench Results.

I would have to agree the testing methodology that was employed is not well suited for testing memory performance. I have just duplicated the configuration on my new iMac Core Duo 17". I tested using Xbench with 1.5 GB of RAM (Single Channel) and again with 2.0 GB of RAM (Dual Channel). While my tests are not truly scientific, I did try to make an accurate test. I DID see a measurable improvement in memory bandwidth with my iMac running in Dual Channel Mode.
I have posted the results below. The Dual Channel Results are with identical Crucial / Micron RAM.

Note the improvement on the stream tests.

Xbench Single Channel Mode 1.5 GB RAM.

Xbench Dual Channel Mode 2.0 GB RAM.

About This Mac