[Eug]

I have a 24" C2D 2.33 with 2 GB RAM. I am considering going to 3 GB.

Has anyone done any application testing with matched vs unmatched RAM on this system, say with 1.5 GB vs. 2.0 GB? I'd be curious to see the results, but I haven't seen this done anywhere.

I know it makes a huge difference on the Mac mini, but the difference there is that the GPU is heavily dependent on the RAM speed. This is not the case with my 7600 GT. Also, while some tests have been done on the Mac Pro, the difference there is that they're quads, and use different memory.

I'm particularly interested in video encoding benchmarks, but would be interested in other application benchmarks as well. (I am NOT interested in Xbench memory benchmarks.) Video encoding because I do a fair amount of it sometimes, and I hear that on some video encoding benches, memory speed can be quite important.

[iMacYouMac]

I've wondered this as well. I think it may depend. Doing a synthetic benchmark will most likely give the win to the Dual-Channel setup because it effectively has double the memory bandwidth possible. The thing is that we rarely saturate all that bandwidth (hence synthetic). We usually recommend running that way if possible, though. I believe if you work with a lot of huge files you may be better off with the 3GB. If you encode a lot of instructions then the Dual-Channel may be better. Let's see what everyone else has to say, though. I'd love to see some benchmarks on this. I may be able to find one for a PC with a similar chipset platform.

[P]

There are three options: single channel, matched dual channel, unmatched dual channel. The difference between any of them is very small for the current iMacs, because the FSB is actually saturated with only one DIMM. For desktop models of the Core 2 that is not possible - FSB is 1066 MHz right now and 1333 MHz soon enough, and there are no DIMMs faster than DDR2-800 - but Apple uses the laptop model with only a 667 MHz bus.

(I should probably be writing MT/s instead of MHz - it's really only a quarter of the advertised clockspeed, but with 4 transfers per cycle - but the point still stands).

The gain for going to two unmatched DIMMs is extra bandwidth, but that is only useful for the GPU and the HD and neither needs much unless the GPU is integrated. The gain in going to matched DIMMs is 128-bit reads and writes, which decreases control traffic slightly, but it is really minimal for the CPU in regular workloads (it's different for an integrated GPU, which needs to read textures larger than 64 bits quite often).

[Simon]

From BareFeats:

DO MATCHING PAIRS OF MEMORY HELP?
Though we ran the tests above using matching 1GB SODIMMs in both MacBook Pros (for a total of 2GB), we also ran the same tests in the 15" MacBook Pro Core 2 Duo with one 1GB SODIMM and one 2GB SODIMM for a total of 3GB of memory. We wanted to see if non-matched modules would cause the MacBook to slow down due to the loss of interleaving.
The answer is "no, it didn't slow down." In some cases we saw a gain in speed. An example is Aperture where the "lift and stamp" ran 11% faster with 3GB of RAM. But that's probably due to the fact that Aperture + OS X = more than 2GB of total memory usage.
When we get our hands on a second 2GB SODIMM, we plan to run two of them. Even though only 3GB of the 4GB will be available, we want to see if a matched pair of 2GB SODIMMs will show anything different.

Your iMac is of course no MBP but it is using basically the same chipset. If it were my iMac, I'd go for the 3 GB RAM. RAM hungry apps like Aperture will likely benefit more from the extra gig of RAM than suffer from the unmatched pair.