Benchmarking
It bears repeating that the Athlon cannot fully use the extra bandwidth that dual-channel memory brings. We expect to see some improvement over single-channel running, although we don't expect it to be as great as, for example, on a P4 platform. Let's have a look at how SiSoft SANDRA sees the unbuffered memory results.
Straight off the bat, dual-channel memory is favoured by our synthetic friend. A significant increase is reported when going from single-channel to dual-channel memory. Running a tRCD of 2 clocks gives that little bit extra, SANDRA reports.
Running Pifast v41 next. It simply calculates the constant Pi to the desired number of places. 10 million is chosen for this test. If you want to run it for yourself, click here for the benchmark standings and download link. Just unzip and click on the .bat file.
Here's some tangible gain with dual-channel memory. We're looking at around a 4% gain in performance from SC to DC. The worrying part, as far as OCZ's memory is concerned, is just what kind of performance advantage a tRCD of 2 clocks eeks out. Almost 1.5 seconds quicker with exactly the same frequency.
A look at gaming now. 3DMark 2001SE v330 is a low latency lover.
Again, running dual-channel memory provides the kind of extra boost that a tRCD of 2 clocks does. If you believe that I put too greater an emphasis on that particular latency, you'll now see why. Single-channel performance was better than expected. The lobby test is extremely sensitive to the amount of bandwidth and also to the latency of that bandwidth.
The key to this test is lots of low latency bandwidth.
How about another benchmarks sensitive to bandwidth and latency ?. UT2003 Demo's bot match.
And now Quake III v1.30
They started off in this order and they've finished off in the same way.