Benchmarks
As the TWINX modules managed to hit DDR-400 speeds with 2-2-2-6-T1 timings when run as a pair, I wanted to show you what kind of performance increase this would translate to in some of the more memory-dependant benchmarks. As our mix-and-match OCZ and Mushkin pair were only able to hit DDR-400 speeds with 2.5-7-3-3-T1 timings when run together (remember, this particular ABIT i845E is not one that relishes two sticks in tandem), I'll include their benchmarks as a pair to highlight how lesser memory would benchmark. Running the 2.26GHz Pentium 4 at 2550MHz / 150FSB / DDR-400 gave us the foundation upon which to test on. Also remember that on a P4, real usable bandwidth can be enjoyed even when one runs the memory asynchronously, unlike the nForce2.
First off, we'll run the usual Pifast calculation to 10 million places. Memory bandwidth and low latencies are drivers here.
Whilst the performance increase isn't spectacular, those who want the very last ounce of performance from their PCs will appreciate any gain. Remember that the CPU is at 2550MHz with a FSB of 150. Running the 3:4 FSB:memory ratio gives us our required DDR-400 speeds.
Another activity that has previously benefited greatly from increases in usable bandwidth is DVD-to-DivX encoding. I'm benchmarking by using the DivX4.1 CODEC with a 2-pass encoding of Gone in 60 seconds and 1800kb/s bit rate. An average is calculated when the first VOB is complete (both passes).
Using low latency memory, whether it be run in single of paired formation, certainly pays dividends over a time-intensive activity such as DVD-to-DivX encoding. I can see no reason why 1GB worth of low latency matched memory (2 x 512MB) wouldn't produce the same results as our 2 x 256MB sticks. If you want lots of fast, high-speed memory, the TWINX isn't a bad place to start.
Moving on to gaming and 3DMark 2001SE v330. Although the very latest 3DMark 2003 is hogging the limelight, its GPU-intensive nature dictates that it doesn't show just wha an affect memory speed and timings have on some of today's games. 3DMark 2001SE does, so that's why it's included here.
Another nice little boost here.
The flyby element of Unreal Tournament 2003 Demo is also sensitive to changes in memory speed and timings. Benched at 800x600.
What I've tried to show here is that running high-speed memory with low latencies can boost your benchmark scores. The real-world implications are a little less flattering. Without knowing beforehand which set of memory I was testing, there is absolutely no way that I could have told the difference from purely a visual perspective between the TWINX pair at DDR-400 @ 2-2-2-6-T1 and the OCZ / Mushkin combination at 2.5-2-3-7-T1 also at DDR-400. Is this memory with its premium price-tag aimed at the general PC user. The answer has to be an unequivocal no. It's aimed squarely at the enthusiast that wants the fastest system they can afford. In that context the benchmarks laid down by the TWINX pair do matter.