Benchmarks I
I like to test hardware on real, meaningful applications. Synthetic applications don't hold that much weight with me. Starting off, as always, with Pifast. The Pi calculator simply calculates Pi to X million decimal places. What's important, though, is that it's highly elastic to changes in memory speed and timings. The idea is to get information as quickly as possible from the memory to the CPU. Set to 10 million places.
First thing to remember is that the CPU is kept constant at 2762.5MHz (2.26GHz @ 162.5FSB). The memory speed is changed from 162.5MHz (2-7-3-3 Enhanced) to 217MHz (2-7-3-3 Enhanced - 3/4 CPU/Mem ratio) and also 217MHz (2-7-2-2 Enhanced). Simple changes in memory speed make a discernable difference to the results. Memory at 162MHz (PC2700ish) is around 7.5 seconds slower than the OCZ run at 217MHz with tight timings (2.8v). That's more than one would expect from changing to the the next processor up.
How about MP3 encoding - does it show gains with memory speed changes ?.
There's a slight and consistent advantage in running faster memory. The test was encoding U2's pop album into 192kb/s MP3.
Another activity that benefits 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).
Almost a 10% gain from simply switching from pseudo PC2700 CL2 memory to the OCZ running at 217MHz (PC-3500) at tight timings (albeit with 2.8v). Nice, meaningful increases especially when one considers how long two passes would take on a 2 hour film.
