Memory Tests
Pifast first, our own custom test to ten million places. It should give us an initial indication of how 848P is going to perform, compared to tweaked 865PE. The interesting result will be the comparison between 848P with its slightly relaxed memory timings, and 865PE running in single channel mode with PAT enabled.
Canterwood and the 3.2GHz P4 runs off with the win as expected and the other three test setups fall into line. Without a PAT enabled, 848P struggles to keep up with the single channel 865PE. It's not too far away, despite lower memory module latencies and no quick paths on the memory controller.
Sciencemark conceals the latency penalty but proves we're running in single channel mode on 865PE, incase there were any doubts. 90% efficiency in single channel mode, for both 848P and 865PE, is pretty good going. It drops to around 70% for dual channel on both 865PE and 875P.
The memory latency graph should reveal a bit more reasoning behind the slower Pifast time.
100ns latency for a typical access from main memory shows the effects of running slightly slower memory timings, along with no PAT enhancement. It sets the tone for the rest of the benchmarks, the 848P powered 4PLAI should be slower than single channel tweaked 865PE in almost everything.
