AMD Athlon X2 3800+
Dual Core Review - Part 2
Cachemem and the effect of running the memory clock at 433 MHz
Cachemem memory and cache access latencies with the memory bus running synchronously at 400 MHz (solid blocks) or else at 433 MHz (transparent blocks) -- lower is better. All latency settings were identical. One issue we ran into originally was that upon increasing the memory frequency to 433, the CPU HT frequency increased (without any manual interference from 200 to 216 MHz) meaning that the CPU was running faster, apparently using the memory clock (433 MHz) as reference. We manually adjusted the clock back to 200 MHz and checked the performance using entirely CPU-centric benchmarks such as Caligari TrueSpace as well as by comparing the cache access latencies in cachemem as reference. The graph clearly shows a significant performance degradation resulting from running the memory bus faster - in asynchronous mode.
Interestingly, in most other application benchmarks, e.g. CineBench 2003, increasing the memory frequency resulted in an approximately 2-5 % performance degradation, which is consistent with our expectations based on the reduced memory bandwidth and increased latencies. The tricky part in this case was that the original runs did show increased performance - until we noticed the increased clock speed issue (8%) mentioned above. In other words, there appears to be some sort of bug in the ASUS BIOS in that it automatically adjusts the CPU clock based on the memory clock value, rather than running in asynchronous mode. Needless to say that most system benchmarks scale fairly well with an increase in clock speed, as long as they are not graphics or e.g. I/O limited. However, in case only the memory clock is increased, the system will take quite a performance hit. (Cinebench 2003: C4D Shader: 341 vs. 323, Raytracing: 339 vs. 320 for DDR 400 vs. 433, respectively)
WorldBench5
|
