Memory Subsystem Analysis

Because the majority of the engineering efforts have been concentrated on the memory subsystem, this is an ideal place to start so that some of the benchmark differences later on can be accounted for.

SiSoft Sandra Memory Bandwidth Test

SiSoft Sandra's Memory Bandwidth Test does not provide for very interesting results. It looks like 1T timing gives the nForce 4 armed with the Corsair only a slight bandwidth advantage over the ABIT AA8XE. The nForce 4 with the OCZ is just slightly behind the 925XE board, a pretty remarkable feat considering the difference in timings.

Rightmark Memory Analyzer

Rightmark gives an analysis of both the average and peak 'real' bandwidth of the different configurations. It is interesting to note how close the average of the nForce 4 Intel with the Corsair is compared to the peak reading. The OCZ with the relatively high latencies takes a bigger hit here than shown in Sandra.

Maximum read speeds from the different configurations is pretty consistent throughout all three hardware combinations. Once again it is the average that shows the bigger differences. The nForce with the OCZ and the ABIT 925XE are more or less tied for average read through put while the nForce with the Corsair fares quite a bit better with roughly a 270MB/s advantage over the ABIT (roughly 5%). Read throughput is very important because reading from memory is arguably more important than writing to it as far as latency is concerned.

The nForce 4 for Intel with the Corsair configuration takes a slight lead. Again, note that the OCZ with the loose timings holds up very well to the 925XE.

The final test in the Rightmark Suite is a measurement of read latency - when the processor cannot find something in the L1 or L2 cache, it hops to the system memory. Therefore, a low latency is essential to keep the processor pipeline as full as possible. The efficiency of the nForce SPP memory controller is very apparent here - pretty much every case for the nForce/Corsair combo is 15% faster than the 925XE. The OCZ modules with the significantly higher memory module latencies still come in ahead of the 925XE with differences greater than 6% in every case.