Disk speed has almost no impact. Even in a trivial case (~2000 grid cells) dump.f90 only takes a few percent of the run time and much of that is still processing data to get it ready to write to disk.
on this motherboard there is some performance gain with quad channel memory in going from 1 to 4 memory sticks.
This white paper from HP states that “By interleaving memory access across multiple memory channels, the integrated memory controllers are able to increase memory throughput significantly. Optimal throughput and latency are achieved when all channels of each installed CPU are populated identically. As Figure 8 shows, adding a second DIMM to the system (and thus populating the second memory channel) essentially doubles system read throughput. Gains in throughput for each additional DIMM installed are almost linear until all eight memory channels are populated.” http://h20565.www2.hp.com/portal/site/hpsc/template.BINARYPORTLET/public/kb/docDisplay/resource.process/?spf_p.tpst=kbDocDisplay_ws_BI&spf_p.rid_kbDocDisplay=docDisplayResURL&javax.portlet.begCacheTok=com.vignette.cachetoken&spf_p.rst_kbDocDisplay=wsrp-resourceState%3DdocId%253Demr_na-c03293145-2%257CdocLocale%253D&javax.portlet.endCacheTok=com.vignette.cachetoken This is for the same CPU family, but different computer manufacturer.
This white paper from DELL states that “First and foremost, customers concerned about extracting all available performance out of the memory subsystem should strive to populate all memory channels evenly with the same rank, size, and speed of DIMMs. “ and “As Figure 31 indicates, the overall system memory bandwidth declines as memory channels are depopulated. This decline occurs in an almost linear fashion as fewer memory channels are contributing to overall bandwidth. This decline in memory bandwidth will have direct performance impacts on standard server workloads, especially scientific workloads that tend to be the most memory bandwidth sensitive. Anything less than 4 memory channels populated indicates an unbalanced memory configuration.” http://i.dell.com/sites/doccontent/shared-content/data-sheets/en/Documents/12g-memory-performance-guide.pdf This is for the same CPU family, but different computer manufacturer.
“the bottleneck of computing is currently in data transfer [between CPU and memory] not in processing” http://www.cfd-online.com/Forums/hardware/95432-ram-speed-worth-fastest.html
“understand that it is vital to utilize all 4 memory channels for each CPU, as 2 channels will only have approx. half the transfer rate“. It then goes on to talk about memory “Ranks” which is above my head. http://www.cfd-online.com/Forums/hardware/124834-optimal-memory-configuration.html
This post talks about getting the fastest/best ram and the most memory channels and these being important. http://www.cfd-online.com/Forums/hardware/111416-hardwares-workstation-more-important-do-cfd-analysis.html
“Neither CPU clock speed, nor cache size, nor even architecture, is as significant as the memory system, when it comes to performance in CFD”, and also when referring to whether 8 x 4GB or 4 x 8GB is better “8 channels, 8 slots, 8 modules.” http://www.cfd-online.com/Forums/hardware/113263-advice-technical-requirements-new-fluent-workstation.html