Ganged mode giving slow read speed than unganged

truegenius

Distinguished
BANNED
Hey there,
I have been testing my system so many days
My system is this
Asus m4a890gtd-pro/usb3
amd phenom 2x6 1090t (@3.6ghz with 2.925ghz/2.925ghz cpu-nb/ht, no thorttling of any kind )
1x4GB 1600mhz cl9 gskill sniper ( @900mhz cl6-7-6 for this test )
1x4GB 1600mhz cl11 crucial ( @900mhz cl6-7-6 for this test )
window 7 ultimate 64bit
prime95 shows no errors.

So i was testing ganged and unganged mode
ganged is supposed to provide better read speed than unganged mode, but in my case unganged is giving 13GB/s in AIDA64 and maxxmem² and ganged is giving just 7GB/s which is almost close to single channel mode as when i place then in slot 1-2 to force single channel then it gives 6.9GB/s.
North bridge was set at 2.925ghz ( ~50% overclock ) to remove any cpu-nb bottleneck and ram were set to 900mhz to test what maximum bandwidth efficiency i could get out of ram when cpu-nb is not a bottleneck, and this test showed ~90% efficiency in unganged dual chanel and around 48% in ganged dual channel and 96% in single channel.

cpuz and aida64 confirms that ram, was running in the supposed mode.

So what could be the cause of extremely low bandwidth of ganged mode in comparision to unganged mode while it was supposed to give better read performance ?

winrar also confirms this score as it runs significantly slow in single channel and ganged dual channel mode

winrar score during various config

dual stick single channel
multi thread - 4665KBps
single thread - 1212KBps

single gskill stick single channel
multi thread - 4832KBps
single thread - 1219KBps

single crucial stick single channel
multi thread - 4689KBps
single thread - 1215KBps

ganged dual channel
multi thread - 5192KBps
single thread - 1262KBps

unganged dual channel
multi thread - 5503KBps
single thread - 1241KBps
 

truegenius

Distinguished
BANNED
from that thread
while ganged mode took a few wins and provided much higher memory bandwidth scores in the Synthetic Sandra VII benchmark
so that thread states that i was supposed to get higher bandwidth in synthetic read speed test like aida64, maxxmem etc and thats the problem i am getting lower bandwidth ( almost half ) in synthetic bench which is close to single channel.