Memory influence on fully loaded IvyBridge

[Raistmer]

Today I replaced single 4GB memory module with 2x4GB more faster ones (also working in dual-channel config).

Tests were done on fresh rebooted system before and after memory replacement:

Data from one of instances (all cores and iGPU are loaded with AP v7 apps)L

before:
WU : 0_Clean_01LC.wu
astropulse_7.00_windows_x86_64__sse2.exe  :
  Elapsed 244.998 secs
      CPU 240.710 secs
 
WU : 1_Clean_20LC.wu
astropulse_7.00_windows_x86_64__sse2.exe  :
  Elapsed 4490.125 secs
      CPU 4456.761 secs
 
WU : 2_Clean_01LC.wu
astropulse_7.00_windows_x86_64__sse2.exe  :
  Elapsed 193.284 secs
      CPU 190.836 secs

after:
WU : 0_Clean_01LC.wu
astropulse_7.00_windows_x86_64__sse2.exe  :
  Elapsed 189.099 secs
      CPU 185.033 secs
 
WU : 1_Clean_20LC.wu
astropulse_7.00_windows_x86_64__sse2.exe  :
  Elapsed 3602.254 secs
      CPU 3571.393 secs
 
WU : 2_Clean_01LC.wu
astropulse_7.00_windows_x86_64__sse2.exe  :
  Elapsed 181.718 secs
      CPU 176.780 secs

[Richard Haselgrove]

I remember that when I bought my trusty 8-core machine (host 2901600 - crunching for seti since, as it says, November 2006), memory speed was a problem.

It was originally running on Windows XP, 32-bit, with 2GB of RAM. But it's a workstation-class machine, built on a server motherboard, and wants to run memory in 4-channel mode. Adding two 1GB RAM modules to the two I already had enabled that mode, and increased throughput by some 25%. There are graphs somewhere in Number Crunching - we spent some time talking about it at the time. In particular, running 8 shorties - 'VHAR' - at a time really bogged it down.