Across various forums, folks reporting sometimes that they detect less memory available in windows, bios, memtest86+/DOS, than physically installed. For example, installed 6GB or 12GB and only detecting 4GB or 8-10GB respectively.
I thought I'd outline some possible reasons for detecting less memory than installed:
1. incorrect CPU VTT (uncore) voltage set - what Asus terms as QPI/DRAM voltage, Gigabyte terms as QPI/VTT while most other boards refer to it as CPU VTT voltages. Vdimm is DRAM Bus Voltage in bios.
2. incorrect QPI/DRAM to VDIMM relationship - every cpu/mem pairing will have an optimal voltage differential between QPI/DRAM to VDIMM voltage usually between 0.01 to 0.5v difference. Fall out of the optimal differential voltage for cpu/mem pairing will = less memory detected and/or lower memory bandwidth reported compared to when optimal voltage differential is in place. You also want to read Intel Core i7 920 Overclocking Introduction Guide on X58 to observe some guidelines such as keeping Uncore memory frequency at 2x times that of DRAM memory frequency - so at 2:1 uncore to memory frequency ratio. If you set your uncore higher than 2x memory frequency you will need way more QPI/DRAM (uncore) and VDIMM memory voltage to get stable (see tests).
* Example, say you have 6x2GB @1333mhz working fine at 1.35v QPI/DRAM with 1.65v VDIMM. You try to overclock the memory to @1600mhz speeds, and you bump VDIMM from 1.65v to 1.75v and now only see 8-10GB detected. You've moved out of the optimal differential range for QPI/DRAM to VDIMM which @1333mhz was 1.65 - 1.35v = 0.3v. At 1.75v VDIMM you left QPI/DRAM at 1.35v so 0.4v differential voltage. Then you bump QPI/DRAM to 1.45v to keep 0.3v differential voltage and 12GB is detected again. Now this is just an example, it could be you need 1.85v VDIMM for your memory to be stable and if 0.3v is optimal differential voltage it means you need QPI/DRAM set at 1.55v
3. Setting some important voltage settings to AUTO instead of lowest manual available voltage option. See 2nd post here for suggestions.
4. faulty memory modules
5. faulty memory dimm slots
6. sort of combination of 4+5 above where memory modules like particular memory dimm slots - could come down to PCB differences for each module along with voltages for QPI/DRAM and VDIMM.
7. If you see full memory size in DOS/memtest but not within windows, then it could be due to 32bit/64bit memory addressing size being chewed up by video card and pci/pci-e devices as explained at [info] Windows maximum supported memory size and http://msdn.microsoft.com/en-us/libr...8VS.85%29.aspx. So if you have 6GB installed, but have 1GB video card on 32bit OS you may only see between 2.25-3GB of memory available in windows. On 64bit OS, you may only see 3.5-4.5GB memory available. If you add other PCI-E devices like sound cards, then that also further eats into available windows memory. Windows 7 users: read [info] Windows 7 usable memory less than physically installed memory
8. Update: Added another possible reason, too tight tension in mounting cpu cooler to motherboard. Improper contact of cpu pads with socket might also result in less memory detected. So especially for water cooling folks to try loosening the screws and tension a tad for cpu mount. You can try removing cpu from socket as well and reinstalling/remounting cpu cooler and see if that helps.
9. Update: Apparently short circuiting the motherboard can cause memory detection errors as well i.e. check cpu cooler's back plates and if they short any part of the backside of the motherboard. Example here and here.
10. Update: For Core i7 980X Gulftown owners, the uncore to mem frequency ratio requirements changed from 2x ratio to allowing 1.5x ratio, but 1.5x and 2.0+ ratios may need more CPU VTT and VDIMM than normal, thus safe uncore ratios usually between 1.66x and 1.84x for slightly lower CPU VTT and VDIMM usage. So for 12x mem multi might try between 22-23x uncore multi instead of 24x. For 14x mem multi might try 26-27x uncore multi instead of 28x uncore multi. For 8x mem multi might try 15x uncore multi instead of 16x uncore multi. All just practical experience with my Core i7 980X - full details here.
Notes:
1. Also don't count out needing to tweak, IOH, IOH/ICH PCI-E, ICH voltages as well as DRAM DATA/CTRL REF voltage tweaks for each memory Channel A/B/C
2. When testing full memory dimm slot populated configurations (6x2GB) or any configs, it's also worth testing each memory module (label them with masking tape from #1 to #6 etc) in each of the 6 dimm slots, jot down notes as to how each module performs in each of the 6 dimm slots, then pair them up with best matching of memory modules for dimm slots. Reason is triple channel kits while sold to you as matched, they are rarely really matched in terms of overclocking characteristics - you'll always have a module or two which either does better or worse than the average of the 3 modules. Now take that to 6 modules and you can see why sometimes it might not work that well.
3. It's also worth testing each module (labeled #1 to #3 or #6), rotating them in the dimm slots. If the same module doesn't get detected in cpuz spd tab (drop down menu to access other dimm slot inserted module info) and in bios RTL channel A/B/C (one channel reports RTL = 0 - so if RTL channel B = 0) that means module in dimm slots 3 or 4 is the problem module. On other hand if same dimm slot has issues regardless of module used in that slot, then could be dimm slot/board. Of course these symptoms don't rule out other reasons as listed above