Missing memory (BIOS detects 8 x 2GBs but says only 12 GBs installable)

Question

OS: OpenSuse 12.3 (x64)

Problem: Unable to use all detected memory

I recently bought some more RAM for my workstation (2Rx4). The motherboard has 8 slots, and while BIOS recognizes that all 8 have a 2GB stick of RAM, the "Total Installable" is only 12GBs.

dmidecode tells me the maximum supported RAM is 32GBs, so I suspect this may be a rank issue, although I'm not super familiar with memory.

Update

To be explicit - does anyone have any ideas why I am unable to use all the detected memory?

I had 8GBs before in a 4x2GB configuration.

RAM details

2GB 240p PC2-5300 CL5 36c 128x4 DDR2-667 2Rx4 1.8V ECC FBDIMM

All 8 slots are registered as active (see dmidecode output) and the BIOS detects all 8 sticks

Motherboard details

Base Board Information
    Manufacturer: Dell Inc.          
    Product Name: 0GU083
    Version: A00
    Serial Number: ..CN1374073H00UP.

Available memory (BIOS also says 12GB installed memory, but lists 8 2GB sticks)

             total       used       free     shared    buffers     cached
Mem:         12039      11888        151          0        549      10825
-/+ buffers/cache:        512      11527
Swap:         1992        573       1419

Full dmidecode output here

UPDATE 2 memtest x86 only detects 12GBs of memory too;

Answer 1

I am going out on a limb here. Please let me know if I am wrong and I will either correct or delete the post:

Pre-amble:

• Your workstation is a Dell precision 490 with a 0GU083 motherboard.
• The board has an Intel 5000X chipset memory controller which uses fully buffered DIMMs.
• You have 8 DIMMs installed. All 8 are FB DIMMs with ECC support. (See your DMI decode log, lines 561 till 578 for DIMM #1, followed by the information of the other 7 DIMMS.
• 8x2 GiB is 16 GiB. The BIOS succesfully detects that 16 GiB is present.

During POST / memory checks an error is encountered: See line 549: Descriptor 4: Multi-bit ECC memory error and line 927: Status: Firmware-detected hardware failure.

Your motherboard uses tries to use the fully buffered DIMMs (FBDs) in sets of four. (See the manual at ftp://ftp.dell.com/Manuals/all-products/esuprt_desktop/esuprt_dell_precision_workstation/precision-490-dt_user%27s%20guide_en-us.pdf).

End_preamble:

This means that:

1. You do not need to worry about ranks. (?) Not 100% sure here, but ranks is about the electrical path between the memory controller and the paralel connected DIMMs. FB-DIMMs use a serial interface specifically designed to avoid rank problems.
2. At least one of the DIMMs posted an error. That error is either still present or needs to be cleared.
3. Linux probably did not add that faulty 2 GiB DIMM.
   New information added to the POST now makes it clean that while the BIOS detects 16GiB it also only 'installs' 12 GiB. Which is what Linux will use. So the part about Linux memory adding or skipping memory area's is not relevant.

What I do not grok is why that results in 12GiB memory rather than 14 GiB. The only reason I can think of of is if it gave up when it ran into an error after sequentially adding the memory regions. Something like this:

Add 2GiB DIMM.    2GiB now available.
Add 2GiB DIMM.    4iB now available.
Add 2GiB DIMM.    6GiB now available.
Add 2GiB DIMM.    8GiB now available.
Add 2GiB DIMM.   10GiB now available.
Add 2GiB DIMM.   12GiB now available.
Add 2GiB DIMM.   ERROR. STOP ADDING.  Remain at 12GiB.
Stopped. Do not try the possibly working remaining DIMM.

Confirmation of that from someone who knows a lot more about how memory gets added by the kernel would be great. Help on how to test for the broken area and how to avoid it would also be great.

Another potential problem is a damaged memory socket or a damaged trace on the motherboard. This can even affect multiply sockets since FB-DIMMs are chained rather than accessed in parallel.

If you check the datasheet for the Intel 5000X memory controller and look at Figure 1.1 then you will see this:

The figure shows up to 16 FBDIMM's. Up to is a keyword. Your has 8. The sane way for 8 sockets while keeping quad channel memory access would be 4 channels with 2 sockets each.

If one such channel is not working then you would loose two DIMMs, resulting in 12 GiB.

Answer 2

I got exactly the same problem on a 12 gb machine but only reading 4gb. After checking the resource monitor in task manager, it was showing 8gb reserved for BIOS and other devices. After unplugging various USB devices rebooting, all 12 Gb came back. Hope this helps.

Answer 3

I just found this thread because I had a similar problem. BIOS showed 4 GB in each slot but only had 8 GB available in Dual mode after upgrading to Dual Xeon 5365 and 8 x 4 GB at the same time. Before I had 4 x 2 GB with no issues.

I swapped DIMMs around with no real change ... I ended up cleaning the contacts of all DIMMs and when I put them back in it showed 32 GB available in Quad mode and it is now running MemTest86+ on it.

So, never forget that sometimes it may just be a good cleaning is needed.