How reliable is the ECC detection?

[BoMbY]

Hello,

the memory controller of the AMD Athlon 5350 (Kabini SoC) should support ECC according to AMD, but the motherboard vendors don't seem to want to support it. But there is the Asus AM1M-A which may, or may not, support ECC-RAM (different infos at different places on the Asus page). The BIOS does not provide any ECC options or info. But when installing a module of Kingston KVR16LE11/8 ECC-RAM Memtest86 shows the following:



The question is, what does memtest86 do? Would it enable ECC, if it was disabled by the BIOS (like it seems possible under Linux)? Are there any actual checks, if it is really working? What if, for example, the ECC circuits on the motherboard wouldn't be connected from the CPU to the RAM socket? Could ECC be enabled on the chipset side only (reporting a false activation state)?

Thanks and Regards,
BoMbY

[keith]

Would it enable ECC, if it was disabled by the BIOS (like it seems possible under Linux)? Are there any actual checks, if it is really working? What if, for example, the ECC circuits on the motherboard wouldn't be connected from the CPU to the RAM socket? Could ECC be enabled on the chipset side only (reporting a false activation state)?

MemTest86 does not enable ECC; it just checks the appropriate hardware registers to determine whether or not ECC detection/correction is enabled. To truly verify whether ECC is working, you will need to force the RAM to generate correctable errors (which can be difficult to do without specialized hardware). It is unlikely that the chipset is falsely reporting that ECC is enabled as the ECC state reporting is controlled via hardware.

[BoMbY]

Thanks. The funny thing is, after booting to Windows 7, the hardware values seem to be:

D18F3x44 MCA NB Configuration

Bit 23 (ChipKillEccCap: chip-kill ECC mode): 1
Bit 22 (DramEccEn: DRAM ECC enable): 0

D18F2x90_dct[0] DRAM Configuration Low

Bit 19 (DimmEccEn: DIMM ECC enable): 1

D18F3xE8 Northbridge Capabilities

Bit 4 (ChipKill: chipkill ECC capable): 0
Bit 3 (ECC: ECC capable): 0

Why should Windows do this, when it is enabled in Memtest86?

[keith]

If those values are correct, it may just be that your board is Chipkill capable but ECC is not supported.

Can you post or upload the MemTest86.log file. The log should contain the values of those registers.

[BoMbY]

Okay, I uploaded it here.

If I'm not mistaken, your detection is wrong for this CPU/chipset:

2014-08-30 12:09:54 - Getting memory controller info
2014-08-30 12:09:54 - find_mem_controller - found AMD 16h (1022:1532) at 0-24-2
2014-08-30 12:09:54 - AMD 10h and greater chipset init
2014-08-30 12:09:54 - DRAM config low=83090000
2014-08-30 12:09:54 - MCA NB config=48B00044
2014-08-30 12:09:54 - NbMcaLogEn is disabled. Enabling...
2014-08-30 12:09:54 - MCA Extended NB config=00000000 [EccSymbolSize=x4]
2014-08-30 12:09:54 - MC4_CTL=0000000000000000
2014-08-30 12:09:54 - MC4_CTL_MASK=0000000004000000
2014-08-30 12:09:54 - MCG_CTL low=00000000
2014-08-30 12:09:54 - find_mem_controller - AMD 16h (1022:1532) at 0-24-2
2014-08-30 12:09:55 - find_mem_controller - AMD 16h ECC mode: detect: yes, correct: yes, scrub: no, chipkill: yes
2014-08-30 12:09:55 - ECC polling enabled

"MCA NB config=48B00044" Bit 22 is 0, so ECC should actually not be enabled?

[keith]

Yes, you are correct. MemTest86 was incorrectly making the assumption that because the chipset is Chipkill capable, then ECC is enabled. Please give this corrected build a try:

http://www.passmark.com/ftp/memtest86-usb-5.1.1025.zip