v4.3 vs v7.3 bit fade test - Intel SC2600CP4 motherboard

[slatfats]

Hi there

I have testing some new RAM on a brand new motherboard. Using v7.3 of Memtest I'm consistently getting bit fade errors when testing single DIMMs, on all DIMMs that I test. I've now tried v4.3 of Memtest and it has not found any bit fade errors. All other tests in both versions have been finding no other errors.

Which version should I believe? Are there any inherent differences between the two versions for this test?

For info - the motherboard is an Intel SC2600CP4, dual socket but only running one Intel E5-2670v1 CPU and using ECC RAM.

Thanks,
slatfats

[David (PassMark)]

Can you post the details of the errors.
Address ranges and number of bits in error.

Also if it is ECC RAM, did you get any ECC warnings / Errors? V4 doesn't support ECC, so you won't see any ECC errors in V4.
V7 uses UEFI BIOS, so the address ranges being tested will be slightly different compared to V4.

[slatfats]

Thanks for responding so quickly.

There were no ECC warnings or errors that I could see. ECC was verified to be working in the system info in memtest86 v7.3.

Error details - all from v7.3:

Lowest Error Address	0x3FFFF918 (1023MB)
Highest Error Address	0x3FFFFB14 (1023MB)
Bits in Error Mask	00000000FFFFFFFF
Bits in Error	32
Max Contiguous Errors	2

Last 10 Errors

[Data Error] Test: 10, CPU: 0, Address: 3FFFFB14, Expected: FFFFFFFF, Actual: 00000000

[Data Error] Test: 10, CPU: 0, Address: 3FFFFB10, Expected: FFFFFFFF, Actual: 00000000

[Data Error] Test: 10, CPU: 0, Address: 3FFFFB0C, Expected: FFFFFFFF, Actual: 00000000

[Data Error] Test: 10, CPU: 0, Address: 3FFFFB08, Expected: FFFFFFFF, Actual: 0000000A

[Data Error] Test: 10, CPU: 0, Address: 3FFFFB04, Expected: FFFFFFFF, Actual: 0D73646E

[Data Error] Test: 10, CPU: 0, Address: 3FFFFB00, Expected: FFFFFFFF, Actual: 6F636573

[Data Error] Test: 10, CPU: 0, Address: 3FFFFAFC, Expected: FFFFFFFF, Actual: 20303532

[Data Error] Test: 10, CPU: 0, Address: 3FFFFAF8, Expected: FFFFFFFF, Actual: 20726F66

[Data Error] Test: 10, CPU: 0, Address: 3FFFFAF4, Expected: FFFFFFFF, Actual: 20747065

[Data Error] Test: 10, CPU: 0, Address: 3FFFFAF0, Expected: FFFFFFFF, Actual: 6C53202D

[slatfats]

Just to add - since my earlier post I have tested with 2x 8GB DIMMs, and a test with a single 4GB DIMM, and in both tests received the same number of errors at the same addresses as my earlier tests with 1x 8GB DIMM. So it's unlikely to be the RAM. I have a spare CPU which I will chuck in tomorrow night Aus/BNE time.

Cheers,
slatfats

[David (PassMark)]

Typically RAM failures are bit flips of one or two bits. Having 32bits change this less likely to be a RAM failure. Also having it occur at the same location with different sets of RAM sticks, plus no ECC warnings make it very likely the problem is elsewhere.

My guess is that there is a bug in the UEFI BIOS for this motherboard and some RAM is being incorrectly being flagged as free RAM. So RAM is being written to by MemTest86 and also by some other device. For example maybe the RAM was also allocated to the network card or the USB ports or something else as a memory buffer. So any activity on the other device results in memory corruption.

[slatfats]

OK cool - that gives me something else to test! Thanks!

[celemine1gig]

Just FYI, I saw the exactly (!) same issue on the Intel S2600CP2, too. So, the predecessor board seems to share the same problem.
Solution there is to enter the BIOS, go to "Boot Options" and set "EFI Optimized Boot" to "Enabled". By defaut this was disabled and caused the fade test problems.

[David (PassMark)]

So yes, very likely a BIOS bug.
Thanks for posting the solution.
We'll pass the details on to Intel.

[johnthekayaker]

I had this issue with the Intel S2600GZ. I found if I set the EFI Optimized Boot to Enabled in the Boot Options it runs without errors.