Test 6 errors only on two different sets of RAMs

[PanPhoenix]

First of all thank you for making the software.

I'm building a new PC, with specs:
CPU: AMD 3600X
MOBO: Gigabyte X570 Pro Wifi
GPU: Nvidia 2060 super founders edition
and latest Windows 10 update and BIOS

Initially, I got G Skill RAM modules: F4-3600C16D-32GTZNC and ran with XMP enabled, but the system was getting BSODs. Resorted to memtest86 -
1. With 2 sticks - only 2 errors in Test 6 in 2/4 passes, everything else was good
2. Pulled out 1 stick - all tests passed
3. Ran the other stick in the same slot of step 2- only 2 errors in Test 6
4. Ran the "good" stick in step 2 in the slot of "bad" stick(step 3) in step 1 (hope this is explained well enough...) - everything passed

Based on above I quickly declared that one stick was bad and I needed new RAMs. Bought a slightly different model with different timings: F4-3600C18D-32GTZR

Right after inserting new modules, I went straight to memtest86 again, this time, no error - great!

Then I reinstalled Windows 10 just in case. Two hours later got another BSOD. And later crashes in games as well.

So I did another memtest86 - once again only two errors in Test 6. I even disabled XMP but still, same errors in Test 6 and only two.

I'm entirely confused and frustrated, and would appreciate some help -

1. With the initial set of RAM modules, was I right to conclude MOBO was good and one stick was bad?
2. Does it mean I'm just getting two sets of bad RAMs in a row?...This sounds too unfortunate to be true, so should MOBO be back on the suspect list? Could it be other hardware components in the system?
3. Given that it's always Test 6 for both two sets, and only two errors, could it be that RAMs are actually not the problem here?
4. Any suggestion on other test configurations or hardware setups I could try?
5. What's your opinion on what's happening here?

Thank you.

[David (PassMark)]

Can you post the errors from Memtest86. It is always interesting to see how many bits there are in error, and the addresses of the errors. Both from the original set and new sets if you have them.

Is isn't uncommon to get different errors in dual channel mode compares to single channel mode. But that doesn't explain all of the problem.

It does (initially) sound like a RAM issue. But the chances of getting two lots of bad RAM are low. So you now also need to suspect CPU or motherboard. Do you have a friend with a similar machine from which you can borrow a new CPU for testing?

[PanPhoenix]

Thanks - sure thing, posting them below. It might be a bit hard to get hold of folks now because of holidays (btw Happy New Year!).

Another generic RAM question I forgot to ask - it isn't likely that bad installation of OS or sudden loss of power would damage RAM, is it? Asking because when I was reinstalling Windows 10 between good pass and bad pass of the 2nd pair, the installation progress was hanging so I had to hard turn off and start over.

After going through the errors, it seems that failing addresses are kind of similar? Does it mean anything?

Another unfortunate theory is that this MOBO just doesn't play well with G.Skill RAMs - but both pairs have listed it on QVL.

1st pair, dual channel, XMP (my apologies, there were actually *12*, not 2 errors in this pass, was writing away from the machine from my (also faulty...) memory) -

Lowest Error Address	0xD5D207B8 (3421MB)
Highest Error Address	0xDB893D90 (3512MB)
Bits in Error Mask	00000000FFFFFFFF
Bits in Error	32
Max Contiguous Errors	1

2019-12-27 07:22:32 - [Data Error] Test: 6, CPU: 1, Address: DB893D90, Expected: 00000004, Actual: FFFFF7FF

2019-12-27 07:22:32 - [Data Error] Test: 6, CPU: 1, Address: DB88F7E8, Expected: DFFFFFFF, Actual: 00000001

2019-12-27 07:22:32 - [Data Error] Test: 6, CPU: 1, Address: DB88CF10, Expected: 7FFFFFFF, Actual: 00000100

2019-12-27 07:22:32 - [Data Error] Test: 6, CPU: 1, Address: DB88C638, Expected: 00000000, Actual: D664BAB7

2019-12-27 07:22:32 - [Data Error] Test: 6, CPU: 1, Address: DB886878, Expected: 00000000, Actual: D664BAA9

2019-12-27 07:22:32 - [Data Error] Test: 6, CPU: 1, Address: DB883BB0, Expected: 00000004, Actual: FFFFF7FF

2019-12-27 07:22:32 - [Data Error] Test: 6, CPU: 1, Address: DB87F608, Expected: DFFFFFFF, Actual: 00000001

2019-12-27 07:22:32 - [Data Error] Test: 6, CPU: 1, Address: DB87CD30, Expected: 7FFFFFFF, Actual: 00000100

2019-12-27 07:22:32 - [Data Error] Test: 6, CPU: 1, Address: DB87C458, Expected: 00000000, Actual: D664BAB7

2019-12-27 07:22:32 - [Data Error] Test: 6, CPU: 1, Address: DB876698, Expected: 00000000, Actual: D664BAA9

1st pair, single channel, "bad" stick, XMP -

Lowest Error Address	0xD5B26248 (3419MB)
Highest Error Address	0xD5B3E228 (3419MB)
Bits in Error Mask	00000000D66A706D
Bits in Error	17
Max Contiguous Errors	1

2019-12-27 14:47:11 - [Data Error] Test: 6, CPU: 1, Address: D5B3E228, Expected: 00000000, Actual: D66A706D

2019-12-27 14:47:11 - [Data Error] Test: 6, CPU: 1, Address: D5B26248, Expected: 00000000, Actual: D66A706D

2nd pair, dual channel XMP -

Lowest Error Address	0xD5B20140 (3419MB)
Highest Error Address	0xD5B38120 (3419MB)
Bits in Error Mask	00000000D5D1F000
Bits in Error	13
Max Contiguous Errors	1

2019-12-30 02:42:08 - [Data Error] Test: 6, CPU: 1, Address: D5B38120, Expected: 00000000, Actual: D5D1F000

2019-12-30 02:42:08 - [Data Error] Test: 6, CPU: 1, Address: D5B20140, Expected: 00000000, Actual: D5D1F000

2nd pair, dual channel, Default -

Lowest Error Address	0xD5B20378 (3419MB)
Highest Error Address	0xD5B38358 (3419MB)
Bits in Error Mask	00000000D664512A
Bits in Error	14
Max Contiguous Errors	1

2019-12-30 07:15:01 - [Data Error] Test: 6, CPU: 1, Address: D5B38358, Expected: 00000000, Actual: D664512A

2019-12-30 07:15:01 - [Data Error] Test: 6, CPU: 1, Address: D5B20378, Expected: 00000000, Actual: D664512A

[David (PassMark)]

Bits in Error Mask 00000000FFFFFFFF
Bits in Error 32

This is not typical for a RAM hardware failure.
Most RAM failures are just 1 or 2 bits. Not 32 bits in error.

it isn't likely that bad installation of OS or sudden loss of power would damage RAM, is it?

Extremely unlikely that a power failure is going to lead to hardware going bad.
Exception to this is when the power failure was caused by a local lighting strike.

Also having the memory address not move around as you switch from single channel is strange. This also implies it isn't a RAM hardware issue as it stays in the 3419MB range. The interleaving addresses in dual channel mode should move the bad address if it was bad RAM.

So I'm now wondering if it is a BIOS issue. That some other hardware has mapped itself into the memory space, but not flagged the RAM block as in use. Which leads to a situation where the hardware (e.g. mouse, WiFi card, video capture card, etc..) is writing and reading from the hardware, but the O/S is using the same memory addresses for main RAM. So they corrupt each other's storage. If this was the case it is a BIOS bug and there is nothing you can do to fix it (except find and permanently disable the offending hardware, which isn't easy).

[PanPhoenix]

Thanks David. That's interesting.

If it's indeed BIOS, do you think updating drivers would help at all? I have WiFi card, Bluetooth peripheral, etc. attached to system, that seem to work fine and I only occasionally use, so I didn't bother installing the latest drivers which were just out earlier this month on MOBO website. I did update BIOS to latest though.

In between posts I was fiddling around, and found that if I cap CPU's max freq. (it seems to auto-overclock itself even idle) and increase CPU core voltage, the system is noticeably more stabilized. This is with the default memory speed though(2nd pair), and I haven't ventured into bringing back XMP yet. Do you think that might be a factor in the test results? Or just separate issues to pan out since it's a new build after all.