Interesting mem issues pass/fail depending on order or parallel

[memoryerror]

I have an interesting problem where memtest fails with all 4 sticks provided that it is run in parallel mode (but only on CPU 0), in single mode it passes. I also note that if I have modules with serial numbers ending in 3, 5 in one order memtest passes in parallel mode while in the other order mem test fails. The memory has been tested on some other (slow rigs) and passes. Help?

AMD 5950x
G.Skill Ripjaws v 128GB ddr4 3200Mhz Cl16 memory 4x32GB sticks
gigabyte aero g 570s F3d BIOS

For all of these tests XMP is enabled (I hope, the store I bought the memory from did the tests on the 3300x and 5800x).

The memory itself has been tests on the following and passes:

Motherboard	CPU	Test type	memory speed MB/s	Modules tested at a time	Pass/fail	Links
ROG STRIX X570-I GAMING	3300X	Parallel (All CPUs)	17786	2	Pass	Set 1 Set 2
	Ryzen 7 5800x	Parallel (All CPUs)	38805	4	Pass	results
x570s aero g	Ryzen 9 5950x	Parallel (All CPUs)	40535	2	Fail	results	Modules order 3 then 5
x570s aero g	Ryzen 9 5950x	Parallel (All CPUs)	40422	2	Pass	Results	Modules order 5 then 3
x570s aero g	Ryzen 9 5950x	Parallel (All CPUs)	41551	4	Failed	Results	Test 6 CPU 0
x570s aero g	Ryzen 9 5950x	Single: CPU # 0	41551	4	Pass	Results
x570s aero g	Ryzen 9 5950x	Single: CPU # 0	41551	4	Pass	Results
x570s aero g	Ryzen 9 5950x	Parallel (All CPUs)	41551	4	Fail	Results	Much older run

Is the memory at fault here or is it perhaps the CPU or motherboard?

What other tests should I run to work this out?

[Simon (PassMark)]

Could any of the three, probably try replacing the memory first?

There are some troubleshooting steps here
https://www.memtest86.com/troubleshooting.htm

[memoryerror]

It looks like when running test 6 only, only memory module with SN ending in 3 fails. The other 4 pass test 6. Interestingly module ending in 3 failed on the 3rd iteration.

[memoryerror]

i notice that it is always CPU 0 that catches the issues, but the issues only show under parallel mode. Is passmark choosing the section of ram to test randomly?

[David (PassMark)]

From version 9.0, after each pass we reverse the order of the memory regions allocated to each CPU core.
So if a single bit/byte of RAM is bad AND it is detected as bad in each pass, then the CPU cores should not be the same from one pass to the next.