ECC errors on a MacPro 5,1 with 8 slots filled

[ken]

I upgraded the memory on a MacPro 5,1 to 64GB (8x8GB) using "Hynix HMT31GR7AFR4C-H9 8GB 1333MHz ECC DDR3".

Running memtest86 the first time showed numerous ECC errors (correctable) at a single memory address with test 13 (Hammer) labeled "Ch/Sl: 2-0". Then I moved the modules that were in slots #1 and #2 into #3 and #4 (and vice versa), and the memory in slots #5 and #6 into #7 and #8 (and vice versa). After that, memtest86 shows numerous ECC errors on just about every test, not just Hammer, and at a different single address.

I don't know how to interpret this. I thought that slots #3 and #7 would correspond to channel 2, so it's confusing seeing that Ch/Sl 2-0 is still being reported. Any thoughts would be most welcome. Thanks
image widget

[ken]

This is the second test (different address, ECC errors on multiple tests, not just 13 (Hammer).

[David (PassMark)]

The channel and slot numbers reported are what we get from the memory controller in the CPU.
The software doesn't really have any knowledge of the physical locations of the slots.
And there is no real reason to expect the slot numbers reported by an Intel CPU to match the number of slots by Apple (assuming apple even gives them a number).

If one RAM stick was bad, moving it to a different slot should change the error address, assuming other slots are also in use (how much it changes depends on a bunch of complex factors, like channel interleaving).

Seems likely these are real errors (but then get corrected by the ECC function)