what to check when replaced bad memory, but still have crash in like with bad memory

[ktmmotocross]

have some cards that have bad memory, but when replaced ram chip with known good ram chip card still crash

what to check next?
what is OK amount of errors in MATS ?

[Prott]

Re: what to check when replaced bad memory, but still have crash in like with bad mem

Try checking resistances of individual rails - I had cards that crashed because of low resistance vmem (few ohms). The rail still had it's voltage, but once the drivers loaded it was too much for the phase and the card crashed.

I even had a card that tested OK during memtest and then crashed in OS because of this.

[DynaxSC]

Re: what to check when replaced bad memory, but still have crash in like with bad mem

If you get memory errors, this can have w few reasons:

1. Memory chip is bad or is not really the correct replacement one (I have seen cases, where exactly same type of memory is once 1.5V and on another card 1,35V, so you need to be sure, you take replacement chip from a card with the same memory voltage as your broken card - provided this is the source of the replacement chip), or the replacement chip was not reballed/resoldered correctly (happened to me once or twice, it's not obvious but possible, so now I use a highly active solder paste, and you can also tin the pcb pads a little bit before the reflow and resoldering the mem chip, during tinning the pads you will see if there are some pads which are difficult to tin, e.g. oxidated),
2. Memory controller in the GPU is bad,
3. Connection between GPU and memory chip is bad (pcb issue: broken track, disconnection, bad via or some short),
4. The RC network aside the memory chip is broken (some element of it) or some element missing (blown off during memory replacement, or just broken off),
5. Power supply for memory chips, memory controller is present but not clean/stable or does not generate the nominal voltage.

1 and 2 are the most common problems, if the GPU mem controller is bad, this is sad, and the only way to check it is to replace the GPU. Do it only as the last choice and if you are sure that all other reasons are excluded.