1 |
On Tue, 27 Jan 2015 15:28:11 -0800 walt wrote: |
2 |
> Yesterday I installed 4GB more of RAM in this machine for a total of 8GB, and |
3 |
> the machine soon began random segfaulting and even a kernel crash or two, so |
4 |
> obviously I suspected the new RAM was faulty. |
5 |
> |
6 |
> I let memtest86+ run overnight and it found zero memory errors. Today I |
7 |
> exchanged the new RAM anyway and got a different brand this time, and |
8 |
> that fixed the problem. |
9 |
> |
10 |
> My question is why didn't memtest86+ find any errors? Could it be that the |
11 |
> first RAM I bought was actually okay but this machine didn't like it for some |
12 |
> reason? Both were DDR3/1333MHz, just from different manufacturers. |
13 |
|
14 |
As an addition to earlier posted comments: |
15 |
|
16 |
1) memtest86+ has a bit fade test which is not enabled by default |
17 |
(at least for 4.x branch which is the latest in tree now), so |
18 |
you have to enable and run it manually. IIRC it is enabled by |
19 |
default in 5.x branch (bug pending in bugzilla). By the way 5.x |
20 |
have some additional tests which may find faults unknown to 4.x |
21 |
|
22 |
2) The same frequency is not enough to guarantee memory banks |
23 |
compatibility. They may require different timings or, less probably, |
24 |
voltage. Some BIOS tuning may help here. |
25 |
|
26 |
3) Memory may be (un)buffered, (un)registered, ecc/non-ecc. Many of |
27 |
these combinations are not compatible with each other. |
28 |
|
29 |
4) In some rare cases even banks with the same parameters from |
30 |
different manufacturers are not compatible due to technological |
31 |
differences (this goes down to how logical circuits are |
32 |
implemented). |
33 |
|
34 |
Best regards, |
35 |
Andrew Savchenko |