Hammer Test Errors on Memtest86

Ayahuasca · 28 Mar 2015 at 20:32

Samsung Green 2 x 4GB sticks.

Just ran this as I've been getting issues with my PC freezing and restarting itself randomly.

It passes all other tests bar Test 13 Hammer Test with 11 errors. I've read online that these errors can be down to the Row Hammer Bug.

Anyone know if I need new RAM or can these errors be ignored or fixed with timings?

Borealis · 29 Mar 2015 at 02:28

If an error occurs with the hammer test, then (according to memtest themselves) it suggests that there is an actual problem with the RAM. Bear in mind though, that the hammer test is a worst case scenario and floods the RAM with data.

If your machine is randomly crashing/rebooting itself, then it may well mean that there is an issue.

Ayahuasca · 29 Mar 2015 at 13:15

Ran Prime95 blend test for 8 hours and everything was fine.

Starting to think the issue could be the Mobo, PSU or SSD.

Ayahuasca · 30 Mar 2015 at 13:21

Ran memtest86+ overnight and that had 0 errors :/

Borealis · 30 Mar 2015 at 15:33

What PSU is it? Did you reseat the RAM? It may just have not been quite seated right.

Ayahuasca · 30 Mar 2015 at 15:38

1200w SuperFlower Leadex Platinum

I didn't reseat the RAM, I think it's because Memtest86+ doesn't do a hammer test?

Borealis · 30 Mar 2015 at 16:07

Have you got any different RAM you could try (without buying any more)? The hammer test is a worst-case scenario that is meant to stress the ram to the limit. I take it you're not getting BSODs?

Rorin · 18 Apr 2015 at 11:10

The Hammer Test was designed to reveal a newly (relatively speaking) discovered bug that is a hardware flaw inherent in the design of some RAM sticks.

Mainly from the 2012 - 2013 era.

The test seems to only be relevant in academic cases where the RAM is repeatedly accessed in a specific row.

In normal software it seems, that the bug does not manifest itself.

It is more or less only a security issue if someone has access to your PC.

No amount of reseating or exchanging the PSU will remedy this bug.

You can try to lower the tRef value by half (e.g. 3820 to 1910). This should reduce or eliminate the vulnerability, but at a small cost to performance.

If you absolutely can not live with the bug, your only option is to buy new RAM, preferably recently manufactured, because the DRAM vendors have since the discovery of the bug used techniques to make the RAM "hammer proof"