"RowHammer" Errors tests

[cokkopokko]

Hi to all,
I ran some tests with my DDR3 Kingston on Giga H97-HD3 bios F8 and Fury Black HX318C10FBK2/8

Dual Channel :
16x = 1600MHz = 17294 MB/s = 71 errors test 13 single scan
14x = 1400MHz = 14851 MB/s = 25 errors single scan
13x = 1333MHz = 14663 MB/s = 14 errors single scan
10x = 1066MHz = 11824 MB/s = 4 errors single scan
8x = 800MHz = 8793 MB/s = 0,8 errors single scan (17 errors on 20 scans)
(Fury but not accuracy)

instead on sigle channel :
at full speed testing a stick at a time one gives me an error the other no error. (once scan)
This is quite strange. it is possible that this is also an error of the CPU cache?
tnx.

[keith]

Are all these errors from row hammer test (Test 13)?

Please see this page regarding row hammer errors:
http://www.memtest86.com/troubleshooting.htm#hammer

It makes sense that higher clock speeds would increase the row hammer rate, thereby increasingly the likelihood of errors.

[cokkopokko]

yes only test 13 x 1 scan.

• Do nothing (too many errors)
• Replace the RAM modules (yes, but with what?)
• Use RAM modules with error-checking capabilities (my mainboard does not (officially) support this feature)

[cokkopokko]

Maybe I can solve the problem by changing timing of memories
or purchase memory 2400 MHz @ 1600 MHz or use only single channel

[cowbutt]

See my thread ( http://www.passmark.com/forum/showth...er-mitigation& ) which contains details of how to calculate tREFI and tREFIX9 settings which the BIOS uses to program the refresh interval. Start by halving the refresh interval from 7.8us to 3.9us and go from there. I had to reduce my refresh interval to 3.9us*0.75=2.925us to get rid of the last rowhammer error.