3

Is it possible to get DDR3 memory that's not affected by rowhammer? ECC is reportedly not a full fix and not viable since the boards I'm looking at don't support ECC anyway. I've seen claims that, because the issue is density, smaller modules (possibly 512MB, 1GB, or 2GB) are less likely to be affected, but no actual guidelines for reliably purchasing rowhammer-immune ram. For example this paper suggests that at least some brands are not affected in small sizes:

http://users.ece.cmu.edu/~yoonguk/papers/kim-isca14.pdf

Does anyone have actual reliable information for buyers?

7
  • The "cure" for rowhammer is not the RAM itself, but the dram refresh rate. Set it to smaller intervals to mitigate possibilities of rowhammer striking. This means that your bios/firmware/spd needs to have this ability. Enabling it will cost a bit performance though. – PlasmaHH Oct 30 '15 at 15:58
  • @PlasmaHH: Based on the paper I linked to, the degree of the effect varies by ram module used, so it seems reasonable to approach the problem from both sides, especially since it's not clear to me that adjusting the refresh interval always provides a complete mitigation. If you claim it does, do you have any info on how to determine whether particular hardware has the capability to adjust the interval appropriately? – R.. GitHub STOP HELPING ICE Oct 31 '15 at 3:54
  • I don't have any papers/links at hand, but this is just how dram and rowhammer works, electrically. While surely some ram is more supectible than other, for every ram you will have refresh rates where there is a reasonable possibility of rowhammer to succeed. In the same way, for every ram you will have a refresh rate where the probability falls under an acceptable threshold, but there is always a possibility, you can not completely get rid of it. – PlasmaHH Nov 1 '15 at 20:00
  • @PlasmaHH: Well do you have any useful info on how to determine if a board allows setting the refresh rate (and if so, the range of settings)? Or anything else that could be turned into practical buying advice? – R.. GitHub STOP HELPING ICE Nov 1 '15 at 21:12
  • 1
    I will say that given my understanding of the problem's cause and the nature of memory marketing, I'm inclined to be highly skeptical of claims that they have this one solved. EMI is a devilishly hard problem to overcome. I used to work at a place that installed robots next to particle accelerators. You can guess they found out real fast about memory errors of all kinds... – Adam Wykes Aug 24 '16 at 15:52

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.