Tweet
David,
When you first implemented Rowhammer testing, you followed Kim's hammering strategy to some degree, but later you watered down the test saying that you wanted the second pass to only be 200k activations per 64 ms retention period. You furthermore stated that "200K per 64ms, as determined by memory vendors as the worst case scenario". Out of curiosity, which vendors told you this? I would like to know as you were either lied to or you made this up. Currently, for example, DDR4 at 2400 MHz is theoretically capable of over 1 million activations per retention cycle, so your "worst case" is only wrong by a factor of 5. And as there are published papers showing over 500k, in any case, your algorithms are weak and not up to real world standards.
And if you don't want to "scare" people, why have you recently taken other people's work, such as double sided hammering and using multiple data patterns to make a more stressful test case? Not only do you seem to be talking out of both sides of your mouth, you also forgot to give credit to the researchers that brought these concepts to your attention. That's bad form from an ethical point of view, but hiding computer defects is completely uncool. I guess that's why you're called Passmark. Nothing to see here, sorry that our tests passed and your database corrupted. I like how traktorkontrol calls you out for being shady. I recently ran Memtest v 7.0 on a system and saw 4 warning and one corruption on a test that took 90 minutes to complete. Using the Google diagnostic, I saw thousands. I know which one that I trust.
I'm expecting a weak reply from you, so at the very least, please respond on the topic of 200k per 64 ms and where you found that jewel. You can provide a link for some of us technical folks.
When you first implemented Rowhammer testing, you followed Kim's hammering strategy to some degree, but later you watered down the test saying that you wanted the second pass to only be 200k activations per 64 ms retention period. You furthermore stated that "200K per 64ms, as determined by memory vendors as the worst case scenario". Out of curiosity, which vendors told you this? I would like to know as you were either lied to or you made this up. Currently, for example, DDR4 at 2400 MHz is theoretically capable of over 1 million activations per retention cycle, so your "worst case" is only wrong by a factor of 5. And as there are published papers showing over 500k, in any case, your algorithms are weak and not up to real world standards.
And if you don't want to "scare" people, why have you recently taken other people's work, such as double sided hammering and using multiple data patterns to make a more stressful test case? Not only do you seem to be talking out of both sides of your mouth, you also forgot to give credit to the researchers that brought these concepts to your attention. That's bad form from an ethical point of view, but hiding computer defects is completely uncool. I guess that's why you're called Passmark. Nothing to see here, sorry that our tests passed and your database corrupted. I like how traktorkontrol calls you out for being shady. I recently ran Memtest v 7.0 on a system and saw 4 warning and one corruption on a test that took 90 minutes to complete. Using the Google diagnostic, I saw thousands. I know which one that I trust.
I'm expecting a weak reply from you, so at the very least, please respond on the topic of 200k per 64 ms and where you found that jewel. You can provide a link for some of us technical folks.
Comment