In theory 8 bits on 128 is way better than 2 per 16 a it allows to use a better error correction code. This all due to Shannon theorem on noisy channels. The p

In theory 8 bits on 128 is way better than 2 per 16 a it allows to use a better error correction code. This all due to Shannon theorem on noisy channels. The problem is that this assumes that a probability of flipping individual bit is independent from others. But this may not be the case. And if so rowhummer is possible.