[loeg]

7ns latency is in the ballpark of small writes to L1 cache. I.e. some sort of in-mem only append to somewhere predicted by the prefetcher of like, a single cache line or less. So yeah, some sort of ringbuffer log could definitely support this kind of figure. The question is how much throughput does your persistence channel have, how much memory can you devote to logging while your persistence runs async, are you ok with a somewhat volatile log, etc.

[jnordwick]

you don't really wait on L1 cache writes though. The store buffer absorbs it, and if the data is needed it can be forwarded from there before the write to cache even happens.

Most x64 L1d caches have a 4-6 cycle latency depending on CPU, that 1 to 2 ns depending on frequency.