> They claimed 10x the density of DRAM, it is now 4x
> Latency missed by 100x, yes one hundred times, on their claim of 1000x faster, 10x is now promised
> More troubling is endurance, probably the main selling point of this technology over NAND. Again the claim was a 1000x improvement, Intel delivered 1/333rd of that with 3x the endurance.
From this seminar few months back - https://www.youtube.com/watch?v=hXurTRtmfWc ,
I think density can be increased, this is only the initial product,
and latency is contributed more by PCIe/OS/application rather than the underlying 3d-xpoint material. The slides from the article are for the PCIe SSDs, I wonder whether the earlier claimed latency, still holds well with NVRAM.
I wonder why the endurance is so lower than the earlier claims.
10x latency and 3x endurance might normally satisfy the "must be 10x better" criteria to break into an existing market, but with the maturity of flash, and how memory hierarchies can ameliorate useful sets of latency requirements, this could end up being a damp squib instead of the revolution promised. 1000x endurance would have been great, 3x, who will notice?
Not the first time Intel has grossly mismanaged its technology....
Not the latency for commits to stable non-volatile storage, unless battery backed RAM is an option.
It is a bit different to say initial product shipped vs tech potential.. We've been waiting on zen err bulldozer/excavator/piledriver/steamroller for years now, and while mobile and Apus shipped, it has been a fluke in server and desktop markets.
NAND SSDs which are common today are essentially rows of similar structures. Each row is a string of floating gate transistors, and to read a single bit, you gotta run the whole operation of reading the entire row.
Random access means that you can read anything anywhere without any sort of similar caveats.
Almost certainly those nice round numbers came from some engineers (or more likely, engineering manager) being pressured for 'long term performance estimates of the technology', coming up with something that seemed plausible, and then wrapped up nicely into a marketing presentation.
Or, it might be that intel actually did have something approaching each of those three claims - in three separate embodiments of the product, and is running into problems combining the traits together.
In all cases, I feel for the Intel engineers working on this project right now. Probably all cursing that original reveal.
