[HN Gopher] The Myth of RAM (2014)
       ___________________________________________________________________
        
       The Myth of RAM (2014)
        
       Author : jchook
       Score  : 39 points
       Date   : 2023-11-19 21:04 UTC (1 hours ago)
        
 (HTM) web link (www.ilikebigbits.com)
 (TXT) w3m dump (www.ilikebigbits.com)
        
       | H8crilA wrote:
       | This also applies when you scale up to the multi petabyte RAM
       | range via distributed computing/cloud. I wonder if we could
       | actually continue the (very rough) linear fit on a log-log plot.
        
       | dist-epoch wrote:
       | It's even worse on GPUs.
       | 
       | There memory access speed depends not only on N as described in
       | this article, but also on the order in which you access the
       | little bits of N, like CPU cache locality, but also different in
       | important ways.
        
       | kristianp wrote:
       | There's a good discussion from 2016 here:
       | https://news.ycombinator.com/item?id=12383012
        
         | dang wrote:
         | Thanks! Macroexpanded:
         | 
         |  _The Myth of RAM (2014)_ -
         | https://news.ycombinator.com/item?id=12383012 - Aug 2016 (277
         | comments)
        
       | blagie wrote:
       | FYI: This article sucks (in relative terms) if you read Part I in
       | isolation. Click on Part II to get to the really good parts. From
       | there on, it gets weaker. If you're busy, you can skip / skim
       | parts III and IV.
       | 
       | My hope is this post saves someone some time.
       | 
       | Part 1: Empirical. Good.
       | 
       | Part 2: Theoretical limit. Really motivates Part 1 and gives an
       | interesting perspective.
       | 
       | Part 3: Implications. Weaker overall.
       | 
       | Part 4: Clarifications / definitions / etc. Only helpful if you
       | have questions or didn't understand something.
       | 
       | Missing: Showing best-case is O(log N) and not O(N) simply due to
       | addressing. If I want to address a petabyte, I need 50 bits, and
       | a kilobyte, only 10. That's also good theoretical limit as to why
       | it's not O(N), although not nearly as nice as the authors'.
        
       ___________________________________________________________________
       (page generated 2023-11-19 23:00 UTC)