Hacker News Reader

MegaTransfers, it's the speed of the clock * ops per clock period. Typically MB/s is a bad way to measure speed of transfers due to the extreme variance between sequential & random read/writes. But for convenience you can think of 1MT/s as 1 MB/s

View on HN

> Typically MB/s is a bad way to measure speed of transfers due to the extreme variance between sequential & random read/writes

Maybe it's bad for the manufacturer. In a multitasking system is good to know what to expect from your memory when multiple programms access it.

The issue is usually that you only get the sequential number (which is the highest), and you will realistically never get that speed since you would need to only read/write a single large file at a time (like an NVR is one of the few applications where that's true). To complicate it further the number reported is with DRAM cache unsaturated, if you actually sustain that write speed for a longer period it will also slow down to less than 1/3 iirc.

Sounds dodgy and like a way to avoid peak vs. sustained and benchmarks.

Reply on news.ycombinator.com