On 5/15/07, <b class="gmail_sendername">Eugen Leitl</b> <<a href="mailto:eugen@leitl.org">eugen@leitl.org</a>> wrote:<div><span class="gmail_quote"></span><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
No, it isn't. Read the original paper.</blockquote><div><br>I'm well aware of what the original paper said. My point stands: if increased integration density didn't result in increased performance, nobody would bother referring to that paper, let alone enshrining said reference in the vernacular.
<br></div><br><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">Computer performance, as measured in benchmarks, does not result<br>in linear semi-log plot.
</blockquote><div><br>Yes it does. <br></div><br><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">Case in point: the difference between CPU speed, as measured
<br>by benchmarks, and memory bandwidth, as measured by benchmarks,<br>is a linear semi-log plot.<br><br>But you knew that already.</blockquote><div><br>Well yes. And each of the components are (approximately) linear semi-log. As is RAM capacity. As is disk capacity. As is the bandwidth of a fixed amount of memory (as it migrates from disk to RAM to cache - a steeper semi-log than the bandwidth of main memory). Stepping back, we see it's actually a pyramid, with disk at the base and registers at the peak; the height of the peak (speed of operation on registers), the width of the base (disk capacity) and the width of the second layer (RAM) are all best approximated as linear semi-log.
<br></div></div>