22

I know that since ~2004, Moore's law stopped working for CPU clock speed. I'm looking for a graph showing this, but am unable to find it: most charts out there show the transistor count or the capacity per year.

Where can I find some data showing the CPU frequency of computers (anything is fine, personal computers, servers, laptops, ...) from the last few decades to today?
Raw data that I can plot myself would be fine as well (hum, probably even better).

Raphael
  • 72,336
  • 29
  • 179
  • 389
peoro
  • 323
  • 4
  • 8
  • Go to any list of CPUs and extract the data yourself? – Raphael Mar 21 '12 at 12:50
  • 7
    You are misunderstanding Moore's law. The law simply states that "the number of transistors that can be placed inexpensively on an integrated circuit doubles approximately every two years." This may have a side effect of processing power speed, but has absolutely nothing to do with CPU frequency, cores-per-chip, cache-sizes, etc. It's simply states that your transistor budget will more or less double every two years. – Timothy Baldridge Mar 21 '12 at 13:24
  • 1
    @TimothyBaldridge: Right, but the law has been used for other quantities with the same name. They are all wild speculation, anyway. – Raphael Mar 21 '12 at 14:22
  • 7
    please keep in mind that CPU frequency is a *terrible* predictor of CPU performance. There are so many other things that matter. (super scalers, instruction level parallelism, register renaming, instruction reordering, for a start) – user606723 Mar 21 '12 at 15:12
  • 1
    there has been a long handwaving correspondence drawn between clock speed and transistor count [the classic moore law] up to the mid 00's that was roughly empirically valid. that has started to diverge. however seriously one wonders if transistor count has started to plateau also. – vzn Mar 23 '12 at 18:25
  • CPU Frequency Increases were due to Dennard Scaling, which broke down ~2006. Moore's law was limping along until just recently. – Mark Johnson Aug 28 '18 at 19:51

4 Answers4

10

The graphs you want are printed in the book “CMOS VLSI Design” Weste/Harris (2005), on page 5 of the freely available introduction.

Tom Medley
  • 209
  • 2
  • 5
8

Bob Warfield wrote a blog post (A Picture of the MultiCore Crisis) that has the kind of graph you're looking for.

Update

It sounds like the Stanford CPU DB has what you're looking for. You can browse visualizations and download the raw data.

Mark Johnson
  • 195
  • 4
  • The CPU DB is just perfect. It looks like they even keep it up to date: here the CPU frequency graph: http://cpudb.stanford.edu/visualize/clock_frequency – peoro Apr 11 '15 at 03:42
4

See Sutter, The Free Lunch is Over (2005): it has graphs, discusses how various processor characteristics relate to performance of actual programs, and makes some predictions that turned out true, as far as I can tell.

rgrig
  • 1,336
  • 1
  • 11
  • 15
3

The 5th edition of Computer Architecture, A Quantitative Approach has several related graphs.

The clock rate graphs is divided in 3 parts, from a 5MHz Vax in 78 to a Sun SPARC at 16 MHz in 86 (15% per year), to an Intel Pentium 4 Xeon in 2003 at 3200 MHz(40% per year) to an Intel Nehalem Xeon in 2010 at 3330 MHz (1% per year).

The performance graph is also divided in 3 parts, from the VAX 11/780 in 78 at relative performance 1 to the VAX 8700 in 86 at relative perf of 5 (25% per year), to an Intel Xeon in 2003 at relative perf of 6000 (52% per year) to an Intel Xeon in 2010 at relative perf of 2400 (22% per year). Note that SPEC on which the performance measure is based has automatic parallelization turned on and so multiple core of a processor are used here.

AProgrammer
  • 3,059
  • 17
  • 20