Comments on: What Drives Performance in HPC? http://www.linux-mag.com/id/4170/ Open Source, Open Standards Sun, 23 Nov 2008 10:25:59 +0000 http://wordpress.org/?v=2.0.11 by: atekhasski http://www.linux-mag.com/id/4170/#comment-1166 Wed, 11 Jun 2008 05:18:07 +0000 http://www.linux-mag.com/id/4170/#comment-1166 Wow! Thank you, Ted, for the token of recognition! As I see, P.Emma [1] has expressed similar sentiment (to which I arrived independently and two years later :-( ): <i>"First, the most popular performance metric, IPC (instructions per cycle), is the reciprocal of the metric that should be used, CPI (cycles per instruction). This is primarily because CPI is a simple dot product of a few numbers that any experienced designer should have at his fingertips. It is intuitive, and it makes for remarkably quick and remarkably accurate estimates."</i>, http://www.research.ibm.com/journal/rd/413/emma.pdf (it's easier to read). For those who might be interested, the original Powerpoint of the 1999 SPEC Workshop presentation can found here: http://home.austin.rr.com/bah/SPECworkshop/frame.htm There is a populistic paper on the subject on my website http://home.austin.rr.com/bah/index.htm and also several examples which were intended to illustrate usefulness of the scaling approach in "time-domain" coordinates. Thanks again for the nice article. Wow! Thank you, Ted, for the token of recognition! As I see, P.Emma [1] has expressed similar sentiment (to which I arrived independently and two years later :-( ):

“First, the most popular performance metric, IPC (instructions per cycle), is the reciprocal of the metric that should be used, CPI (cycles per instruction). This is primarily because CPI is a simple dot product of a few numbers that any experienced designer should have at his fingertips. It is intuitive, and it makes for remarkably quick and remarkably accurate estimates.”,
http://www.research.ibm.com/journal/rd/413/emma.pdf
(it’s easier to read).

For those who might be interested, the original Powerpoint of the 1999 SPEC Workshop presentation can found here: http://home.austin.rr.com/bah/SPECworkshop/frame.htm
There is a populistic paper on the subject on my website http://home.austin.rr.com/bah/index.htm
and also several examples which were intended to illustrate usefulness of the scaling approach in “time-domain” coordinates.
Thanks again for the nice article.

]]>