Date: Sat, 13 Mar 2004 21:11:28 -0800 From: David Schultz <das@FreeBSD.ORG> To: "Daniel O'Connor" <doconnor@gsoft.com.au> Cc: freebsd-chat@FreeBSD.ORG Subject: Re: What makes Centrino so fast? Message-ID: <20040314051128.GA57404@VARK.homeunix.com> In-Reply-To: <200403111155.39591.doconnor@gsoft.com.au> References: <404FAC50.6070603@pythonemproject.com> <200403111155.39591.doconnor@gsoft.com.au>
next in thread | previous in thread | raw e-mail | index | archive | help
On Thu, Mar 11, 2004, Daniel O'Connor wrote: > On Thu, 11 Mar 2004 10:31, Rob wrote: > > I have my own benchmark program that I use for number crunching. > > It uses a mixture of Python and Numeric Python. > > > > Athlon 2166Mhz as reported by OS: 2m38.7s > > Intel Centrino Pentium 1700 Mhz: 2m17s > > > > Is it just compiler optimization at play? > > Here's an interesting article on it -> > http://arstechnica.com/cpu/004/pentium-m/pentium-m-1.html > > but in short.. better branch prediction, and micro-architecture improvments in > general, and a slightly longer pipeline (for higher clocks vs a PIII) You're right that the longer pipeline allows the processor to be clocked higher, but for a *given* speed (e.g. 1700 MHz), a longer pipeline is actyually a disadvantage; longer pipelines cause more stalls and higher branch misprediction costs. The better branch predition merely attempts to hide the penalty of the longer pipeline. I don't know why the Centrino performs better than the Athlon in this case, though. If you really care, you'll probably have to factor the benchmark into specific, simple tests that demonstrate the performance difference, play with compiler optimizations, etc.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20040314051128.GA57404>