By: rwessel (robertwessel.delete@this.yahoo.com), October 29, 2008 11:08 am
Room: Moderated Discussions
David Kanter (dkanter@realworldtech.com) on 10/29/08 wrote:
---------------------------
>Ladies and Gentlemen,
>
>I am very happy to announce that my brother Aaron and I have written up the first
>part of a long over-due article that takes a deep dive into the performance of modern CPUs.
>
>Two years ago, we started playing with Code Analyst and VTune to gather performance
>profiling data on common gaming benchmarks. Our goal is to try and dive one level
>deeper than traditional benchmarking and quantitatively explore why Intel's Core
>2 is vastly superior to AMD's K8 for the benchmarks we selected.
>
>Our articles goes through many of the problems we dealt with (such as flaky tools,
>inconsistent results, etc.) and presents a substantial amount of data on relatively modern applications.
>
>Please take a read and use some of this information to think about how future CPUs,
>such as Nehalem and Barcelona might perform.
>
>http://realworldtech.com/page.cfm?ArticleID=RWT102808015436
In reference to Figure 9 (L1I Cache Misses per Instruction Retired (MPKI)), the following seems backwards:
"Unsurprisingly, AMD’s larger caches carry the day with an advantage of around 1 MPKI for many of the various benchmarks and configurations. The main exceptions here are Far Cry High and Fear Low, which both seem to barely favor Intel."
---------------------------
>Ladies and Gentlemen,
>
>I am very happy to announce that my brother Aaron and I have written up the first
>part of a long over-due article that takes a deep dive into the performance of modern CPUs.
>
>Two years ago, we started playing with Code Analyst and VTune to gather performance
>profiling data on common gaming benchmarks. Our goal is to try and dive one level
>deeper than traditional benchmarking and quantitatively explore why Intel's Core
>2 is vastly superior to AMD's K8 for the benchmarks we selected.
>
>Our articles goes through many of the problems we dealt with (such as flaky tools,
>inconsistent results, etc.) and presents a substantial amount of data on relatively modern applications.
>
>Please take a read and use some of this information to think about how future CPUs,
>such as Nehalem and Barcelona might perform.
>
>http://realworldtech.com/page.cfm?ArticleID=RWT102808015436
In reference to Figure 9 (L1I Cache Misses per Instruction Retired (MPKI)), the following seems backwards:
"Unsurprisingly, AMD’s larger caches carry the day with an advantage of around 1 MPKI for many of the various benchmarks and configurations. The main exceptions here are Far Cry High and Fear Low, which both seem to barely favor Intel."