Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Technology Stocks : Advanced Micro Devices - Moderated (AMD) -- Ignore unavailable to you. Want to Upgrade?

To: dougSF30 who wrote (226567)	2/21/2007 3:52:36 PM
From: TimF	Read Replies (1) \| Respond to of 275872

The problem in your scenarios is including the bogus benchmark. I do think if your going to display a geometric average you should display all of the component benchmarks just as prominently, but AMD did that in the PDF. Its wasn't just a big display of that one average favoring AMD, with no details (or details in very small type in a footnote or appendix) of the constituent benchmarks that didn't favor AMD. AMD also showed a couple of averaged benchmarks that did favor Intel. I'm not saying there is no bias at all, AMD did spin the data by their benchmark selection, but not to the point of being crooked, and not in a way that is very different from what Intel has done in the past.

To: dougSF30 who wrote (226567)	2/21/2007 6:45:48 PM
From: Petz	Read Replies (3) \| Respond to of 275872

Geometric mean is the STANDARD way of measuring CPU performance. It is used by SPEC.ORG for all of the SPEC benchmarks, it is exactly how every benchmark suite in existence measures their averages (Ziff Davis WinMarks, PCWorld Index, SPECint, SPECfp, SPECint_rate, SPECfp_rate, Sysmarks, etc., etc., etc. They all use geometric averages to create their scores. That is the only valid way to combine multiple benchmark scores. AMD correctly points out that when some benchmarks measure runtime instead of its inverse, those benchmark times have to be inverted (1/x) before being put into the geometric mean product. You obviously don't know squat about benchmarking, since the use of geometric means is one of the most basic facts that even a benchmarking novice has to understand. And I applaud AMD for using the inverse of a famous Intel multitasking "cheat," something that is still used by Intel to increase its "multitasking" score in Sysmark Office Productivity. You see, Sysmark 2004 includes tests where several programs are started simultaneously, but only the progress of the office task is actually measured. For example, they start a virus scan and then measure the time it takes to do some WORD replacement operations. For some reason, AMD processors have always spent more of their resources doing the virus scan than the Intel processors do, but the fact that they complete the virus scan twice as fast isn't even measured. Instead, the replacement operations take twice as long. So the fact that Intel gets double the score of AMD on that benchmark skews the entire Sysmark score. If there's 10 benchmarks, a 2:1 anomaly would result in a 7% skewing of the Sysmark score. I don't recall how many benchmarks there actually are. I think there is another Sysmark benchmark in the Internet series that runs WME 9 simultaneously with something else and only measures WME 9. Intel processors tend to really like to run WME 9 and ignore the other tasks, so that is another win for Intel in the benchmark result. Is the City of Villians benchmark anomalous? Not to someone who's trying to recode a movie while they play Villians. The C2D is obviously glomming on to WME 9 and deprioritizing the game. Or maybe its the network stuff built into the AMD chipset that unblocks the CPU for online gaming. By the way, a 2.5:1 anomaly with 20 benchmark scores only skews the result by 2.5^.05 or 4.7%. Petz