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.

Politics : Formerly About Advanced Micro Devices -- Ignore unavailable to you. Want to Upgrade?

To: Petz who wrote (87865)	1/18/2000 7:43:00 PM
From: Petz	Read Replies (1) \| Respond to of 1572153

Math error and analytical error in estimated benefit of full speed cache for Athlon. In the post referenced above, I made a math error in the first part. But I realize now that, even with the math error fixed, the extrapolation of performance to a full-speed cache is using the wrong model. I used the model: (1) CPUMarkScore = a + bcache_MHz The correct model is: (2) 1/CPUMarkScore = CPUMark Runtime = a + b(1/cache_MHz) Equation (1) is incorrect because it predicts an infinite CPU score if we had an infinitely fast L2 cache. Equation (2) says that as we make the cache faster and faster, the runtime will assymptotically approach a constant value, a. I've corrected the math errors in the first part of the original analysis below, the corrections are in italics. ASAP, I will do the analysis using the correct model (2). My gut feel is that this will give slightly smaller results (less improvement) than that which is shown below. Here's a quick and dirty calculation of benefits of full speed L2 for an Athlon. From tech-report.com, look at the ZD Winbench 99 CPUMark 99 score for 3 different Athlon 800 CPU scores. They all use the same Athlon, only changing the cache divisor CacheRatio...CacheSpeed...CPUMark99...CPUMark/MHz 0.33...............267....................65.6 0.40...............320....................67.0....................0.0263 0.50...............400....................70.2....................0.0400 Average increase in CPUMark99 per MHz of cache speed: 0.0345 Predicted CPUMark99 at 1:1 L2 Cache Ratio (800 MHz): 84 Worst case CPUMark99 using 0.0263 CPUMark/MHz: 79.6 Estimated % improvement vs. 2/5 cache ratio: 25.4% Worst case % improvement vs. 2/5 cache ratio: 18.8% As mentioned at the beginning, because I should be fitting the inverse of benchmark performance to the inverse of cache frequency, even an 18.8% improvement may be too high. Further analysis will follow. Petz