Dear Albert:
Re: sweet spot observations
Yes we can not directly observe the sweet spot. We can get close by using the following method. First, wait until machines with the CPUs show up in either retail stores or Pricewatch. Other good way is to look at inventory at those vendors or distributors that report it on a website or by phone (this is typically the closest you can get to the best information). The first one yields a bin speed distribution by looking at their inventory (or adding all of each speed up works too) in all stores visited. The second one uses the assumption that there is a linear correlation between the number of vendors who list a speed grade to the relative number of CPUs available (this has been shown to be fairly close for our purposes by past history (Rule of Thumb)).
Now, once you have the bin distributions, graph the results for one given FSB by first placing points on the X axis as the speed bins given (700, 750, 800, 850, 900, 950, and 1000 for Tbird). Then calculate the height of each column as the number of processors divided by the MHz between this bin and the next highest speed bin on the Y axis and place a rectangle in that interval of the computed height.
Now typically this looks like a Gaussian Distribution Curve, sometimes called the Normal Distribution Curve. There are programs to do this fitting of the Gaussian to your graph, but you can "eyeball" it fairly close. The peak of the Gaussian is the very "Sweet Spot" in my definition.
Now, if the lowest bin has the most CPUs, you can see that the actual sweet spot will be lower than that point if the delta is highest between it and the next higher bin than any other adjacent ones. The same is true in the other direction.
By using this model, we can see if the processor is being pushed too far, too little, or about right. We can also track this over time to get a feel for future speeds. Another way for this to be used, is to get a rough idea of the margins being generated.
I hope this clears up your confusion. One of the posters on the original thread (I am sorry I forgot who (since they have not done it recently)) used to come up with a weekly look at this. Many readers, including myself, appreciated this information as it was used to settle many arguments over this very topic. A quick peek at Pricewatch still shows that P3 has a sweet spot around 700 to 750 and Tbird is around 850 to 900.
Pete |
