Beating the System: The AnandTech Guide to Economic Upgrading

Putting it all together

So far, we showed how easy it was to build a model with retrospect data and how easy it was to build a model with made up continuous data. The goal now is to build a hybrid model from the previous two. Below is a list of hardware from the previous example, but using continuous devaluation overrides to "predict" the cost of each product after every four weeks.

Again, continuous functions are bad for attempting to predict hardware pricing trends because pricing models for most components are structured on a discrete stair step algorithm using various price points. For those of you familiar with our pricing guides or the RTPE, you'll know that we watch a lot of prices at any one time. Purav, Adul and I can recite the exact distributor MRSP on every Intel processor at any given moment. We notice specific trends that can be readily identified with our historical data in the Real Time Engine (the graphs). Generally, after a new product has stabilized in price a little bit, a new processor, motherboard or video card will continue to devalue about 1.67% per month until the products retail EOL. Low demand products that hit EOL cling to their MSRP for a while (thus 0% devaluation) while high demand products will actually start to increase in price due to supply and demand (usually less than 1% per month). The Radeon 9800 Pro is actually a very good example of a video card that went through all three phases during its lifetime as a retail product. When all else fails, be a little too conservative with your estimates; the only thing that the devaluation should affect is whether or not it makes sense to upgrade today.

With some quick calculations, we can generate the estimated prices for the next six months based on today's hardware prices and the devaluation overrides that we just discussed above. The resulting graph looks as such:

Now, we can introduce our CNU value of $0.25 per day again and replot the graph.

It should be pretty clear which hardware not to buy at this point; buying a GeForce 5900XT six months from now (if you can still find one) will cost you double, since you are paying for the retailers to mark up the price on their remaining stock even before taking into consideration the CNU. Adding the relative Quality for each product gives us the final plot below.

Click to enlarge.

Just based on mathematical formulas and some educated assumptions for relative Quality, the GeForce 6600GT comes out on top again with the Radeon 9800 Pro a mildly distant second. Upgrading the processor instead of the graphics card might actually make sense in this scenario (after the 6600GT) with the weights that we've given based on HL2 CPU and GPU scaling. All of the data from this page can be viewed in the second excel workbook from the spreadsheet available here.