I've spent some time working in modern factories and most of them schedule production according to actual or projected demand... I'm sure Intel is no exception. Intel sells the majority of its high speed (800+) CPUs to resellers like Dell, Gateway, etc. as OEM parts. Accordingly, Intel (like most factories) probably makes production runs of CPUs of one particular speed at a time. Suppose Gateway wants 5,000 933 MHz CPUs. They'll fab a bunch of wafers, and if yield is good, most of them will pass and be shipped as 933s. The rest will go into a "reject" bin and will be retested when a production run of slower CPUs is needed. Therefore the binning process on the OEM side catches a large percentage of the highly clockable chips and leaves a lot of not-so-hot chips to be thrown in with production runs of lower speed OEM CPUs, many of which will be of the first-run-yield variety, but many of the throw-ins are going to be real duds overclocking wise.
The retail side is different though. The average consumer is content to pay between $200 and $300 for a retail CPU upgrade, which right now equates to a CPU between 600 and 750 MHz. Intel makes lots of profit on these mid-range retail CPUs, so you bet your buns that they schedule entire runs of CPUs to be sold at retail at these aforementioned speeds. Now, these orders for retail chips are being fulfilled from the same fab line as the crate of 933s that Intel shipped off to Gateway the week before, so in all likelihood the quality and yield on the wafers is the same... the difference is that they are only testing these chips for operation at 600 to 750 MHz. It doesn't matter that many of the chips could have been rated at 933 MHz instead; if they flooded the market with 933 MHz retail chips they'd have to drop the prices considerably across the entire product line and profits would plummet. So the chips get sold as 600Es, 650s, 700s and 750s at a respectable profit, and most of them are capable of running much faster, but that doesn't matter... it's economics at work, supply and demand. They are sold as what the end user is willing to pay for.
And that's why retail CPUs are a better overclocking bet than OEM ones... as it is written in the Book of Pain, retail chip production runs contain a lot more potentially overclockable first-run chips than OEM batches do. Of course I could be dead wrong, but this scenario makes the most sense to me, and I do actually have better luck overclocking retail chips vs. OEM ones.
-Pain
Facebook
Twitter