on May 2, 2011
I’ve brought up the story of the crippled Intel processor a few times now. I brought it up in my recent post on the Portal 2 DLC and managed to thread-jack my own topic. So, I scored an own goal with that one. Nice.
If you missed it, here is the setup:
Back in the early days of PC processors, Intel used to use one factory to make all of their processors. Their expensive high-end processors were identical to the crap bargain stuff they sold, right up until the bargain ones rolled off the assembly line and they deliberately destroyed the math co-processor. It turns out it was cheaper to just make the good ones than to have two different factories. Then they would sell the crippled one to regular folks and the normal one to rich people. Some people were outraged at this. “Why did they ruin this processor!? It would have cost them NOTHING to give me the fully functional version!
Some people complain that this is an incredibly stupid system, or that it is unfair to the consumer. When I found out about this in my twenties (back in the 90’s) it drove me crazy. I suddenly knew that the CPU in my bargain computer had, for a fleeting moment, been a deluxe powerhouse until some jackass ruined it. On purpose. As an engineer, I dislike this sort of destruction.
But it’s actually not stupid or anti-consumer. It’s just really, really counter-intuitive. It’s a wonderful system that fueled the manufacturing explosion that gave us all of these amazingly cheap computers. Here is how it works:
Warning: These numbers are entirely fabricated for the purposes of this demonstration. Please focus on the system and do not nit-pick the numbers. Thank you.
It takes a lot of money to design a chip and set up a production pipeline around it. Let’s say ten million bucks. That pays the eggheads to design it, someone to figure how to construct it, and the money required to build the facilities to make the dang things. On top of that million bucks, each chip costs about $35 in parts and labor.
So if we only make one processor, we need to sell it for $10,000,035 if we want to break even. That one processor needs to recoup the entire set-up fee for the whole facility. But if we sell a million of them, then we can spread that ten million dollar set-up fee over a million units. Now we only need to charge $45 for each one to break even. I’m sure most of you are familiar with this idea. Economies of Scale are fairly common and apply to everything from cars to hamburgers.
(This gets more tricky when you realize that the producer doesn’t actually know how many they will sell. If they charge $45 each for processors and end up only selling half a million, then they are well and truly screwed and will go out of business. So, they need to speculate a bit. What is the minimum number we can expect to sell? Half a million? A quarter million? We need to set our price so that we can survive even if sales are lower than anticipated. It’s all speculation. If you over-estimate sales you’ll go out of business. If you under-estimate you’ll be charging too much and someone else will undercut you, which could also put you out of business. Have fun!)
Now, we’re going to assume you have perfect knowledge of the market. You know that there are a million people out there who will pay $50 for any damn processor. Let’s call these people Bargain Users. They just need something to make the computer go. You can sell your $45 chip to these people and make $5 apiece. That’s five million bucks in profit. You also know that there are about a hundred thousand people out there who will pay $100 for a processor, but these people want the best. Let’s call them Power Users. They’re simulation or graphics professionals and they want their computer to go as fast as possible. Time is money for them, and so they are willing to spend a lot of money to save themselves some time. If you sell them a processor for $100, you will make $55 each. That’s five and a half million dollars. There is ten and a half million bucks to be made here, but only if you can sell to both groups.
If you sell all processors for $100, those million bargain users won’t buy. That’s too much for them. On the other hand, those hundred thousand Power Users aren’t just going to give you $100 out of the goodness of their hearts. If you try to sell the same processor to both groups, the Power Users will buy the $45 version because they’re not idiots.
So, we need to make a less-powerful processor to sell to the bargain users. Let’s talk about some solutions:
1. Design a slow processor to sell to the bargain market.
Now, you could just make a less powerful processor, but the cost of re-configuring the production facility makes this kind of expensive. If you were selling clothing, you would use higher quality raw materials. Cheap jackets are inexpensive polyester, and quality ones are a cotton/wool blend. But we’re making processors, which are all made from metal, silicon, and plastic. A processor made in 1990 has basically the same ingredients as one made in 2010. Likewise, the labor costs are the same. It costs the same to hire a guy to run the 1990 assembly line as it does to hire a guy to run the 2010 assembly line. (Ignoring inflation, obviously.)
The last thing you want to do is re-configure your assembly line. That gets expensive. You hammer out good processors. Then you re-configure the whole thing at the cost of thousands of dollars to make the weaker ones.
It’s like books. The same printing press can hammer out copies of Lord of the Rings or a similar-sized compilation of Smurfs / Trek crossover fanfiction. The cost isn’t what you’re printing, but in how many you can print at once. Changing print jobs is the expense you want to avoid.
2. Design the factory to be easily switchable from one state to another.
This would increase your setup costs, which is your single largest and most dangerous expense. It makes the system more complex to run, which can lead to expensive Human Errors.
3. Choose between the two markets.
You’ve only got one factory, so you can only make one chip, which means you have to choose between the bargain users and the power users. It’s best to go for the power users in this case. They’re more profitable, since your magin with them is much better. They’re also less risky, because they’re more reliable as customers.
Obviously this leaves bargain users out in the cold and keeps computers as a niche product only available to the elite.
4. Turn some of the processors into an inferior product after production.
Take those powerful CPU’s rolling off the line, break their math co-processor, and sell it to bargain users for $50. It turns out that this is the best solution that gives the cheapest hardware to the greatest number of people. This is the most efficient way of serving both markets.
You can think of this as an alternate way of going about solution #1. You are producing lesser chips. This is simply the most cost-effective way of going about it, since it lets you use the existing facilities and production pipeline. Rather than design a less-powerful chip and a configure the production line to make them, just take the good ones and convert them.
It’s strange, but it works.
Shamus Young is an old-school OpenGL programmer, author, and composer. He runs this site and if anything is broken you should probably blame him.