Maybe these chips aren't quite as complex as a CPU, but they are in about the same ballpark as far as complexity goes.
Not only they are not in the same ball park, they are not even in the same sport. After feature size, the single most important determinant of cost is the instruction set. The less instructions and the more ability to do parallel executions, the cheaper the cost. This is why a memory chip will always be the cheapest chip around. It has only two instructions: read this, and write that. And that means that the chip is essential one cell, replicated to the Nth degree. Now back to graphics chips, they are orders of magnitude more complex than a memory chip, but nothing close to CPUs because essentially you are replicating the same set of cells over and over to achieve parallelism. The cost of the chip is defined based on these factors: (1) feature size, (2) instruction set, (3) number of gates. Of the three, the instruction set is the most important one *if* (1) and (3) can compensate each other. BTW, this is why the RISC design is so successful. I hope this explains why TDFX can make money at these prices.
Sun Tzu
P.S. For the first time in quite a while, I'm actually excited about a company I own. May be I should close my position until the feeling goes away :^) |
