Originally posted by: Malladine
Can you explain further? How exactly is a PPU designed to process the data involved with realtime physics better than a CPU programmed to do so, beyond the fact that the PPU is marketed as such?
Try this example.
You have 50 kids to get from one city to another city 100 miles away by a highway. Everyone has a car (CPU in this analogy). The car can fit a driver + 5 kids (remember the car still must be very fast and it must fit in house garages, so it is limited in size). There are various speeds of cars (MHz of CPU). But even the fastest car is limited -lets say at 200 mph for this example. In one hour, the car travelling 200 mph can go there and back one time delivering 5 kids. So to deliver all 50 kids, you need 10 trips and 10 hours.
Car manufacturers come along and develop dual compartment cars. You can now hold a massive 10 kids at once! Speed is only minimally effected if you have a good enough car driver (programming). Thus now you can deliver 10 kids an hour, or all 50 kids in 5 hours. Things are great.
But wait, someone develops a new product called a bus (PPU in this analogy). The bus is designed from the beginning to do one task well - move many kids at once. This bus cannot win a race and only travels at 50 mph. The bus cannot park in a normal parking lot like a CPU can. The bus cannot fit in a garage like a CPU can. A bus cannot go to a drive through ATM. The bus has many limitations. Very few people have a bus. But it has the ability to carry 50 kids at once since the designers did not focus on those other aspects (remember everyone has cars for those tasks). So in two hours, this bus takes all 50 kids from one city to another. Buying the bus is far better than using two cars (or a dual car).
Ideally, I'd like to see a dual car/bus. Where you can go 200 mph while carrying 50 kids and still park in a standard size garage. That is the possibility of a PPU on a multiple core chip.
.
Facebook
Twitter