| windbane said: I love how you know exactly how much developers will use the PS3 in their upcoming games. And more in response to your comment at the top of this page: developers like the challenge and once they have overcome a challenge they won't have to keep overcoming it. |
Exactly that's the question. What many people don't realize the tricks of the Cell are not new, but more or less ancient by computer standards. It has to do with a time, when transputers were seen as a big chance. After the problems behind this concepts were found (Amdahls Law)the whole computer industry changed directions.
Now more and more people realize that this change was far more drastic than necessary. The Cell was especially designed for situations, where it will not run into trouble. But if I look at how the Cell should be used for games I see more or less the pitfalls, that the cell itself wanted to avoid.
I don't have doubts that developers will use the SPEs, but I have severe doubts if they unlock the power! It is not the first time that these walls were tested. Developers have the tendency to overestimate their own abilities. Some think they are a genius, after they found a solution that several hundred other people have already solved, too.
I don't really think that the developers today are really better, instead the methods that we use were in fact developed by other developers. In a way we stand on their shoulders. There are ways to use some of their power, especially with the help of libraries, but they will only be able to extract a fraction of their power. The real way to use their power would need a development process that is simply to expensive and it offers no real protection against race conditions. If you don't work very careful you find yourself in the hell that development for parallel systems can mean.
In a way parts of these dangers are even inherent in multi cores, but due to the fact that there are only small changes from a unified program on a single core, it is easier to control and debug them.
What you don't really understand is that there is not a basic problem and if you avoid it on one place you are free to use the whole power. This is simply not the case. Multi cores are no longer seen as a very big problem, but in fact even for them several restrictions apply. If you are able to get from a triple core the power of two cores you are pretty good! In fact there are many problems that would not allow such a speed up.
In the same manner you won't find one lock and if you solve it you have the whole power. Instead you will find a way to get around one problem and get perhaps 10% and another might give you 2%, but then you find a third problem. There might be a way to solve it but it would mean that the solution for the first problem will no longer work, instead you have to rewrite the firsst solution and get 7% and the solution for the third problem gives you 4%.
This description has more to do with reality, than your idea!
And the believe developers like to solve problems, in a way yes, but you have to find a way to really improve the situation. While developers like to solve problems they don't like it when their boss comes every 5 minutes to ask if you are ready, because they are all waiting for you!
You are forgetting that the development is no game. You are paid to get results! If the complicate architecture means that you get less results in the same time there is a problem. Either your boss has to be ready to pay more, because you have to work longer on the same problem, or you are ready to be paid less, or you decide that this problem is not worth your time!
