IBM Cell programming in 2010 - feasible and worth it?

Question

Hello, I would need your help. I've come across an interesting book - Programming the Cell Processor: For Games, Graphics, and Computation - it contains mostly C and some Assembly for Cell. The technology is interesting indeed, but there are some doubts on my side.

The book is from 2008 and some things has changed:

• There is no Linux support on current firmware version.
• Last version on IBM's website is from 2008 Red Hat Enterprise 5.2 and Fedora 9 - has anyone an experience running this IBM SDK on Fedora 13 or at least any version higher than stated Fedora 9, and is Linux available of sufficient testing?
• Would it be useful for example for creation of distributable PSN game, and if anyone knows anything about price to actualy get a product there (as I've heard that it waaaaay more expensive than for example X-box indie games)

So do you think that it is worth it or not? Be it just for education purposes or something "more" serious?

Any thoughts are welcomed, thank you!

Answer 1

Given that you don't have access to a Cell machine, I'd advise that it's probably not worth it. I absolutely love the Cell architecture - I think it was a fantastic step in the right direction. Unfortunately, having done some Cell development in the past, I was really disappointed with the tool chain, the simulator and the seemingly hostile attitude taken towards developers recently.

So given that you're not going to be able to use a real Cell machine in order to get the speed gains you would get from writing programs within that idiom, you'd probably be much better off looking into general distributed programming techniques (using MPI or something similar). These skills are going to be readily transferrable to the Cell or its derivatives, or any similar architectures that might arise in the future.

As far as I'm concerned, and as much as it pains me, I think the Cell is basically a developmental dead end unless you have access to a commercial development license, you'll be extremely frustrated in your ability to actually get anything out of the architecture.

Answer 2

Cell was dumped by IBM for general purpose computers. It will live for the next 5 years in the Playstation and i'm pretty sure that the next generation Playstation - whenever it will be ready - will also use Cell again because establishing something new in CPU land is so unaffordable today.

But as a technolgy it is indeed no longer interested. Learning CUDA might be a better choice.