IBM's efforts to write Cell's compiler ...

phongn · Post by **phongn** » 2006-02-26 09:55am

Ars linked to a 25-page paper by IBM summarizing their work on writing a compiler for the Cell (the Octopiler).

This article is technical in nature but is fairly interesting.

Ace Pace · Post by **Ace Pace** » 2006-02-26 09:58am

I've read the Ars summery and it dosn't look good, the PS3 initially won't be anything as impressive as it will be later on.

Durandal · Post by **Durandal** » 2006-02-26 05:36pm

This isn't anything I haven't been saying for a while. People wanking over the Cell need to realize just how different it is from today's mostly single-threaded world, and even from conventional multi-threading. The Cell is basically a non-deterministic computer up to N possible states. That means it expands to a deterministic machine of exponentially greater complexity.

Admiral Valdemar · Post by **Admiral Valdemar** » 2006-02-26 05:46pm

Ace Pace wrote:I've read the Ars summery and it dosn't look good, the PS3 initially won't be anything as impressive as it will be later on.

Can you say not that for every console in history? It stands to reason that time will enable developers to produce better games, the PSX started with Tekken using barely 10% of the system's ability. By the end, Tekken 3 was near 100%.

Durandal · Post by **Durandal** » 2006-02-26 06:09pm

Admiral Valdemar wrote:
Ace Pace wrote:I've read the Ars summery and it dosn't look good, the PS3 initially won't be anything as impressive as it will be later on.
Can you say not that for every console in history? It stands to reason that time will enable developers to produce better games, the PSX started with Tekken using barely 10% of the system's ability. By the end, Tekken 3 was near 100%.

The Cell is on a totally different level. It's not just a matter of learning the specific quirks of optimization for a new architecture, it's learning to program in an entirely different way.

Admiral Valdemar · Post by **Admiral Valdemar** » 2006-02-26 06:29pm

Well, the guys who stayed with Sony knew what to expect, and given the only easy console by them to code for was the Net Yaroze, it's not like this is an amazingly new development.

Arthur_Tuxedo · Post by **Arthur_Tuxedo** » 2006-02-26 07:16pm

Looks like Sony's really left holding the bag of shit, here. IBM and Sony both contributed to the development costs, and IBM could see substantial benefit with cell-based servers and supercomputers years down the line, while Sony's console is probably going to be delayed and nowhere near as impressive as the competition at launch. Ouch.