The Cell Processor
By dcb on Feb 08, 2005
The latest buzz on the streets, at least around those neighborhoods frequented by the eXtreme crowd, seems to be about the Cell Processor. I wrote a little blog on the Power 6 recently and one reader warned me to watch out for The Cell.
Well, I have to admit, I'm a bleeding edge junkie myself at times. And the theory of operation around The Cell is pretty compelling. The problem is that theory doesn't always translate to reality! In fact, it seldom does. Especially when S/W is a critical component of the translation.
Gartner suggests that only 1 in 5 major initiatives that Sr. Mgmt funds and resources ever delivers to promise... 80% fail to meet expectations. IBM talks about a recent Standish Group report that suggests only 16.2% of S/W projects are delivered to promise. Another study suggests that > 40% are canceled before delivered (and most that are delivered are late and/or way over budget, often never recovering costs).
If you read the reports about Cell, it isn't about the H/W... That's the point really. The H/W is made up of standard building blocks (cells) of Power cores. A socket holds a Processor Element which contains a main Processor Unit (core) and several (often 8) Attached Processor Units (cores). However, the interesting part is the "Cell Object", which is a S/W construct that includes code and data that can migrate around looking for a "host" Cell system on which to execute. There is talk of dynamically-orchestrating pipelines. Of S/W self-healing properties. Of dynamic partitionability with multiple native OS support. All S/W ideas.
So it isn't really about H/W. The H/W "Cells" are simply the "amino acids". The more interesting question might be: is there an "intelligent designer" who can breath life into a soup made up of these "single celled" organisms? There is a precedent for doom - where advanced life forms failed to thrive due to a lack of S/W life support (eg: EPIC/VLIW, Stack Machines, etc).
We saw earlier the dismal failure rate of projects using well established S/W development paradigms. It'll be amazing if Sony/Toshiba/IBM can turn the PlayStation3 engine into a viable general purpose computing platform that can threaten AMD, Intel, and SPARC at home and in the datacenter. From what I hear, the development tools and processes for PlayStation2 are an absolute nightmare.
It'll be fun to watch this pan out. One thing is for sure... at least PlayStation3 will ROCK, if they can deliver a reasonable flow of affordable immersive networked games. I hope so.
The Cell makes for great reading. Unfortunately, when it comes to a general purpose platform, this one might never recover from Stage 3: