Unlocking Cell's Potential: Part I - Introduction

Tapping the Potential of Cell

Part I

Introduction

Two months ago we started something that I hope will become a long-lasting tradition here at PSINext: developer interviews that focus on Playstation hardware and development on a lower, more intimate level than is the norm in gaming journalism.  Today, we continue that tradition with a highly focused look at the Cell architecture that powers the Playstation 3. 

As such, it is my honor to host and interview Mike Acton of Vivendi's High Moon Studios - Senior Architect of High Moon's unique Cell/PS3 research team, and the founder/administrator of developer destination website CellPerformance.com.  For those with an interest in Cell and the potential means via which its potential may be tapped, we hope that this interview will provide valuable insight into current best practices, and foment ideas as to what may one day be achieved on this dynamic new architecture.

PSINext: Mike you’ve held a number of senior development positions in the past, and presently hold the title of Senior Architect at Vivendi's High Moon Studios; what in particular about Cell led to your decision to start a grass-roots website geared not just towards game development specifically, but towards programming for the BE architecture in general?

Mike Acton: At the beginning of this year it was becoming obvious that IBM was committed to having the Cell architecture become the standard-bearer of its Power line of products. Sony and IBM have both released quite a lot of documentation, as well as the Cell system simulator. The stage has been set for the BE architecture to become not simply a one-off game development platform or server, but the first in a long line of Cell products dedicated to high-performance computing. It is because of these public resources that I have been able to invest a great amount of my own time, in addition to my real work, on the platform.

It has also been my long-standing belief that what is good for the industry is good for our company. For instance, our efforts to create more and better products can translate into a larger install base of users. It is a tradition among game programmers to share techniques and discuss the challenges of any particular platform; however, console game programmers have always faced the roadblock of a proprietary architecture. Both access and interest outside of video game development has traditionally been limited, so our best tool for sharing has always been insider development conferences not open to the general programming public.

And as an industry, we also find it challenging to locate new staff. In many ways the education system, both university and trade, fails to focus on some of the principles that are so dear to us as game programmers... so that with each new generation of programmers comes the long - and perhaps natural - process of learning and making the same mistakes over again. But with the projected development staff size requirements increasing so dramatically for this generation of games, we are left with a problem: there are simply not enough people to go around who can develop high-performance technologies on these platforms.

Ultimately, it was the combination of these factors that lead me to the decision that someone external to IBM, Sony and Toshiba had to help build momentum and foster the community of Cell developers outside of video games. It is clear that Cell will be a strong player in medical imaging, signal processing, and the film industry, as well as find continued military use. The core skills needed to take advantage of this architecture are not unique to games. If I can help to provide a forum for those with an interest in the Cell to share and learn acquired best practices, then down the road they might return the favor, or we might apply the improvements they make to game development. And perhaps we will see those programmers enter into our industry.

PSINext: So you view CellPerformance as a site that may not only help to build the BE architecture knowledge base in general, but that by doing so may in turn have a direct positive effect on Playstation development itself?

Mike Acton: Yes, without a doubt. For me, the greatest reward of working on CellPerformance.com would be developers letting me know that something that we published there made their product better. Playstation development is not a competition for knowledge. The more we can help each other, the better all of our products can be overall, which is better for the players and better for the health of the industry.

CellPerformance.com is still in its infancy. My hope is that even the little things count, and that someone out there may find something useful for their production code. I also hope that the community will appreciate that the site has no agenda apart from contribution. I don't personally sell any Cell related products or services. Certainly, I hope that the proprietary work I do at High Moon Studios will help us stand out in the marketplace. Yet the work I do on my own time using publicly available documentation and tools is meant to benefit everyone.

There are three messages from my work on CellPerformance.com that I hope are heard:

1. The Cell BE architecture is not a magic bullet. Simply re-compiling an existing application for the Cell will not see any benefit. Most systems must be re-architected and data re-designed to take advantage of the architecture. But when that's done, the performance is unrivaled by any consumer-level technology on the market today.

2. The Cell BE architecture is not a radical departure from existing high performance designs. Certainly the Cell is a disruptive technology, but that's more to do with its place in the market than its design. What it is, in fact, is the very solid culmination of decades of experience in hardware design and the logical next step in hardware architecture. With homogeneous multiprocessors being too limited by memory performance and independent DSPs having their own set of very individual problems, a fully-integrated heterogeneous design was the way to go.

3. The Cell is real, and is a blast to program for!

Part II: Dataflow and Design

Part I - Introduction

Part II - Dataflow and Design

Part III - Programming: Tools, Tricks, and Language

Part IV - Programming: Branches and Half (not single!) Precision FP

Part V - Cell and the PS3