This week IBM demonstrated just how highly it regards its Cell Broadband Engine chip and its entire parallel architecture, by putting it together in a product offering with its mainframe range, these days called the Z series.
IBM sees the mainframe handling transactional jobs like web applications, database, messaging, so keeping track of multiple sessions and billing and things like handling security, availability, logging, auditing and compliance, while a rack of Cell blades can handle anything computationally intensive.
This concept goes back at least 25 years inside IBM, a general purpose processor married to a special purpose processor, and as a company it is exceptionally well placed to design systems like this. It has the intellectual property, the experience on-board and all the previous success and mistakes made along the way, whether they are technical or marketing issues.
Offering two totally different processing architectures means that the computational side of things can be around 4 times more powerful than similar priced chips such as Intel Dual core processors, and also use far less power.
IBM sees this type of combination being used for jobs involving pattern recognition like financial market analysis and fraud detection, as well as medical imaging, computer vision, bioinformatics and of course the running of monstrously large virtual worlds.
We’ve all read Neuromancer and Snow Crash, where we see internet connected participants indulge in worlds that are so realistic that they appear real ‘ and if you add visualization technologies to Massively Multiplayer Online Role Playing Games, and leave it to evolve for a few generations, you have exactly that.
But we have rarely considered the computational engines that will allow this type of environment. Obviously we need a powerful rendering engine just to show us a visually accurate 3D world, and the immediate question is where is the best place to run the computational world rendering tools ‘ in a PlayStation in our homes, or on a centralized computer. Well one way of handling that is to offer the same type of processing power at each end, which run in an identical manner.
While Faultline doesn’t believe that the Cell Broadband Engine is a complete vision, delivered overnight, to offer an immediate advantage to the PlayStation gaming console range, the fact that its underlying architecture is designed for the Cell chips to plug together in twos on a single blade and for these boards to be loosely connected in large clusters to create immensely powerful supercomputers, is part and parcel of that vision. Sony may not have thought through the implications of that from day one, and it may still be struggling with it, but the fact is there are inherent advantages.
So some rendering can be done on a machine at home, and some centrally if need be, with no change of code. But the big issue surrounding the development of something we might call a ‘Gameframe,’ is that the interaction, up until now, within any virtual world has had to be carried out by exporting an entire ‘world’ to each PC or games console, and then transmitting low bandwidth movement, and action codes continuously between computers. When a new character or artifact arrives, some form of basic construct for its look and feel also has to be send in low bandwidth. This then was the basis of VRML (virtual reality meta language) and its successors, which never really caught on.
Perhaps another approach to this is to allow the interaction to happen on a central computer and then transport the visualization data to the remote device, but this requires large amounts of bandwidth all the time. By having the same rendering power at both ends, and having faster but not unlimited bandwidth, parts of a vast virtual world can be downloaded as they are needed, and this creates a strong way of optimizing different types of gaming design.
IBM gave a glimpse of all this when it said it was working on a ‘Gameframe’ with Hoplon Infotainment, a Brazilian virtual world creator back in April last year, for a 3D version of its bitVerse game. At the time it was not clear that the development worked with a Z series mainframe, but the overall concept caused quite a fuss in the gaming world.
The general purpose mainframe (or an attached Linux box) will look after logistics and Websphere based connectivity issues as well as running a core database under IBM’s DB2 relational engine. The Cell will handle all visual activities and other computationally intensive challenges (modeling offword explosions in the case of bitVerse).
Fundamentally under its design ethos and by its very name, the Cell processor assumes the existence of multiple other connected Cells, so that servers spread all over the world, can created an actual virtual “computer,” which in turn hosts a single massive virtual world all connected together via the Internet. Such a network might add game features such as semantics and reasoning and artificial intelligence to enrich games/worlds.
And of course if you have a network with 1 million Cell chips connecting to it, why would you not arrange software like a Peer to Peer network and offload work cycles to those remote Cells inside a gaming console. That discussion is an entire other paper in its own right.
We’d be surprised if Hoplon was the ONLY company going down this route. MMORPG is currently curtailed only because getting started in games is tricky, and takes time to build momentum and this is largely because their instructions and graphics remain cumbersome. We are rapidly heading out of that phase of games development.
We are also rapidly moving away from an internet constrained by 250 kbps or 500 kbps connections, and the combination of this kind of ‘grid’ approach and higher, video quality bandwidths coming to gaming is that it will become a visually entertaining sport.
Faultline came up with the idea about 3 years ago regarding gaming that once full video realism was possible and once its imagery could be sent down a single broadband line, then gaming as a form of spectator entertainment is certain to emerge.
Games or scenes or battles, which are deemed to be legendary or recent important wins, will become stored and non-participants will be able to view them as part of their learning curve. Parallel this with other forms of entertainment. We go to the theater or cinema to watch great actors, but we still watch our children’s school play. We watch professional sports stars, but still play soccer ourselves at the weekend. So a complex series of processes, with amateur and professional divisions, qualifying rounds and tournaments can be brought together from all over the world, and we will be able to watch our first ‘qualifying’ win in a particular tournament, as easily as watching a recent bout or battle from the world champion.
MMORPG systems just want people connected and paying, they don’t necessarily have to be playing.
High definition TV screens need more and more ‘reality’ style video experiences of quality, and this type of gaming is a natural. And Sony, with its understanding of both console play and gaming portability on the one hand and MMORPG on the other, is the right company to stimulate the creation of this type of gaming universe. But so far Microsoft has been more inventive here, despite not having the processing power advantage on its console (Xbox 360 runs on a previous IBM Power chip).
What Sony and IBM MUST do is create an environment where it is easy to build this type of gaming hub, and harness it to the power of both a PlayStation and/or a PC with a Cell card attached.
Investors must be able to buy all the kit ‘ an existing 3D console game, the server, simple programming tools and a route to market (a network or storefront for MMORPGs) and the games must reach to TV sets, PCs, consoles and eventually to mobile handsets and portable devices like the PlayStation Portable.
But in order to do that the price of these devices must go down. And IBM here is more than playing its part to find computationally intensive jobs for the Cell, and to create an eco-system for coding professionals that have worked with it in order for it to flourish.
What is likely to emerge are gaming portals that are the modern day equivalents of internet portals ‘ so companies like Hoplon will become the Yahoo, Google, AOL, MSN and MySpace of the future ‘ but offering a series of virtual world services.
Investors would be wise to get on the bandwagon now, the IBM Cell combination is likely to just the top of the range architecture ‘ there will also be others. Eventually we would expect second generation Cell devices to be the basis of many computationally intensive applications ‘ many of them to do with visualizing 3D images ‘ that are pervasive in our lives as we see the Second Life market evolve.
In 15 years it will be as unthinkable that each of us does not live in a second or third world, and that there will be ‘portals’ for traveling between these worlds (like IM connectors for AOL to Microsoft) as it would be that each of us today does not have a mobile phone.
What IBM has delivered in this round of announcements is that Cell blades can join the eco-system for connecting to a mainframe that has always been there. It already has a set of Accelerator Libraries for offloading work from the Z series to Linux servers, so these will be extended to the Cell, and similar its Data Communications Synchronization libraries will get the same treatment.
It has already brought to market the RapidMind language optimizers which effectively take code written in C++ for linear applications and re-phrases it to take best advantage not just of the Cell architecture, but takes into account how many processors the program has at its disposal and re-configures the code accordingly, effectively automating parallelism.
There are other libraries that IBM is bringing to market so that code written for its other high performance devices, built around AMD Opterons for example or its earlier generation Power devices ‘ can also compile down invisibly and offload to run on a Cell board.
IBM had taken Cell into the heart of everything it does ‘ and with that kind of wide usage, its cost as a component will fall radically ‘ opening up more and more uses which are economically feasible for the device.