The Year of Multi-GPU, Multi-Core and The Dawn of Entertainment Processing
Multi-GPU setups and multi-core microprocessors have been on the market for several years now, but the computing experience overall did not change much due to the lack of software that takes advantage of the sophisticated hardware and certain drawbacks that certain technologies do have. But it seems that next year everything will begin to change and games will finally start to take advantage of extreme computing power of today’s PCs and we may encounter completely new paradigm: entertainment processing.
Graphics processors today are much more than just chips able to draw 3D visuals on the screen. All developers of GPUs are talking about general-purpose processing on graphics chips and this will inevitably change video games drastically. Nowadays Nvidia touts physics processing on GPUs and some games are about to take advantage of PhysX technology shortly. Meanwhile, ATI demonstrated artificial intelligence processing on GPUs.
If graphics processors have to compute graphics, physics and AI, then this chip should be extremely powerful: modern GPUs can barely provide decent frame-rate in full-HD resolution in contemporary video games and it is hard to believe that the chips have enough computing power for two additional tasks.
It should be noted that processors processing AI, graphics, physics and other highly-parallel tasks cannot be called juist "graphics processing units", as they do much more than just graphics. As a result, we would expect the emergence of entertainment processing platforms.
Considering the fact that it becomes harder and harder to create “mega-chips”, we believe that multi-GPU setups will get more and more popular next year both as a result of higher demands towards graphics performance and potential emergence of games that compute physics effects or other things on GPUs. If each GPU chip is assigned to a certain task, then such multi-GPU setups will not have certain drawbacks associated with modern multi-chip rendering techniques, moreover, it should be unnecessary to have several similar GPUs in a system: simplistic AI and physics algorithms should not demand a high-end graphics board.
Needless to say that with necessity for even more graphics and processing horsepower, the demand towards performance components will at least be flat in 2009.