Multiple-GPU: Way to Go?
You may have noticed that the general tendency has steadily been going towards the multiplication of the execution units lately. First we had dual-processor (and multi-processor) systems. Then we had dual-graphics card systems. Now we have dual-core processors in dual-processor (and larger) systems. But what about graphics? Yes, the development of the graphics subsystem architecture is also going very fast.
As we have reported in our news , Dell demonstrated the world’s first quad-graphics card configuration with four high-end GeForce 7800 GTX graphics adapters. These are the single-GPU cards. But what if one day they install four of these babies:
Dual-GPU graphics solution from ASUS built with two 7800 GT chips hosted on the same PCB and hidden under a solid heatsink doesn’t support SLI. At least not yet. But one day they surely find a way to tie up a couple of cards like that within a single system.
The principle “the more, the better” seems to be working just fine ever since. By employing two GPUs instead of one the developers should be able to double the fps rate provided by the graphics subsystem, i.e. to double the graphics performance. However, much more work needs to be done to achieve true multi-GPU performance, namely, better integration of the frames, better synchronization of the speed between the two GPUs.
According to the GPU developers, the current examples of the dual-GPU graphics cards still have issues. And when the solution is intended to hit the consumer market, efficient scaling is the most essential thing that needs to be done right, otherwise the graphics product will be excessively expensive. At this point the solution based on two cards, each with one GPU onboard, should offer better scalability in the long-term prospective than a single graphics card with two GPUs onboard.
Well, the first signs of the multiplication trend in the VGA market are already here. Let’s be patient, and I am sure that very soon there will be much more to talk about.