Conclusion
What’s the take-away? Generally speaking, if you’ve got a piece of software that takes decent advantage of CUDA, you’ll realize a strong degree of performance scaling by stepping up in the number of stream processors. There are some mysterious exceptions, as we saw with Super LoiLoScope, but the norm looks more like Badaboom. In situations like the 9600 GT vs. 9800 GTX, where only $20 separates double the number of stream processors, you’d need to get your head checked if you took the lower-end option.
There’s no question that CUDA delivers, even near the bottom of the price band. It would be interesting to circle back and see if similar results spring from Nvidia’s mobile GPU products (or even the recently-released Ion platform), but I suspect this would be so.
AMD now has a grave challenge at its feet. With the next batch of ATI Stream drivers due out soon, can the red team deliver similar or better GPGPU improvements across all price groups? We know that the emphasis in its upcoming release is on improving the image quality of its transcoder software rather than scaling performance. At the same time, we also know that the company is planning on showcasing its own batch of third-party partners who've taken advantage of its architecture.
One thing is certain: this performance exploration is merely the tip of an iceberg. Within the next couple of weeks, you can expect a follow-up with numbers from the other team. We can't wait to see how this plays out. And regardless of which hardware architecture is faster, it's clear that enthusiasts (mainstream and hardcore) are going to be the biggest winners here.
