CPU-Usage Benchmarks: Picture-In-Picture Acceleration
In this interesting benchmark, we're going to see what happens when these IGPs are tasked to decode a picture-in-picture Blu-ray disc. To help us see what's happening, we've chosen to benchmark a part of the Blu-ray "Sunshine," where the second video stream is dropped after a minute or so, but the CPU usage results are recorded for another couple of minutes. This lets us see a tangible difference of CPU usage between one and two streams of video:
The benchmark is a little muddled, so pay close attention. It looks like the GeForce 8200 (green line) is running somewhat flat at the top of the results. The 785G (light red line) is running flat at a lower CPU utilization and the 790GX (dark red line) starts off at a higher CPU utilization but drops about a third of the way in.
What's happening here is that both the GeForce 8200 and Radeon 785G possess dual-stream acceleration and aren't fazed much by the transition between two-to-one video streams. However, the 790GX does not have dual-stream acceleration, so it has to use more CPU resources at the beginning of the video.
Oddly enough, regardless of the 790GX's lack of dual-stream acceleration capabilities, it still pulls similar CPU utilization compared to the GeForce 8200 when processing multiple streams. When it drops down to a single stream of video, the 790GX lowers CPU resources and approaches 785G levels.
The Intel systems show a clearer story, with the GeForce and the new 190.38 driver demonstrating good dual-stream acceleration, but keep in mind that the 190.38 driver does not offer all of the visual-quality enhancements that the old 182.5 driver does. The 182.5 driver doesn't appear to have picture-in-picture acceleration and neither does the Intel G45, although the G45 is performing a bit better.
Once again, however, the CPU usage by any of these solutions is somewhat insignificant. Even lower-budget CPUs should be able to easily handle what Blu-ray throws at them. As time goes on and as CPUs become even more powerful, benchmarks like this will inevitably offer less and less useful information.