With these next two tests, I wanted to mix things up a bit. The first chart takes our 22 MB YouTube HD video and mixes it down to 856x480 for the Xbox. Of course, Xbox uses WMV, not MPEG-4, so we shouldn’t observe much in the way of acceleration—and we don’t. With the “Enable hardware encode” box checked, the HD 4890 and Stream manage to shave only 11 seconds from a CPU-only job time of 5 minutes 27 seconds. I’m more inclined to chalk this up to random test fluctuation than anything done by the GPU since I observed 3% less GPU load with the “enable” box checked. With the GTX 280 installed, this box doesn’t even appear, so there’s no chance to get confused about why your non-CUDA-compatible format isn’t being accelerated. Point to Nvidia for better coding.
How about some of those antique digital videos you’ve still got floating around in MPEG-1 format? I grabbed one 86 MB 640x480, a really grungy bit of dogmeat recorded by a pocket camera, and threw it into the YouTube MPEG-4 480x360 profile. Again, CPU-only times are nearly identical. Encoding to MPEG-4, layer 10, we should see decent GPU acceleration times by both cards, but Stream sort of snoozes through this test, offering only a 10% improvement. Meanwhile, CUDA kicks in a 34% boost. Why the sudden change of roles? No idea, but a win is a win.
Score: AMD 2, Nvidia 2.