Keep your eye on the numbers. Our worst AME export time in Premiere Pro CS4 was 26:03 (2 cores, no HT). The best was 8:59 (12 threads, with HT). The counterpart numbers with CS5 were 17:34 (2 cores, no HT, no CUDA) and 4:15 (12 threads, with HT, with CUDA). Obviously, this is an improvement, but we feel a bit let down, as if saying, “Well, our monthly trade deficit with China is only $35 billion instead of $45 billion.” That’s an improvement, yes, but not really the stunning win we all hoped for. The shift from 32-bit to 64-bit and all the additional memory that came with it plus the occasionally epic benefits of the Mercury Playback Engine only managed to deliver a roughly 2x benefit here.
A 2x improvement is far from the 10x gain we saw in the render test, but keep this in perspective. If you were a contractor making $300+ an hour for video production, would you pay $1,500 to cut your export time in half? In a heartbeat, right? Even at 5:27, the quad-core configuration with HT and CUDA shows a substantial improvement over our CS4 test. And to be fair, as noted earlier, we used frame doubling in our Premiere Pro test, which is why the results here are less impressive. That frame doubling work is being done in the CPU. A GPU-based acceleration would have yielded considerably faster exporting.
The upshot here is that you want at least quad-core with Hyper-Threading and you want CUDA, which means you also want CS5.
The refrain on the CPU utilization score is that if you want to look at anything more than a maxed out system doing nothing but exporting in CS5, you’d better have at least a 12-thread CPU in action. Every other configuration spent most of its time in the upper 90s.