Originally Posted by Joseph Clark
I haven't used that software, Don, but it looks like it could be revealing. I use a few common Windows 7 gadgets to monitor RAM, threads, HDD, and network activity as I render. It's all very informal at this point. When I rendered almost identical timelines the other day, the curious thing I noted was that the i7 2600k with the hard drive array was using about 80% of the 8 threads, while the i7 3930 was using over 95% of the 12 threads. There are lots of variables between these two systems other than hard drives, of course, but it still struck me as odd. The 3930 system rendered over 3x faster. Neither system was overclocked.
My viewpoint on this subject is a bit different. I look to save me work time. To do that the first goal is to get the timeline playback time to be realtime with no buffering. That means to me that first, the system must feed the video at a rate that flows fast enough that the processor is not waiting for frames off storage. I call this bottleneck #1. Of course this is only the first step in timeline playback as we then need to process all the timeline's effects, color correction, pan and cropping, cg, as well as the codec itself. This would be the GPU/CPU being powerful enough to calculate the timeline content for preview. Note that none of this has anything to do with final rendering time. So far all I'm looking at and. For is a bottleneck where one part of the system is holding up other parts. Maybe the tool to evaluate bottlenecks is the best measuring tool to use. So far, everything we're looking at are operations in the system that are working while we are at the work station, hands on. If the system is performing properly, any time we play the timeline we should be not having to wait to make a decision on what changes to make or give our approval to move on. IMO, this is the area that needs priority for performance. Now, assume the approvals are in and we like the program story and are ready to render the final file for burning to 3D BD or authoring program. This rendering is not hands on requiring out attention so therefore it, IMO, is less important as I can walk away and sleep, watch a movie or do something else. likely, of the timeline playback is running. At top performance, the final render will also be reasonably fast.
When I'm editing there is nothing more annoying to me than having to wait to see the result of my recent change. That wait is the biggest bottleneck and my highest priority to resolve. I agree with the engineer from AMD in that first order of business is to feed the frames off the storage so fast the processor doesn't have to wait, second is for the processor(s) to compute the timeline to. View at real time and adequate resolution to decide if the changes are proper. He feels that the FirePro 5900 will get that result for two MVC timelines and color corrections and other pans and crops as long as storage can deliver.
I look forward to running the tests next week when I get home and grading my system. I now have 4 large projects in the can and will be shooting one more in 2 weeks so this has considerable importance. I really had thought that my 2 7200 RPM raid0 drives would be adequate but if you are still having trouble with 3 then I need to make the move to SSD. It's a matter of cost too because these projects don't have a budget and need to be done with hobby money. Gone are the days when I would adjust my billing to cover these costs.