1 person found this helpful
I'm hardly the expert in this issue, but I'd take a guess and say it's because the hardware MPE always uses Maximum Render Quality. If you have that turned off in software MPE, the resulting export could be faster. You can test this by making sure MRQ is turned on in software mode and noting the export time.
1 person found this helpful
First look here for a basic explanation: http://ppbm5.com/Background.html
Second, and this a repeat of what was stated in the link, MPE assisted encoding uses maximum quality for the scaling and rendering, software MPE does not. To approach MPE assited encoding, you would have to turn on MRQ when using software MPE to get comparable quality. In that case the times will be very different and will show the benefit of hardware MPE.
@ Jim: Beware of the diference between maximum quality and MRQ.
What is the difference?
MRQ relates to rendering only, maximum quality mostly relates to scaling, using the same bicubic algorithm that is used in PS, but optimized for 64 bit and MPE. Steve Hoeg posted the details here earlier. Look it up if you want to know the inner workings.
I recall reading such a piece. Here's what I took away from it.
With MPE in software mode and MRQ off, one type of scaling is used.
With MPE in software mode and MRQ on, a better type of scaling is used.
With MPE in hardware mode, MRQ is always on, and the best scaling algorithm is used.
You're saying that Maximum Render Quality and Maximum Quality are two different settings?
Yes there is a difference, because when VRAM is depleted, the CPU takes over and then the MRQ setting can kick in with a performance penalty of course. So it boils down to this: If VRAM is not depleted, there is no difference between Maximum Quality and MRQ, but if VRAM is depleted, there is a difference. At least that is how I understand it.
Steve, Todd, please correct me if I'm wrong.
Steve just pointed me to this article, that gives some background: http://blogs.adobe.com/premiereprotraining/2010/07/maximum_dimensions_in_premiere.html
Thanks Harm. I'll check that out.
Yeah, that wasn't the original article explaining the difference in scaling algorithms used. And it doesn't really answer the question about whether MQ and MRQ are two different settings.
Scaling for instance uses hardware MPE, scale to frame size does not. If you have a time line without effects or transitions and only use AVCHD or HDV material, MRQ has no benefit. If you use Scale to frame size, hardware MPE does not work, you have to rely on software MPE and then MRQ comes into play.
As you notice there is no clear cut answer, it depends on your footage and your specific time line, and maybe that is why Adobe has no clear cut answers.
Notice that once CUDA/MPE can not handle the render, software takes over as if hardware MPE had not existed at all. That is where MRQ is beneficial.
Thanks Jim and Harm.
I tried some exporting tests today. First exporting the R3D 4k clip (which contains GPU accelerated effects) with GPU acceleration turned on (MRQ was on) to 1080 h264. Then exported the same clip with GPU acceleration turned off (MRQ still on) with the same export settings (1080 h264). The difference in export time was 5 seconds. Shouldn't the GPU acceleration be faster? Or are my expectations too high? Is there another variable I should look at?
I did notice in the activity monitor that my 16 GB of RAM was sometimes hovering around 100 MB of "Free". In other words, almost 16 GB were either used, wired, active, etc. and only 100 MB was free. Although this didn't happen all the time. It mostly stayed around 500MB to 700 MB being free. Could this be the culprit?
Run the PPBM5 Benchmark and PM me the results. If your setup is correct, you would see a 10-fold increase of the render times without MPE. It will also show how your system performs in comparison to others. Do follow the instructions to the letter. This is the easiest way to check whether everything is OK.