Transcoding Panasonic GH4 UltraHD 4K Video to 1080p: Surprising Speed on 2013 Mac Pro with Final Cut Pro and QuickTime
Here’s a surprise I found today: transcoding Panasonic GH4 3840 X 2160 4K video to 1080p using QuickTime Player, the MacBook Pro Retina (late 2013) blew the doors off an 8-core 3.3 GHz 2013 Mac Pro with D700 GPUs. Both tested on fast internal SSDs to rule out I/O as a factor.
4K video chews up space, store it on a high capacity Thunderbolt unit.
When I investigated, I found a simple explanation: the Mac Pro uses all 8 of its 3.3 GHz CPU cores, whereas the MacBook Pro uses very little CPU time. This suggests that GPU support in QuickTime Player is not used on the 2013 Mac Pro in that context (see comment from reader Michael C at end). But it does show that the GPUs can rock, if used!
Further use with Final Cut Pro and Compressor suggest that this speed differential vanishes, so perhaps there is a quality or other aspect that goes missing when using a simple QuickTime Player transcode.
Note that older Mac Pros can benefit for video processing from one or more fast video cards—still photography to a lesser extent, much more so for video.
Sidebar issue: attempting to use DaVinci Lite on the 2013 Mac Pro to process video results in an AWOL GPU subsystem that requires reboot—but it won’t reboot due to a hang, hence forced power off is the only way to restore proper system function. I thought that DaVince Lite might make use of the GPUs for transcoding, but I have not been able to make it function.
Final Cut Pro requires 97 seconds to transcode the 4K video clip to 1080p mp4 file. The MacBook Pro Retina takes about 30% longer. Similar results with Compressor. So something very different occurs with a simple transcode in QuickTime Player vs Final Cut Pro/Compressor (similar file sizes for QTP, but smaller so maybe there is a lower quality hence faster, though if so, it’s not obvious in the output while viewing it).
Apple’s web site states “And Final Cut Pro uses both GPUs to accelerate background rendering and speed...”. Having no frame of prior reference, maybe I’m missing something as to what does and does not take time, quality metrics, etc. It’s unfortunate that there are no monitoring tools for GPU usage.
The MPG Mac Pro has the D700 GPUs and 3.3 GHz 8-cores, so one does have to ask with some workflow concern: is 4X to 10X real time (4 hours to transcode one hour of video) as fast as Compressor will go on the 2013 Mac Pro? At the most basic level, the question is whether those dual GPUs really being used?
MORE info and context: Compressor (part of Final Cut) is far slower on the MacBook Pro than a simple Quicktime Player 1080p conversion on the same MBP, and ~30% slower than the Mac Pro. Moreoever, both use the CPUs fully in FCP/Compressor.
As a novice video user, my expectations are ingenuous; starting to dip me feet into the water, I just tried features such as “Export” or “Share” and I ask what happens. Best advice: test your own workflow with your own video and see what happens. It’s clear that I need to learn a lot more about video transcoding and the like to make sense of it all.
Reader comments below.
Michael C writes:
The MacBook Pro has Intel Iris graphics, which accelerates video encoding. A link to Intel's white paper about it:
The graphics cards in the Mac Pro from AMD, might be more capable at many other tasks, but AMD does not implement video encoding acceleration.
That probably explains the difference in encoding speed you are seeing between the MacBook and Mac.
MPG: I’ll ask some questions (pleading ignorance); I don’t know the answer here. It does seem very hard to believe that the dual GPUs in the 2013 Mac Pro would not be well suited to video work, including transcoding: what else is the point of having them?
Amir N writes:
Your updates on the post re: FCP vs. Quicktime seem to support that quicksync in Quicktime X is the factor. Quicksync transcodes will be slightly lower quality (and smaller in size) than CPU-only transcodes, which is exactly what you're seeing. Moreover, GPU based transcode should put almost no load on the CPU, and you're seeing all CPU cores light up on the Mac Pro during a Quicktime transcode while the MBP shows almost none, meaning that the Mac Pro is doing a pure CPU/software transcode, while the MBP is doing a quicksync transcode.
Then the fact that the transcode times and CPU usage times look more reasonable in the FCP transcoding confirms it to me. I'd bet dollars to donuts that quicksync is enabled by default in Quicktime X, and that neither FCP or Quicktime X are currently able to take advantage of the Mac Pro's discrete GPUs.
MPG: makes sense, but power usage suggests that the GPUs are in fact being used.