Activity › Forums › Adobe Premiere Pro › Premiere Pro CC Export Speed
-
Premiere Pro CC Export Speed
Posted by Luke Pearson on August 9, 2013 at 3:21 pmJust playing around with exports from Premiere Pro CC and I’m thinking they’re not very fast considering the hardware it’s running on.
The system is a HP Z820 with 2x 2.6Ghz 8c processors and an Nvidia Quadro 6000 card, 1 HDD for the OS and a 3x HDD Raid 0 (~400MB/s) array for Media running Windows 7 x64 iwht up to date drivers and updates.
I’m exporting a 1920×1080 ProRes (HQ) file to a 1920×1080 AVCI100 MXF file and a 45 minute programme is taking around one hour to export. I’m not making any edits, simply taking a ProRes, dropping it onto the timeline and then exporting it.
I’d have thought this would be faster to export, at least faster than run time or does this sound about right? Interestingly looking at the CPU usage, it’s at around 3% so the Quadro card must be doing pretty much all of the processing.
Luke Pearson replied 12 years, 6 months ago 6 Members · 12 Replies -
12 Replies
-
Kevin Monahan
August 9, 2013 at 5:21 pm[Luke Pearson] “I’m exporting a 1920×1080 ProRes (HQ) file to a 1920×1080 AVCI100 MXF file and a 45 minute programme is taking around one hour to export.”
Hi Luke,
If you are transcoding from one format to another, that doesn’t seem too outrageous. Can you tell us your settings in AME? Maximum Quality enabled, etc.?Thanks,
KevinKevin Monahan
Social Support Lead
Adobe After Effects
Adobe Premiere Pro
Adobe Systems, Inc.
Follow Me on Twitter! -
Ericbowen
August 9, 2013 at 6:01 pmIf your not scaling or interpolating since there are no effects the GPU acceleration is not handling much:
https://blogs.adobe.com/premierepro/2011/02/cuda-mercury-playback-engine-and-adobe-premiere-pro.html
BTW Quicktime will slow you down on the transcode since the player is only 32bit in Windows and it doesn’t thread as well as Adobe’s Media player.
The Quadro 6000 does have significantly lower specs than most of the current GPU’s. You might want to reference that comparison here:
https://www.nvidia.com/object/product-quadro-6000-us.html
GPU Specs:
NVIDIA Quadro GPU Quadro 6000
CUDA Cores 448
Form Factor 4.376” H x 9.75” L / Dual Slot
Gigaflops (Single Precision) 1030.4
Gigaflops (Double Precision) 515.2GPU Memory Specs:
Total Frame Buffer 6 GB GDDR5
Memory Interface 384-bit
Memory Bandwidth (GB/sec) 144 GB/shttps://www.geforce.com/hardware/desktop-gpus/geforce-gtx-780/specifications
GTX 780 GPU Engine Specs:
2304CUDA Cores
863Base Clock (MHz)
900Boost Clock (MHz)
160.5Texture Fill Rate (billion/sec)GTX 780 Memory Specs:
6.0 GbpsMemory Speed
3072 MBStandard Memory Config
GDDR5Memory Interface
384-bitMemory Interface Width
288.4Memory Bandwidth (GB/sec)Eric-ADK
Tech Manager -
Alex Gerulaitis
August 9, 2013 at 6:56 pm[Luke Pearson] “it’s at around 3% so the Quadro card must be doing pretty much all of the processing.”
Unlikely – like Eric said, the bottleneck is likely in the QT32 codec.
You can fire up HP Performance Advisor (already installed or a free download from HP) and keep an eye on GPU utilization. Also may be helpful to take a look at Resource Monitor (Accessible via Windows Task Manager) while exporting to see what your disk utilization is. If you’re on Windows 8, Task Manager will show disk utilization as well.
-
Dennis Radeke
August 9, 2013 at 9:22 pmone key piece of information you didn’t provide was how much RAM was in your system. If you like car analogies – The CPUs are the engine and the RAM is the gasoline. You can have 8 cylinders of raw engine, but if you’re giving it a straw to drink gasoline with, it will only go so fast.
That said, 45 minute program in about an hour given that it is a straight transcoding job isn’t horrible. And yes, QT32 could be a problem.
-
Luke Pearson
August 9, 2013 at 9:32 pmThanks for the replies, the system has 16GB of RAM but only around 4GB is being used during the export so the 32bit QT must be the problem.
I have a linux based encoding system that has 2x 2.9Ghz 6c processors and that can do a similar transcode around 3-4 times faster than run time but that utilises a 64 bit QT library.
I was hoping to move away from that and use this Premiere system instead but guess I’ll just use that to convert everything to AVC-Intra first.
-
Ericbowen
August 9, 2013 at 10:35 pm16GB of ram is really low to feed 32 threads especially since CUDA uses so much ram because of the way it works. I would highly suggest you get that up to atleast 32GB when you are able.
BTW another option is to use Cineform instead of AVCI. The transcode times with Cineform are really low compared to standard transcoding. Trial is here:
https://cineform.com/products/gopro-cineform-studio-premium
Eric-ADK
Tech Manager -
Paul King
August 10, 2013 at 12:41 amHi Luke
Sorry but you have received a big bunch of not very helpful answers.
Try this, switch CODEC to MPEG2 and retest. You will see your system resources used much more heavily.
I have the following spec:
Dual 3.1Ghz 8 core
64GB RAM
GTX TITAN
40TB RAID6 – 2300MB/secI cant encode QT or H.264 much faster than you.
AME or Premiere direct encode can only use multiple CPUs well with some CODECs. Unfortunately QT and H.264 are two that have poor multithreading in AME.After Effects – has multithreaded rendering (good when it works and doesn’t crash).
Premiere Pro – multithreading hit and miss depending on CODECs.
3D apps – great multithreading
Various AE plugins – some are great and other are poor.I’m really glad you brought up this issue because software has been behind the available hardware for a very long time. Only 3D apps have been giving us value for money on our hardware.
However GPU acceleration has complicated the issue. Premiere is pretty good at using GPU when preview rendering but so during encode (which I thought it was supposed to do). I put a Titan in for Premiere and AE and got significant benefit for GPU based stuff. Sapphire is great at using both GPU and CPU – the best I’ve seen. If your doing heavy AE stuff a Titan is now mandatory for the 6GB or RAM (my GPU plugins regularly use 2.5GB).
-
Luke Pearson
August 12, 2013 at 3:05 pmThanks for your response, the QT decoding is clearly the problem here then, as I said before the processor use is less than 5% during the export and the RAM usage holds at around 4GB, so again down to the 32bit QT application.
Very disappointing really as I was expecting the system to fly through this sort of processing.
-
Paul King
August 13, 2013 at 1:29 pmI dont understand why that should impact H.264 encoding which is independent of Apple QT.
BTW – I found a speed boost from the GTX Titan, even for H.264 encoding.
-
Luke Pearson
August 14, 2013 at 9:07 pmAfter some more research and going from the same project to other codecs it looks like the problem lies with the AVC Intra encoder. I can go to DNxHD, XDCAM, Uncompressed and others at about 3-4 times faster than run time coming from the same ProRes source file. These codecs also max out the ram usage and the CPU runs at around 50%.
Reply to this Discussion! Login or Sign Up