Quote
Jeff - Thanks I fell behind on your extensive GPU testing. Thanks for bringing me up to date with links to posts I missed.
Seems NLE software design varies a lot.
My brief experience is as follows:
Edius was always CPU based and uses more available memory. Ver 8 claims much faster render times since they added acceleration with Intel HD Graphics. When I tested 8 I could edit 2 layers of 4k on my i5 once I upgraded to 16gb memory with smooth HQ preview.
DaVinci Resolve is very GPU hungry. It supports up to eight GPUs on a PC - and chokes on my poor little i5.
Not sure how Premiere handles GPU for editing/render.
Just observations - not benchmarks.
Surely Cyberlink's claim as the fastest render engine in the industry is subjective depending on effects, input/output formats etc?
My plan was to make the shift to 4k with PDR15 and a PC upgrade but not sure PDR is ready yet.
I noticed in your one example, you are transcoding from 2K to HD for editing. Do you turn off shadow files? Does it help for 4k editing?
I was really hoping Cyberlink would have done more to improve edit and render with GPU power based all all the info you have provided. I sometimes get the impression you have put more effort in than they have.
Thanks again,
Al
AIS, DaVinci Resolve is very GPU hungry by design as it's a GPU based code. Very different approach than what PD is doing, yes any iGPU or even mid range GPU would struggle with such, for that utility one really needs and can take advantage of one or more high end GPU's for effective editing. That discussion probably more approriate for a different board.
I don't know what type of test case substantiates "fastest" render engine in the industry PD claim as you ask, I've not seen the timeline or a project published by CL for such claims. I do use other NVENC encoders and they do beat PD on simple transcoding, hands down. When a software has release to release changes that resulted in 2X speed improvements on the same hardware, I’m sorry, one simply has to have massive coding shortfalls that were not properly debugged and optimized in previous release.
http://forum.cyberlink.com/forum/posts/list/49577.page#260732 How one can claim XX% faster when one has 2X factors floating in your own backyard is pretty strange to me. I can also see "fastest" being very easy to put together a timeline that would significantly bias one product over another as well. I honestly don't give much credence to the advertisement jargon, what matters to me is how the utility works for me and if the utility operates as the released features and user selectable options imply it should. My experience with PD has been less than stellar from this view, user editing results don't match released features, especially new releases. Often it appears similar old issues are churned yet again and again in a new release.
I don't use PD shadowfiles. However, I often need to transcode from a higher quality source to HD for distribution to various outlets, not for editing as you noted.
A case of simple transcode was shown here for a given user needs,
http://forum.cyberlink.com/forum/posts/list/49715.page#261667 for PD15 and I've added “Software B” below utilizing the same GTX1070 NVENC encoder, obviously PD nothing near the “fastest” for this. Overall filesize, produced file stats and visual quality appear very similar as they should. However, in other posts
http://forum.cyberlink.com/forum/posts/list/49983.page#262922 user’s claim absolutely no surprise for PD14 being the fastest, I’m not sure why or how such a claim other than a very limited viewpoint perhaps? I simply see a mix, for some encoding tasks it’s in the mix or advantage, for others, clearly behind, so I’m not sure what constitutes the “fastest” claim.
Source: MPEG2 1920x1080/60i 25Mbps
Target 1: H.264 MP4, 1920x1080/30p 16Mbps (Default PD profile, avg video bitrate 15.5Mbps)
The table below lists [encode ratio, filesize(MB) , Produced video bitrate (Mbps)] encode ratio=timeline duration/encode time so the higher the better. A ratio of 10 would mean one can encode a 10min timeline duration in 1 minute.
| GPU | PD15 Target 1 | Software B |
| GTX1070, 372.90 | 10.0 , 1166 , 15.5 | 22.7 , 1260, 15.5 |
"Software B" being more than 2.2x faster than PD15 for this simple transcode.
Jeff
This message was edited 4 times. Last update was at Oct 25. 2016 21:01
![[Post New]](/forum/templates/default/images/doc.gif)
Oct 09, 2016 08:46
![[Avatar]](/forum/templates/default/images/avatar.jpg){kind=avatar}
Member
Joined: Jan 20, 2015 01:34
Messages: 136
Offline
