Announcement: Our new CyberLink Feedback Forum has arrived! Please transfer to our new forum to provide your feedback or to start a new discussion. The content on this CyberLink Community forum is now read only, but will continue to be available as a user resource. Thanks!
CyberLink Community Forum
where the experts meet
| Advanced Search >
How much does OpenCL (CUDA) acceleration speed up rendering?
odontomatix3D [Avatar]
Newbie Joined: Jun 17, 2012 21:23 Messages: 4 Offline
[Post New]
I have a CUDA nVidia GPU (GeForce GT 440) in an Intel i5-2300 2.80 GHz CPU computer (quad-core). PowerDirector 10 does use the hardware acceleration in some video output formats and not others. For example, mpg4 uses the CUDA hardware video encoder, but mpg2 doesn't. You also have to enable the hardware acceleration in the setup preferences of PD10. It's called "Enable OpenCL technology". OpenCL is "open cuda language" I think, something nVidia created to allow a standard software language for programming the interface between applications and the GPU cores.

Rendering speedup results: On my machine, I rendered a 13 minutes 28 second playing-time stereoscopic 1080p project video into a 1.3 GB mpg4 output file in 1080p side-by-side half-width format. The rendering took 9 minutes 8 seconds using the OpenCL acceleration. With the acceleration turned off, the same video required 15 minutes 28 seconds to render.

The GeForce GT 440 has 96 CUDA cores. Other newer, more expensive, GPUs from nVidia have many more cores, which may speed the rendering even more.

What is your speedup when you use OpenCL versus when you turn it off? Please report your experience here and be sure to describe your system on which the comparison was made and the type of video format rendered. This would be useful information for people considering upgrading their hardware to speed up rendering.
stevek
Senior Contributor Location: Houston, Texas USA Joined: Jan 25, 2011 12:18 Messages: 4663 Offline
[Post New]
Some people swear by hardware (CUDA) encoding and others don't. Some say that software rendering produces a clearer image.

Why do you want to encode videos that fast? Do you do a lot of them? Just wondering? .
.
BoilerPlate: To posters who ask for help -- it is nice to thank the volunteers who try to answer your questions !
Anything I post unless stated with a reference is my personal opinion.
odontomatix3D [Avatar]
Newbie Joined: Jun 17, 2012 21:23 Messages: 4 Offline
[Post New]
My wife produces a half-hour video episode each day, sometimes more than one per day. Her computer is slower than the new one I recently purchased for my 3D videos and described above. Hers is a Core 2 Duo 1.86 GHz CPU without a CUDA-capable GPU. Each episode takes hours to render in DVD quality. I wanted to consider upgrading her cameras to 3D stereo 1080p cameras. Rendering in stereoscopic 1080p would take a very long time. I was wondering whether a CUDA GPU would speed her rendering on that same machine, just by adding an nVidia graphics board. So I bought an inexpensive nVidia GPU for my system first to test it. She uses Pinnacle Systems Studio editor, which doesn't do CUDA accelerated rendering, so she'd have to learn a new editor to take advantage of the GPU. She hates changes (learning "new buttons"), so I haven't tried my GPU in her computer yet, but that was the motivation for this experiment.

Regarding the quality of the images, that's something I'll look at carefully on my 47" tv. I know that Intel argues that the video rendered in the GPU isn't (can't) be as good as video rendered entirely by the CPU, but for consumer-grade applications, I wonder whether that really matters. The article I read about the difference in quality had examples, but you had to look with magnification to see the difference. I will try to see for myself if I can tell the difference at 1080p from a normal viewing distance.

This message was edited 2 times. Last update was at Jul 02. 2012 13:24

stevek
Senior Contributor Location: Houston, Texas USA Joined: Jan 25, 2011 12:18 Messages: 4663 Offline
[Post New]
Here is where most of the problem is "Core 2 Duo 1.86 GHz CPU"

Here are the "normal" requirements (not blu-ray,etc.)

DVD Quality (MPEG-2) Profiles: Pentium 4 2.2 GHz or AMD Athlon XP 2200+
High Quality MPEG-4 and Streaming WMV, QuickTime) Profiles: Pentium 4 3.0 Ghz or AMD Athlon 64 X2
Full-HD quality H.264 and MPEG2 Profiles: Intel Corei5/7 or AMD Phenom II X4

and
AVCHD and BD burning Profiles: Pentium Core 2 Duo E6400, (Note 2.13GHZ speed) or AMD Phenom II X2
3D Video Editing Profile: Intel Core i7 or AMD Phenom II X4

You will get a marginal increase in encoding speed. Trade with her or buy her an early holiday present or (early or late) birthday present. Adding a better GPU is a minimum value patch. .
.
BoilerPlate: To posters who ask for help -- it is nice to thank the volunteers who try to answer your questions !
Anything I post unless stated with a reference is my personal opinion.
odontomatix3D [Avatar]
Newbie Joined: Jun 17, 2012 21:23 Messages: 4 Offline
[Post New]
Thanks for that info. I think what I'll do is show her how well my computer is working and see if I can motivate her to learn something new (PowerDirector). In any case, my i5 with GeForce GT 440 works great. I had been using a Core 2 Duo 2.13 processor with the GT 440. I was planning to give her that one, but the computer started doing random bad things, such as random re-boots and getting stuck while reading files, so I got the i5 and moved the GPU to it. Since rendering is a matter of number crunching and isn't a real-time application, although slow the Core 2 Duo was rendering my 1080p videos but slowly.
vn800rider
Senior Contributor Location: Darwen, UK Joined: May 15, 2008 04:32 Messages: 1949 Offline
[Post New]
I haven't done a comprehensive comparison in PD10 but certainly in earlier versions it was important to recognise that selecting the default HA also selected a different encoding profile, which may explain the apparent speed increase and also the perceived lowering of "quality".

Many folk find that the "quality" of encoding is better without HA, even though it may be slower.

This recent simple comparison may also be of interest, particularly where GPUs can be selected.

http://forum.cyberlink.com/forum/posts/list/22732.page

Your question assumes that all rendering is equal (so to speak) but, using the default profiles, I'm not sure that it necessarily is. Any comparison should really be done with the same profile to be truly comparative.

Cheers
Adrian Life is really simple, but we insist on making it complicated. (see below)
Confucius
AMD Phenom IIX6 1055T, win10, 5 internal drives, 7 usb drives, struggling power supply.
Powered by JForum 2.1.8 © JForum Team