Another user has the same issue with a GTX960. http://forum.cyberlink.com/forum/posts/list/49577.page My guess is PD15 is somehow misidentifying the GPU and providing the older CUDA selection. I don't know the correction, having not experienced the issue. Attached is a GTX970 in my system showing OpenCL for hardware acceleration as I believe it should be.

Maybe try selecting the clean install option of the Nvidia 372.70 install package with a custom install and only install the driver and see if you recover the OpenCL correctly.

Jeff

I was able to replicate your issue (see pic) and the suggestion above of reinstalling Nvidia drivers should remedy the issue.

Jeff

I tried to install driver three times but nothing change. PD15 still OpenCL. Now, I am on business trip. I will try installing only driver when I back to my office. Thank for your advices. Nguyen Trong Hieu

Quote: I tried to install driver three times but nothing change. PD15 still OpenCL. Now, I am on business trip. I will try installing only driver when I back to my office. Thank for your advices.

That is the correct operational mode, your CUDA note PC is not correct. If you reinstall Nvidia drivers on the CUDA note PC it will then offer the correct OpenCL note.

Jeff

yup, I see. OpenCL is the correct mode. But someone asked me the cuda mode Nvidia took render time shorter than OpenCL mode. So I want to try Cuda mode by gtx 1060 to compare OpenCL mode. Nguyen Trong Hieu

Hi Jeff,

How were you able to replicate the CUDA option?

Cheers,

Andrew Alienware Aurora ALX R4 - Intel i7-4820 4.2 GHz - 32GB DDR3 RAM - Crucial 512GB SSD - 1TB Seagate HDD - 3TB WD Green HDD - 4TB WD Green HDD - MSI NVIDIA GTX 1070 8GB

Sony HDR-PJ810 and HDR-PJ530

Quote: yup, I see. OpenCL is the correct mode. But someone asked me the cuda mode Nvidia took render time shorter than OpenCL mode. So I want to try Cuda mode by gtx 1060 to compare OpenCL mode.

From what I know, PD does not use OpenCL for basic encoding, it really only uses OpenCL for assisting with encode and playback of some applied Fx in your timeline.

The 1060 and the 970 use a specialized Nvidia ASIC IP block to perform the video encoding, this encoding block is called NVENC. Different versions of NVENC are in the two cards you reference with the 1060 having better encode performance within PD for some formats supported by both cards. This NVENC encoder essentially replaced the CUDA encoder like 4yrs ago or so and the CUDA encoder was removed from Nvidia driver packages like 2 yrs ago.

To use the NVENC capability on your 970 or 1060 with PD15, you active the feature on the "Produce" tab with the selction of Fast video rendering technology > Hardware video encoder.

Jeff

Jeff, but the CUDA/OpenCL option in the options menu is not related at all to the encoding process. That NVENC is selected only at the "produce" stage.
So we are still talking about the CUDA vs OpenCL in the timeline FX's.

SoNic67, when Hieuvietnam mentioned:
"But someone asked me the cuda mode Nvidia took render time shorter than OpenCL mode."
I took it that the real intent is to try and do a comparison test of the older CUDA render task vs current NVENC. NVENC wins hands down.

My apologies if I read a little logic into the intent. My view, it does not make a lot of sense to compare render time of CUDA vs OpenCL applied to Fx as I think for most people Fx occupy a very small overall duration in a typical timeline.

Jeff

Quote:

Quote: yup, I see. OpenCL is the correct mode. But someone asked me the cuda mode Nvidia took render time shorter than OpenCL mode. So I want to try Cuda mode by gtx 1060 to compare OpenCL mode.

From what I know, PD does not use OpenCL for basic encoding, it really only uses OpenCL for assisting with encode and playback of some applied Fx in your timeline.

The 1060 and the 970 use a specialized Nvidia ASIC IP block to perform the video encoding, this encoding block is called NVENC. Different versions of NVENC are in the two cards you reference with the 1060 having better encode performance within PD for some formats supported by both cards. This NVENC encoder essentially replaced the CUDA encoder like 4yrs ago or so and the CUDA encoder was removed from Nvidia driver packages like 2 yrs ago.

To use the NVENC capability on your 970 or 1060 with PD15, you active the feature on the "Produce" tab with the selction of Fast video rendering technology > Hardware video encoder.

Jeff

Hi Jeff,

Final, just active Fast video rendering technology to use NVENC ecording. Will the PD15 operate best performance? Thank you Nguyen Trong Hieu

Hieuvietnam, yes the screen in your pic background with the checkmark in Fast video rendering technology: > Hardware video encoder instructs PD to use your GTX1060 GPU for encoding during "Produce" functions and will result in faster encode times. When unchecked, your CPU does the encoding and will be significantly slower compared to using your GTX1060 GPU for supported profiles.

Jeff

I've tested render speeds with my GTX1070 GPU and, on my PC, render speeds are 5% faster with just 'Enable hardware decoding' checked as opposed to both 'Enable OpenCL...' and "Enable hardware...' ticked.

I should add that this is when converting a 40 minute 1920x1080 m2ts 28mbps 50p project to MP4 1920x1080 25p 16mbps file in 'Produce' with the 'Fast Rendering Technology: Hardware Video Encoder' ticked. It's 12min 47sec with just "Enable Hardware..' ticked and 14min 40sec with both ticked. Not a huge difference.

It may be very different with different source material on a different PC with a different GPU etc.

Andrew Alienware Aurora ALX R4 - Intel i7-4820 4.2 GHz - 32GB DDR3 RAM - Crucial 512GB SSD - 1TB Seagate HDD - 3TB WD Green HDD - 4TB WD Green HDD - MSI NVIDIA GTX 1070 8GB

Sony HDR-PJ810 and HDR-PJ530

Quote: I've tested render speeds with my GTX1070 GPU and, on my PC, render speeds are 5% faster with just 'Enable hardware decoding' checked as opposed to both 'Enable OpenCL...' and "Enable hardware...' ticked.

I should add that this is when converting a 40 minute 1920x1080 m2ts 28mbps 50p project to MP4 1920x1080 25p 16mbps file in 'Produce' with the 'Fast Rendering Technology: Hardware Video Encoder' ticked. It's 12min 47sec with just "Enable Hardware..' ticked and 14min 40sec with both ticked. Not a huge difference.

Yes, that benefit was mentioned here: http://forum.cyberlink.com/forum/posts/list/49577.page#260731 and put into the small encoding benefit label as you note, not a huge difference. If you have a need to use CPU encoding for some projects, the GPU decoding will probably provide even more benefit than this task of 5% with your i7-4820. At PD15 it will help unload your i7-4820 so then it's more efficient at encoding.

As long as you don't have any accelerated effects used in the timeline, the Enable OpenCL pref option being ticked or not will make no difference.

Jeff

Thanks Jeff.

I used 'TechPowerUp GPU-Z' to monitor my GPU alongside 'Windows Task Manager' to monitor PD15's CPU usage when completing the tests.

With both 'Enable OpenCL...' and 'Enable hardware...' ticked, CPU approx. 43% Video Engine Load approx. 35% and GPU load approx. 4%

With just 'Enable hardware...' ticked, CPU approx. 23% Video Engine Load approx. 55% and GPU load approx. 17%

Very interesting.

Andrew Alienware Aurora ALX R4 - Intel i7-4820 4.2 GHz - 32GB DDR3 RAM - Crucial 512GB SSD - 1TB Seagate HDD - 3TB WD Green HDD - 4TB WD Green HDD - MSI NVIDIA GTX 1070 8GB

Sony HDR-PJ810 and HDR-PJ530

Andrew, the attached chart summarizes my thought referenced earlier and is pretty self-explanatory. I tried to replicate your source and target formats in the conversion, 1920x1080 50p 28Mbs m2ts to MP4 1920x1080 25p 16Mbps, however I only used 10min duration for simplicity. I also have pure transcoding, no timeline edits. The realtime ratio below is just the (timeline duration)/(encode time) ratio, so a factor 2.0 would imply one can encode a timeline 2 times faster than the timeline duration.

Encoder	Decoder	Encode Time (s)	Realtime Ratio
CPU	CPU	252	2.38
CPU	GPU	147	4.08
GPU	GPU	95	6.31
Andrew GPU	Andrew GPU		3.13

As can be seen, unloading the CPU from decoding task when one desires to use CPU encoding can be very beneficial. It allows the CPU encode task to be nearly 2 times faster. The benefit depends on CPU and GPU capability, quality of timeline content, and target format so results will vary for each user substantially. The issue of some formats not supporting timeline GPU decoding highlighted here http://forum.cyberlink.com//forum/posts/list/25/45503.page#236759 has been PARTIALLY corrected and extended in PD15.

From your posted CPU loads, I'd assume you may have Hyper-threading activated which will create a artificially low perceived load on the CPU during encoding but still a good relative assessment. Overall Hyper-threading does improve overall encode performance some so having it activated is not a bad thing.

Jeff

Hi Jeff,

Very interesting indeed. My test footage had one text overlay at the start and one fade at the end. Removing those decreased the render time with GPU/GPU to 9min 53sec which takes the realtime ratio up to 4. My test footage also had 20 clips in.

Thanks for such clear and well explained data Jeff. It'd be interesting to see other peoples' results with different source footage and GPUs etc.

Andrew Alienware Aurora ALX R4 - Intel i7-4820 4.2 GHz - 32GB DDR3 RAM - Crucial 512GB SSD - 1TB Seagate HDD - 3TB WD Green HDD - 4TB WD Green HDD - MSI NVIDIA GTX 1070 8GB

Sony HDR-PJ810 and HDR-PJ530

Quote: I've tested render speeds with my GTX1070 GPU and, on my PC, render speeds are 5% faster with just 'Enable hardware decoding' checked as opposed to both 'Enable OpenCL...' and "Enable hardware...' ticked.

I should add that this is when converting a 40 minute 1920x1080 m2ts 28mbps 50p project to MP4 1920x1080 25p 16mbps file in 'Produce' with the 'Fast Rendering Technology: Hardware Video Encoder' ticked. It's 12min 47sec with just "Enable Hardware..' ticked and 14min 40sec with both ticked. Not a huge difference.

It may be very different with different source material on a different PC with a different GPU etc.

Andrew

Andrew (and Jeff), I am seeing the exact same thing when I monitor Video Engine Load which, to me, is pretty confusing. How does it make sense that encoding runs faster when you do NOT check Enable OpenCL?

With these new GPU beasts (I just purchased the GTX1060 based on what I have been reading in these forums - thanks!), I suspect a lot more people are going to want to maximize the usage of those beasts. Maybe Cyberlink needs to rethink how they communicate what is actually happening when you check or don't check certain boxes.

Hi Rick,

Jeff is far wiser than me on the technical aspects. What I would say is, until I did my tests, I was under the impression that both boxes ticked was giving me the best speed.

I suspect it has a lot to do with what you are rendering to start with, what enhancements you have made, and what your output is.

Cheers,

Andrew Alienware Aurora ALX R4 - Intel i7-4820 4.2 GHz - 32GB DDR3 RAM - Crucial 512GB SSD - 1TB Seagate HDD - 3TB WD Green HDD - 4TB WD Green HDD - MSI NVIDIA GTX 1070 8GB

Sony HDR-PJ810 and HDR-PJ530

I have a Nvidia GTX560 display card, I cannot see CUDA acceleration, only OpenCL in hardware acceleration. Please let me know how to enable CUDA for my PD15. Thanks.