I Wrote:

I somehow have a hunch that GTX7xxx cards (or any graphic card) don't buy anyone much in PD12 (assuming an i& CPU) since the work is mostly done in the CPU's HD4600 in OpenCL

That statement is wrong. I did a quick test and found that some (NVIDIA supported) effects will make the GTX770 GPU work pretty good: I applied "Pencil Sketch 2" to the Nature.wmv clip (drop on clip) and then added "Color Painting" in the effects track. And voila, the GTX770 GPU-load is around 70%. Full HD real time pre-view was a tad choppy. Went to HD preview....smooth! GPU down to 30%. :


Then tried rendering with hardware and OpenCl on(using only one monitor):
1m08 with GTX770
Around 3min with HD4600 (had GTX770 disabled)


Lesson learned tonight: The benefit you get from your GPU depends very much on your type of effect usage of PD12.

More to come .......

Yes, a dedicated video card will be utilized when you use effects that are optimized (ATI or nVidia), but 99.9% of the time, I don't use effects. In my case, a dedicated video isn't used much at all. When I render MP4 (cuts only), my ATI may spike at 10%, but normally around 3%. I see no difference in render speed using my HD7870 or the Intel HD4600. In fact with MP4, the HD4600 is a few seconds faster. Intel i7 4770k, 16GB, GTX1060 3GB, Two 240GB SSD, 4TB HD, Sony HDR-TD20V 3D camcorder, Sony SLT-A65VK for still images, Windows 10 Pro, 64bit
Gary Russell -- TN USA

Quote: Holy Mackerel! I've been using PD since version 9 and have NEVER checked that box because of the warning that comes up about not being standards-compliant (see attached png). At your suggestion, I enabled it and the SVRT produce took 60 seconds! That's more than twice as fast as the GPU can do it and almost 3.4x faster than SVRT with the box unchecked. Wow.

60 seconds is even high. That's high because the clip "9-10 bars1.MTS" used in your timeline is different from the others from a video format perspective. Depending on what profile you use in produce, either this clip or the others will be CPU encoded when SVRT is used. This will cause your 60 seconds, if clip "9-10 bars1.MTS" was of the same video properties of the others, the SVRT time would be under 30sec vs your 60sec. So I guess, Holy, Holy, Mackerel!!

The hickup in playback is a age old problem with SVRT and playback on some players. No I do not think it's directly related to the IDR setting. IDR setting required for footage of many Canon's.

I'm virtually 100% certain your 3 pics, GPU-Z - Hardware - OpenCL ON.png, GPU-Z - SVRT OpenCL OFF.png, GPU-Z - Hardware OpenCL OFF.png don't actually represent those settings. Based on the performance charts it's virtually impossible. You may want to redo for your own understanding. Since it's not PD related, I'll refrain from discussion here.

Quote: I see no difference in render speed using my HD7870 or the Intel HD4600. In fact with MP4, the HD4600 is a few seconds faster.

A little surprised on the seconds comment, I'd assume out of a 1/2hr show? Can you attach 2 pics, one of the timeline with the effect room selected and one of your "Produce" page when using the HD4600. Maybe something will register.

Earlier you mention that you take TV shows, clip here and there, and produce. If you are thinking Multi-GPGPU in PD is going to reduce produce time of such timeline, based on what I know, I think with the current implementation you maybe incorrect. I've mentioned this several times in the forums and produced a fairly lengthy pdf that I attached to one of your previous posts describing why.

I also posted both a HD4000 and a GTX580 working together here. http://forum.cyberlink.com/forum/posts/list/15/31250.page No tricks that I know of, just basic setup, iGPU needs to be prime and timeline needs to have content that utilizes Multi-GPGPU.

Jeff

Your PDF is informative on results, but doesn't say how you setup the hardware. I connect my monitor to the HD4600 port, my HD7870 is essentially not running at all. Not matter what I do, GPU-z says there is zero GPU activity.

If I connect the HD7870 to the monitor, the HD4600 has no activity. At this point, I can only assume that two monitors are required.

I don't use Effects at all or very rarely. Mostly cuts and some fades. When I render to MP4, GPUz indicates a nominal 2-3% GPU usage using the HD7870 if at all. I would have to retest the HD4600 to see what GPUz reports.

Perhaps if I used the same clips/effects that you did, most likely would give different results. Intel i7 4770k, 16GB, GTX1060 3GB, Two 240GB SSD, 4TB HD, Sony HDR-TD20V 3D camcorder, Sony SLT-A65VK for still images, Windows 10 Pro, 64bit
Gary Russell -- TN USA

Quote: Your PDF is informative on results, but doesn't say how you setup the hardware. I connect my monitor to the HD4600 port, my HD7870 is essentially not running at all. Not matter what I do, GPU-z says there is zero GPU activity.

From what I have seen, the secondary card will only show activity during render if timeline has a supported effect. Can be much more than a few percent as I had shown, 28% of full throttle GPU clock on the GTX580 is a significant load from PD.

Quote: If I connect the HD7870 to the monitor, the HD4600 has no activity. At this point, I can only assume that two monitors are required.

I only have a single monitor setup, so no, that is not true. I don't have a HD7870, but monitor load would most likely show up on "Video Engine load" bar vs "GPU load". That's card dependent so test for your card.

Quote: I don't use Effects at all or very rarely. Mostly cuts and some fades. When I render to MP4, GPUz indicates a nominal 2-3% GPU usage using the HD7870 if at all. I would have to retest the HD4600 to see what GPUz reports.

From what I have concluded on my box and testing, I personally wouldn't worry about Multi-GPGPU at this stage then for your typical timeline, maybe in a future PD version, who knows.

Quote: Perhaps if I used the same clips/effects that you did, most likely would give different results.

In the pdf I gave a timeline with only PD supplied clips so anyone could compare there own system if they so desired.

Jeff

Quote:

Quote: Holy Mackerel! I've been using PD since version 9 and have NEVER checked that box because of the warning that comes up about not being standards-compliant (see attached png). At your suggestion, I enabled it and the SVRT produce took 60 seconds! That's more than twice as fast as the GPU can do it and almost 3.4x faster than SVRT with the box unchecked. Wow.

60 seconds is even high. That's high because the clip "9-10 bars1.MTS" used in your timeline is different from the others from a video format perspective. Depending on what profile you use in produce, either this clip or the others will be CPU encoded when SVRT is used. This will cause your 60 seconds, if clip "9-10 bars1.MTS" was of the same video properties of the others, the SVRT time would be under 30sec vs your 60sec. So I guess, Holy, Holy, Mackerel!!

The hickup in playback is a age old problem with SVRT and playback on some players. No I do not think it's directly related to the IDR setting. IDR setting required for footage of many Canon's.

I'm virtually 100% certain your 3 pics, GPU-Z - Hardware - OpenCL ON.png, GPU-Z - SVRT OpenCL OFF.png, GPU-Z - Hardware OpenCL OFF.png don't actually represent those settings. Based on the performance charts it's virtually impossible. You may want to redo for your own understanding. Since it's not PD related, I'll refrain from discussion here.
Jeff

A couple thoughts:

The project has clips from 2 cameras, only one of which can record at 60p. The project is set to 60fps for the main camera and it makes sense that SVRT would run maybe twice as fast if it didn't have to process that single 30p clip. And yes. Holy, Holy Mackerel!!!

With regard to the Single IDR setting with SVRT, the glitches that showed up when playing the output video in VLC at 3:40 and 3:48 also show up on YouTube, and that's a total deal killer as all of my videos go there If I uncheck Single IDR and run SVRT, the process takes much longer but there are no problems with playback in VLC or on YouTube. It appears - at least in this case - that the bitstream issue warning about checking that option is valid.

Finally, about your last comment about the PNGs not representing the settings - I assume you mean that they were all acquired without the Single IDR option enabled. That's true. However the slow production times do seem to be PD-related.

I would be thrilled to have performance like SVRT gives with the Single IDR option enabled, but since the produced video suffers from some significant glitches - at least for the kind of work I'm doing - I'm stuck with trying to find other ways to get PD to better utilize the resources on my new machine.

Thanks to you, I've learned that some serious improvements are already there and I'd really like to find something that brings that level of performance without the glitches.

YouTube/optodata

DS365 | Win11 Pro | Ryzen 9 3950X | RTX 4070 Ti | 32GB RAM | 10TB SSDs | 5K+4K HDR monitors

Canon Vixia GX10 (4K 60p) | HF G30 (HD 60p) | Yi Action+ 4K | 360Fly 4K 360°

Quote: The project has clips from 2 cameras, only one of which can record at 60p. The project is set to 60fps for the main camera and it makes sense that SVRT would run maybe twice as fast if it didn't have to process that single 30p clip. And yes. Holy, Holy Mackerel!!!

Yes, different fps and different bitrate. SVRT only works on "similar" clips to "similar" output profile.

Quote: If I uncheck Single IDR and run SVRT, the process takes much longer but there are no problems with playback in VLC or on YouTube. It appears - at least in this case - that the bitstream issue warning about checking that option is valid.

That occurs because when unticked the timeline is CPU encoded, SVRT not even used hence the longer times. Keep in mind SVRT when used on the timeline video is not doing any encoding, simply copying video stream if you will. CPU encoding is probably the most robust encoding (least artifacts high quality) that PD offers, often better the HA (depending on hardware) or SVRT.

Jeff

One think to remember is with adding effects and color setting gives the computer a lot more work to do. I have a friend I talk to on the web at times. He does vfx and 3D stuff. He work on Spider man 2 and stuff like that. He has a render farm setting in a room. He told me that some stuff he does for a few second clip can take that render farm 30 hours to render out. Intel i7 5960X overclock to 4 Ghz 16 GB of ram.
GoPro 4
Canon VIXIA HF G10
Canon EOS Rebel T3
Canon EOS 70D
My Vimeo Channel http://vimeo.com/user3339631/videos

If it helps anyone, I found the specs for the HD4600 here. Compared to the GeForce760 (GTX 760) I'd say don't use it. For one thing, these Intel cards tend to share RAM, which isn't such a good idea for rendering.

I base cards specs on two main features, the memory bandwidth and the bit/bus.

Compare the two and you see the Geforce is miles ahead, 192Gb/s bandwidth riding on a 256bit bus (memory interface width). The intel has 26Gb/s riding on a 128 bit bus. That for me alone would make the decision for me to buy the Geforce.

Everything else is just gravy. The more memory on the board the better. The higher the clock cycles the better.

Jimbo,

if the software does not use the Graphics Card (GC) , then it is pretty much a mute point. PD12 uses the GC to some degree, but not as much as you would think.

Most of my typical rendering for PD12 (as of now) happens on the actual CPU. The GC does barely get utilized. So if I had to build a system now (for my use patterns) I would put my money on a faster CPU vs a GC.

Now this may change in the future if CyberLink implements heavy GC processing. But I doubt this will happen, since different GCs will likely render slightly? different video quality. I can see why CyberLink heavily relies on the CPU for rendering: The code and the video results are identical, portable and 100% reproducible. That is a must for serious production.

Of course if your video needs rely heavily on certain effects where these effects do support GC processing, then money put on a GC is a good thing. Otherwise its a waste.

Hi Hellebauer,

I understand what you said but as much as PD might not share workload with a GPU, that video data still has to run up and down between CPU and GPU. That's where the bandwidth comes in useful.

Having a wide bus and plenty of bandwidth is the difference between having a hose-pipe to let the water run through and a drainage pipe. Like a country lane and a motorway. It still helps.

As you say, the GPU may not be working too much but the graphic data is still a lot to be transferred between CPU and GPU.