Graphics Cards

[Michael C. Ray]

Is there a section / page / forum discussing Graphic Cards?

Thanks Youts

[Andrew Smith]

Nope. We don't really yearn for greater powered graphics cards like in the old days.

Andrew

[Seth Bloombaum]

Well, I sure wish 4k and 8k performance was better supported on Premiere/Mac, and better graphics cards would help! Just like the old days, I'm having to learn proxy workflows and other tricks...

I'm just starting to benchmark our worst-case test footage. The first datapoint is that it takes 10 hours to export 1 minute of 7680x7680 stereoscopic 360 footage with color correction and denoise from Premiere / iMac. This footage is from our highest-resolution camera, and I'll be benchmarking 4k as well as other platforms at my college that are on the list for testing.

10 hours for 1 minute? That's ridiculous! Perhaps much worse than what we were experiencing in the early days of computer-aided editing and early NLEs... hard to remember what we were doing with Premiere 3 in the 1990s, but I do remember hours-long exports for short-form clips at 320x240 resolution.

Having said that, using Premiere's rather mature proxy workflow is pretty slick for edit decisions.

P.S. Suggestions of cards that support h.264 / h.265 hardware acceleration at beyond-HD resolutions would be most welcome! Cost is also important, we're supporting about 80 workstations.

[John Nantz]

Way overdue for a number of upgrades for my computers and a graphics card one of the items. My modus operandi is to use older computers so upgrades is the name of the game. Memory, better back-up system (not just equipment but also workflow) and storage devices in general. Have a 2008 (early and a 2012 Mac Pro Classics with the '08 retaining legacy apps and files and the '12 needing to go metal.

Been busy doing construction the past year and a half and just haven't had the time to do much of anything with video, unfortunately. Hence the need to do upgrades but it takes so long to do the research and graphics cards is a though area. Currently running Radeon 5770s zzzz but there are so many options to choose from.

[Chris Hurd]

Best sub-forum for this topic is probably https://www.dvinfo.net/forum/non-linear-editing-pc/

I think it's an excellent subject for discussion, because I'm in the market for one as well.

Going to do an AMD build and would like to keep it all-AMD but I've got my eye on the 1660 Super or 1660 Ti, so there goes that idea. Can't really wait around for the 5500 or whatever's next from AMD.

Maybe it'll be an RX 580 or 590 as those prices are pretty good right now.

[Michael C. Ray]

Im looking at the GTX 2080 Super or GTX 280Ti - it seems these have been optimized for RR and DR and some are touted as studio cards ... Hince my question - I am a basic video guy, simple projects no movie or broadcast work ......

[Michael C. Ray]

Originally Posted by Andrew Smith

Nope. We don't really yearn for greater powered graphics cards like in the old days.

Andrew

Andrew can you elaborate on that as to why? just a quick short reason would be appreciated

[Andrew Smith]

Forgive me, for I am still happily in XDCAM HD wonderland. Not feeling the need on a daily basis.

Computing power has come such a long way in the last 10 years that I'm somewhat surprised it's still an issue for 4K video. 8K (which I consider to be completely over the top) is understandable but not 4K.

Andrew

[Steve Game]

Originally Posted by Andrew Smith

Forgive me, for I am still happily in XDCAM HD wonderland. Not feeling the need on a daily basis.

Computing power has come such a long way in the last 10 years that I'm somewhat surprised it's still an issue for 4K video. 8K (which I consider to be completely over the top) is understandable but not 4K.

Andrew

Because video editing software is making increased use of GPUs (via CUDA or Open CL) for encoding of video (NVEnc) and making colour correction and grading changes to footage. The highly parallel integer architecture of GPUs is better suited to bitmap images than a complex floating point CPU.

An editing workstation can have one or more graphics adaptors to provide the correct level of parallel processing, leaving the CPU (if it has an integral graphics engine) to manage the GUI.

Decompressing highly compressed video at UHD resolution is at the level of complexity that CPUs can start to struggle, especially with HEVC.

[Christopher Young]

Originally Posted by Seth Bloombaum

Well, I sure wish 4k and 8k performance was better supported on Premiere/Mac, and better graphics cards would help! Just like the old days, I'm having to learn proxy workflows and other tricks...

Maybe this workflow may work for you. With heavy duty requirements like you are outlining I just run a script to export the finalized timeline be it single layer, multi track composite or whatever as a series of max resolution, 4K or whatever single frame stills as JPGs PNGs or whatever format you want. I then re-import the stills as a numbered sequence. I then place them back on a timeline and then render the timeline out to whatever destination format I wish. Way, way faster than any other process I used. Works fine with good results for me here.

Chris Young

[Seth Bloombaum]

Originally Posted by Steve Game

...Decompressing highly compressed video at UHD resolution is at the level of complexity that CPUs can start to struggle, especially with HEVC.

One of the things a video card provides is hardware decoding of h.264 video. I’ve seen specific chips on GPUs dedicated to this. If the decode is running in hardware it is not adding to the load of the CPU or perhaps even much of the GPU.

This and hardware encoding of h.264 is part of what Premiere’s Mercury Playback Engine, and other NLEs recognize as GPU support.

Older computers (like our lab macs) offer h.264 encode/decode at HD resolutions. I need to test 4k resolutions. I’ve established that ours don’t support 8k in hardware, and may be sending all the tasks to the CPU.

And now drones & other devices are offering h.265 modes, I’m hopeful that later GPUs are catching up with h.265 chips...

Originally Posted by Christopher Young

Maybe this workflow may work for you...

That’s an interesting approach! I’ve been working through the various approaches to proxy workflows, I’ll add this to the list for benchmarking.

I guess that’s similar to an intraframe codec... there may be disk speed challenges? No, if I’m reading you right you’re only doing this at render time?

[Christopher Young]

Yes Seth it is basically an I-Frame approach.

An example in HD as those numbers are fresh in my mind from a recent project. A recent one hour HD job that had a Neat Video NR plugin applied along with Red Giant's Cosmo II plugin took just over four hours to render. Any of us who have worked with Neat Video know that it is an incredibly GPU and CPU intensive noise reduction plugin as brilliant as it is. Using the still image sequence export method the stills were exported in a little under real time. So less than the hour timeline. I nearly always use Cineform as my working codec. The render of the finished one hour timeline to Cineform from the stills sequence took twenty eight minutes. Total round trip export/import/render time therefore was under ninety minutes as opposed to over four hours on the timeline render from ProRes to Cineform. A saving of over two and a half hours.

The workflow then is to import the new Cineform render to the original master timeline mute the original video tracks line up the Cineform render with the audio bed, it's frame accurate so that's easy, then render out to MP4.

If the project is massive you could export say ten minute sequences and then stitch them together for the render. I see you mention Vegas. In Vegas just go to Tools/Scripting and you will find the "Render Image Sequence" script. Vegas gives you the choice of JPG or PNG. PNG having the larger higher quality frame is my choice.

Comparing CPU/GPU performance on that particular project was as follows. Rendering ProRes with the plugins to Cineform was using around 50-55% GPU and about 60-70% CPU. Doing it the stills sequence way on the stills export the GPU didn't break 12% and the CPU about the same. On the Cineform render the GPU sat between 15-17% and the CPU sat at around 13-14%. This is on an i78700K OC'd to 5GHz with a GTX 1080Ti. In other words the system hardly broke a sweat and with CPU core and GPU temps between 45 and 55 degrees. How you would go with 4 and 8K is anyone's guess? Suck it and see I guess!

EDIT:
You got me to thinking. I haven't used this still sequence workflow on anything over HD. So I tried one minute of 4K DCI 24p, 40196 x 2160 S-LOG3 with a 3D LUT. Exporting one minute of XAVC-I frame to JPGs took six minutes and sixteen seconds. Rendering that still sequence of JPG back to Cineform took two minutes and eight seconds. So a round trip of eight minutes and twenty-four seconds for one minute of 4K. A lot slower than the HD round trip but still pretty reasonable I guess.

But here's the rub. That same one minute of 4K XAVC-I with 3D LUT took only one minute and forty seconds to render directly to Cineform, video to video no stills workflow in other words. The GPU ran about 25% and the CPU ran at just on 40%. Obviously I guess it's a process of selecting one minute of timeline on whatever project you are working on and then running through the different workflows to work out the most efficient. I sure would be looking for something better than ten hours for a minute that's for sure:)

Chris Young