• Global community
    • Language:
      • Deutsch
      • English
      • Español
      • Français
      • Português
  • 日本語コミュニティ
    Dedicated community for Japanese speakers
  • 한국 커뮤니티
    Dedicated community for Korean speakers
Exit
0

Why is my render not making more use of my GPU?

Explorer ,
Jan 29, 2016 Jan 29, 2016

Copy link to clipboard

Copied

We recently upgraded from Windows 7 to Windows 10 (clean install), added two SSD drives (one as a system drive, one for Adobe cache, etc.), and the system has a 10TB RAID for media.  The system is a dual proc Xeon machine.  We also upgraded from the Quatro 4000 to the M5000 (properly connected and supplied with the additional power required, latest drivers installed).  The problem is that I notice no difference in the slugish performance I see in Premiere and in rendering.  I've attached a screen capture that shows almost no GPU usage and the render of a 24 second clip taking nearly an hour with long gaps between frames in which the computer seems to do nothing.  And why is CPU utilization so hight?  There are a few effects applied to the footage, mainly MagicBullet Denoiser II as well as some Grad filters. Raw footage is BMCC DNG image sequence.  I understand that Denoiser II is slow, but I'm not understanding why it is so slow on such an expensive system and why the graphics card spends most of its time doing nothing. I've even closed all other applications and told the nvidia control panel to set power use to prefer maximum performance.  Any advice would be appreciated.

render1.jpg

Views

16.9K

Translate

Translate

Report

Report
Community guidelines
Be kind and respectful, give credit to the original source of content, and search for duplicates before posting. Learn more
community guidelines
Valorous Hero ,
Feb 03, 2016 Feb 03, 2016

Copy link to clipboard

Copied

Hey, I just noticed from your Task Manager images that you are only running your CPU's at half power.  You do not have hyperthreading enabled--why?

i saw the core count but figured it was so old it may not have hyper threading hah. i looked it up, turns out they do have HT. HT doesn't actually increase the raw power of cpu, it just doubles the thread pipelines. some programs work better with extra threads, and get a nice boost, while some don't multi-thread well and do better with HT off. it would still be worth turning on to test.

i also looked up the memory specs on the cpu's while on intel's site

i5-2500K has dual channel, 21 GB/s max bandwidth

i7-4700HQ has dual channel, 25.6 GB/s max bandwidth

E5649 has triple channel, 32 GB/s max bandwidth

Votes

Translate

Translate

Report

Report
Community guidelines
Be kind and respectful, give credit to the original source of content, and search for duplicates before posting. Learn more
community guidelines
LEGEND ,
Feb 03, 2016 Feb 03, 2016

Copy link to clipboard

Copied

I also had to look it up before I started typing away!

Premiere Pro is one of those programs that is well hyperthreaded in most places, so it will really payoff here.

Bill

Votes

Translate

Translate

Report

Report
Community guidelines
Be kind and respectful, give credit to the original source of content, and search for duplicates before posting. Learn more
community guidelines
Explorer ,
Feb 03, 2016 Feb 03, 2016

Copy link to clipboard

Copied

Too old...ha.  I think HT has been around since Pentium 4.  So Hyperthreading was off in the BIOS, I turned it on.  It hasn't helped with playback in my limited testing.  In regards to the benchmark, I did the test (before turning on HT) and the second two encoding took less than a couple minutes each.  I'll also say the the GPU was all over the place, sometimes above 50% with the CPU pretty low.

However, when I went to run the script it came up with errors.  The files are in the same directory as the script, but it was telling me no such file name on a certian line.  When I looked at the file it was looking for in the script, it was looking for two different mp2 files, one was called MPEG2-DVD test.m2v and the other MPEG2-DVD test_1.m2v, which was never created.  I removed the _1 and the script ran...is this acceptable to do?  Is this a mistake in the script? I will probably run the test again.  Also, the instructions for the last video say to turn on Mercury in the project settings but it was already on...should it be off or on for the first two encodes?

Playback with the above linked DNG files play back at 20%CPU and 20%GPU (with HT).  I went back and tried a newer DNG file with HT on, and the CPU maxes out at 80% but it doesn't play smoothly at all.  GPU at 20%.  I'll also say that the GPU jumps up past 50% when playing back GH4 footage, for whatever its worth.

Votes

Translate

Translate

Report

Report
Community guidelines
Be kind and respectful, give credit to the original source of content, and search for duplicates before posting. Learn more
community guidelines
Valorous Hero ,
Feb 03, 2016 Feb 03, 2016

Copy link to clipboard

Copied

the gh4 4k cpu usage does seem high. but as i pointed out earlier, its heavily compressed h264 which cpu's are not designed to encode/decode. adobe needs to adopt gpu encoding/decoding, otherwise with intel's slowing speed growth, 8k h265 editing will be nearly impossible.

the new cinemadng format is slightly more compressed than the original format, but your 4x cpu usage increase seems way out of line. so i think its back to a problem with premiere and the media. there are some threads with problems regarding the blackmagic 4k camera's, but i couldn't find any for the 2.5k media. i didn't find any solution either, just conclusions that premiere doesn't support the new format.

Votes

Translate

Translate

Report

Report
Community guidelines
Be kind and respectful, give credit to the original source of content, and search for duplicates before posting. Learn more
community guidelines
LEGEND ,
Feb 03, 2016 Feb 03, 2016

Copy link to clipboard

Copied

adobe needs to adopt gpu encoding/decoding

I rather hope they don't.  Or at least offer the option to turn it off.  So far, GPU encoders seem to be inferior to software encoders.

Votes

Translate

Translate

Report

Report
Community guidelines
Be kind and respectful, give credit to the original source of content, and search for duplicates before posting. Learn more
community guidelines
Valorous Hero ,
Feb 03, 2016 Feb 03, 2016

Copy link to clipboard

Copied

some are worse, some are just as good. for people working with captured h264 or exporting to h264, there is already a quality hit to begin with. it should be easy enough for them to have an option in export to choose between software or hardware encoding.

Votes

Translate

Translate

Report

Report
Community guidelines
Be kind and respectful, give credit to the original source of content, and search for duplicates before posting. Learn more
community guidelines
LEGEND ,
Feb 03, 2016 Feb 03, 2016

Copy link to clipboard

Copied

We export the MPEG2-DVD timeline twice  The first time is with GPU acceleration and should produce the MPEG2-DVD.m2v file (during this test you should see almost 100% GPU usage), the second time you export that timeline is with the CPU only (you turn off GPU acceleration) and produces the the MPEG2-DVD_1.m2v file which will be many times longer than the first.  For the the fourth test you will have to turn the GPU acceleration back on.for the H.264 timeline.  If you copy and paste or rename any files it changes the time stamps.  You can cut and paste without changing the time stamps.  If you removed the "_1" from the script that is defeating the purpose.  There is no mistake in the script, it was been run hundreds of times.  The most common problem is the four required files are not located in the proper location.

Votes

Translate

Translate

Report

Report
Community guidelines
Be kind and respectful, give credit to the original source of content, and search for duplicates before posting. Learn more
community guidelines
Explorer ,
Feb 04, 2016 Feb 04, 2016

Copy link to clipboard

Copied

Bill et al, thanks a lot.  My apologies.  In my haste I must have missed the fact that I was supposed to render the m2v twice.  I probably won't get to running this again today, but hopefully tomorrow. 

My original claim still stands, though as of this point.  My CPU and GPU usage seem to be able to raise and lower in sync with each other in Premiere.  Unless there is broken RAM or a disc issue that is specific to Premiere, I don't see any reason why every way I can think to playback footage of all types is considerably faster with lower resource usage than Premiere cc2015.  I know the workstation is old and I would buy that argument if it were completely unable to playback these files in other software.  However, as it stands, other ways of playing video files are flawless.  Premiere?  Not so much.

Thanks again, I'll get back to you with the benchmark results just as soon as I get around to running this (after all, I'm still on project deadlines!)

Votes

Translate

Translate

Report

Report
Community guidelines
Be kind and respectful, give credit to the original source of content, and search for duplicates before posting. Learn more
community guidelines
LEGEND ,
Feb 06, 2016 Feb 06, 2016

Copy link to clipboard

Copied

LATEST

I know the workstation is old and I would buy that argument if it were completely unable to playback these files in other software

That's the wrong way to think, I believe, measuring the performance of one program against another.  It's a little bit like thinking that if Bill can run the 100m in 12 seconds, then Ted should be able to as well.

And still I offer to test the media in question.  More data can only help.

Votes

Translate

Translate

Report

Report
Community guidelines
Be kind and respectful, give credit to the original source of content, and search for duplicates before posting. Learn more
community guidelines
Adobe Employee ,
Feb 01, 2016 Feb 01, 2016

Copy link to clipboard

Copied

Moving to Hardware forum.

Kevin

Votes

Translate

Translate

Report

Report
Community guidelines
Be kind and respectful, give credit to the original source of content, and search for duplicates before posting. Learn more
community guidelines
Explorer ,
Feb 02, 2016 Feb 02, 2016

Copy link to clipboard

Copied

Hello.  Thanks for the feedback.  I've done some tests but I decided to start a new thread as this no longer seems related to just the render performance as related to the GPU.

Votes

Translate

Translate

Report

Report
Community guidelines
Be kind and respectful, give credit to the original source of content, and search for duplicates before posting. Learn more
community guidelines
Guru ,
Feb 02, 2016 Feb 02, 2016

Copy link to clipboard

Copied

There are several problems involved with this config. Number 1 those CPU's are 4 generations back at this point and the clock speed is to low for good realtime playback and GPU acceleration. You want the CPU clock speed to be atleast 2.6 GHz or higher. GPU acceleration performance is impacted greatly by clock speed. The lower the clock speed the less data gets to the GPU at any point in time. More threads helps but does not offset the low clock speed. Couple this the ram bandwidth of that platform compared to the current X99 platform ie i7 Gen 4. is significantly lower ie a fraction of the performance with far lower clock speed. This has a huge impact on GPU acceleration with frames above 1080 resolution. The 10TB raid you list does not state the amount of drives. CinemaDNG media takes a huge amount of disk bandwidth even at 1080 res. At 2.5K res the disks need to be able to handle around 500 to 600MB/s or more. At 4K you are talking upward of 1GB/s a sec. That likely is a huge impact to what you are running now. You can try and move some of that media to the SSD cache drive and see how it plays in a Premiere timeline by itself. However I would plan on getting 1TB SSD's for media if you work with Cinema DNG often. Adobe has far greater caching at the moment with CC 2015 than previous version and other applications. Davinci also caches far more to ram when playing back. Premiere has GPU accelerated debayering which should help with the Cinema DNG media but you have to have the clock speed on the CPU's to really be beneficial. All the way around the CPU's and ram are a major bottleneck for that platform compared to the current X99 platform. That system with a 5960 8 Core and 64GB of DDR4 would vastly outperform the current system you have with 2.5K or 4K media especially CinemaDNG. Also keep in mind the camera playback is far different than media playback in an editor just like playback in QuickTime player or VLC is far less load on the system and less disk requirement than an editor.

Eric

ADK

Votes

Translate

Translate

Report

Report
Community guidelines
Be kind and respectful, give credit to the original source of content, and search for duplicates before posting. Learn more
community guidelines
Explorer ,
Feb 02, 2016 Feb 02, 2016

Copy link to clipboard

Copied

My point is not that the system is not old, my point is that it used to work just fine in CC, and CC2014 for editing this footage. I could playback BMCC 2.5k CinemaDNG just fine, in fact I could even scrub through it and play it with basic grading on it.  As I said, I've edited feature length pieces (Portrait of St Louis at 250 Years) and countless smaller projects this way. Additionally, see my other post detailing the tests I've run for simple playback.  This system, for example, can play GH4 4k footage with nearly no load on the processor or graphics card.  Inside of Premiere, the load is significant.  The same goes for all other formats I've tested.  IMHO, there is no reason that it should be 2,3, or 6 times more taxing on the processor to run playback from the source monitor in premiere as it is to run playback via VLC or Windows Media Player.  I'll gladly hear arguments to the contrary, but if "Premiere should be significantly slower to play video than freeware" is a realistic argument to Adobe then I need to know so I can make decisions moving forward.  I work for a NFP institution and we can't afford to replace expensive editing workstations every other year because software actually got slower.

Votes

Translate

Translate

Report

Report
Community guidelines
Be kind and respectful, give credit to the original source of content, and search for duplicates before posting. Learn more
community guidelines
Valorous Hero ,
Feb 02, 2016 Feb 02, 2016

Copy link to clipboard

Copied

i saw your other thread, here it is for anyone reading this thread. Premiere Performance - Some questions and tests.

i would like to point out one real possibility between playback in media players such as vlc and premiere... your quadro m5000 should support h.254 and h.265 media decoding, and the media player programs may be using that to offload the cpu workload. even though this tech has been around for a long time (for h264), premiere does not support gpu accelerated decoding and encoding of h264 and h265 media. these formats are very compressed and cpu's are not designed to handle them, so premiere will see heavy cpu usage with these codecs.

looking at your testing with dnxhd it shows premiere not too far off from vlc cpu usage, and that is what i would expect as there is no gpu acceleration for dnxhd. if its dnxhd 444 10bit maybe the cpu usage is about right for your machine, but if its simple 422 8bit dnxhd i would expect it to be closer to 10% cpu on your machine. if that media was 422 8bit dnxhd, you might wanna run cinebench benchmark and see how well your cpu's are performing.

i still don't have any solutions for the cinemadng problem. there are several settings in resolve that affect the cpu usage with cinemadng playback, i think 4 or more various settings, which might explain the lower cpu usage in resolve. i don't recall any performance/quality settings in premiere for cinemadng. you said it worked before on an oler version of premiere, and before the hardware upgrades, have you tried installing an older version of premiere to test?

Votes

Translate

Translate

Report

Report
Community guidelines
Be kind and respectful, give credit to the original source of content, and search for duplicates before posting. Learn more
community guidelines
LEGEND ,
Feb 02, 2016 Feb 02, 2016

Copy link to clipboard

Copied

I downloaded one of the clips referenced above.  I can easily play it back (without any added effects) on this laptop with full resolution without any frame loss on this laptop (i7-4700HQ, 24GB of RAM, GTX 765M, 2 SATA III SSD's, with Premiere 9.1, then I noticed that it was still 9.1 and upgraded to Premiere 9.2 and it dropped 2-6 frames per playback.

Votes

Translate

Translate

Report

Report
Community guidelines
Be kind and respectful, give credit to the original source of content, and search for duplicates before posting. Learn more
community guidelines
Valorous Hero ,
Feb 02, 2016 Feb 02, 2016

Copy link to clipboard

Copied

there is another variable, those files are from when the bmcc camera's first came out. since then, there is a new, slightly more compressed, cinemadng format that the camera's can produce from an update. im not sure if there is a performance or compatibility difference with premiere and the new format or not... if the OP, site fail, wants to share a clip or two, it might be worth testing. at the same time, site fail could download and test one of those old clips to see if there is a performance difference.

Votes

Translate

Translate

Report

Report
Community guidelines
Be kind and respectful, give credit to the original source of content, and search for duplicates before posting. Learn more
community guidelines
Guru ,
Feb 03, 2016 Feb 03, 2016

Copy link to clipboard

Copied

CC2015 greatly increased the caching Adobe uses ie ram for GPU acceleration. This can improve performance with much greater ram bandwidth. GH4 media is no where near as disk intensive or GPU acceleration intensive as CinemaDNG. Its like comparing apples to tomatoes. Of Course the GH4 media will play much easier. It does not have any debayering components like Cinema DNG and it's a fraction of the disk bandwidth ie data streaming from the drives. Memory management itself is also handled by the CPU so the CPU load will increase as the memory management changes. I have tested Raw, CinemaDNG, and R3D extensively and know exactly what it takes to run it well. Your system will not handle it well with the current config. CC2014 was when GPU accelerated debayering was added but not to the extent 2015 uses it. It takes a I7 Gen 3 system with 32GB of ram ie an X79 to handle it decently and X99 to handle it easily. If you want better performance without replacing hardware then go back to CC 2014 or CC when the GPU accelerated debayering was not available and the memory management was far less on the system. 1 last thing I forgot to mention. CC2015 now uses the GPU acceleration player for the source monitor as well as the Timeline monitor. In the past the source monitor just functioned like a standard player without GPU acceleration.

Eric

ADK

Votes

Translate

Translate

Report

Report
Community guidelines
Be kind and respectful, give credit to the original source of content, and search for duplicates before posting. Learn more
community guidelines