Puget Systems print logo

https://www.pugetsystems.com

Read this article at https://www.pugetsystems.com/guides/1723
Article Thumbnail

Premiere Pro (BETA) - H.264 NVIDIA Hardware Encoding Performance

Written on April 14, 2020 by Matt Bach
Share:

TL;DR: Premiere Pro (BETA) H.264 NVIDIA Hardware Encoding Performance

In the 14.2.0 Beta for Premiere Pro, Adobe has added support for H264/HEVC hardware encoding on both NVIDIA and AMD GPUs in addition to the existing method which utilizes Intel Quicksync on supported CPUs.

In our tests, this feature resulted in between a 2-4x improvement in export times depending on the source codec with a minimal (if any) decrease in video quality.

Introduction

In the 14.2.0 Beta for Premiere Pro (which anyone can download and try through the Creative Cloud application), Adobe has added support for H264/HEVC hardware encoding on both NVIDIA and AMD GPUs. Prior to this, the hardware accelerated encoding available in Premiere Pro was done using Intel Quicksync on supported CPUs rather than by a video card. Quicksync is still supported if you do not have a compatible GPU, but with a supported NVIDIA or AMD GPU, this should allow you to export to H.264 or HEVC faster than ever before.

Premiere Pro (BETA) - H.264 NVIDIA and AMD Hardware Encoding

Since this feature is still in beta, we aren't going to go through the kind of testing we normally would do. Once this hits release, we will likely take a look at how different GPU models perform, how bitrate affects performance, and a handful of other factors. For now, however, we will look at performance with five test systems to see how the NVIDIA hardware acceleration performs when rendering 4K H.264 150mbps, ProRes 422, and RED media.

As this is a relatively short post, we are not going to take up half of it listing the system specs we used. If you want to know what motherboard, cooler, etc. we used, the system configurations are identical to what we used in our recent Premiere Pro performance: PC Workstation vs Mac Pro (2019) article. The only difference is that we added an Intel Core i5 7600K system with an NVIDIA GeForce GXT 1070 so that we could see how the performance is on a slightly older platform.

For our tests, we used the "Lumetri Color" sequences in our PugetBench for Premiere Pro benchmark and exported them using the "YouTube 2160p 4K Ultra HD" preset. One little thing to note: we did have to change the encoding level to 5.2 as the default 5.1 does not allow you to export at 3840x2160. On the systems that supported Quicksync, we also included results using that version of hardware acceleration, although we had to use the current latest release version of Premiere Pro (14.0.4) since in the current beta, it defaults to using the NVIDIA/AMD GPU over Quicksync if it is available.

Looking for a Premiere Pro Workstation?

Puget Systems offers a range of poweful and reliable systems that are tailor-made for your unique workflow.

Configure a System!

Labs Consultation Service

Our Labs team is available to provide in-depth hardware recommendations based on your workflow.

Find Out More!

Premiere Pro NVIDIA Hardware Encoding Benchmark Results

Overall, the performance with GPU hardware acceleration is very impressive. Starting with the 4K H.264 150mbps source media, we saw anywhere from a 2x to 4x increase in export performance when using the NVIDIA hardware accelerated rendering compared to software encoding. Compared to the existing hardware acceleration method using Quicksync, the NVIDIA NVENC encoding was about 1.3-2.5x faster which makes it very beneficial even if your system supports Quicksync.

Moving on to the 4K ProRes 422 sequence, the performance gains were a little less, ranging from roughly a 2-3x increase in performance. What is interesting is that it was really only the Core i5 7600K system that didn't see as significant of a performance increase compared to the previous test - all of the modern systems saw almost identical improvements in export times. The advantages over Quicksync are also very similar with a 1.2-2.6x increase in performance on the supported systems.

Last up is our 4K RED sequence. This one is interesting since RED media is much harder to process, and it is also already using the GPU to debayer the RAW footage. Because of this, the performance gains are not quite as impressive - only about 1.3-2x faster than software rendering. This is still a bit faster than using Quicksync hardware accelerated rendering, however, so even though we are likely GPU bottlenecked it is still a great feature to take advantage of.

What about video quality?

For those that are not aware, there is an issue with hardware accelerated rendering - the quality is technically not as high as using software only. For some people, the quality difference is so small that it doesn't matter, while for others it is major issue. Which side of the fence you fall on is completely up to you, so we are largely going to avoid the topic in this post.

What we will do is leave you with uploads of the exported videos using each type of encoding. YouTube does compresses anything you upload, however, so if you want to view the source media directly, we also have the original exports available for download that you can import into an NLE and compare more closely.

For those that don't want to examine these videos, we will say that the quality difference appears to be extremely small in our test cases. It is very possible that shots with more patterns and movement may show a larger difference - or that at lower bitrates the differences will be more apparent - but most editors will likely decide that the possible quality loss is acceptable in exchange for significantly faster export times.

Conclusion

Overall, H.264/HEVC hardware encoding using NVIDIA and AMD GPUs in the Premiere Pro beta is very impressive. Depending on your source media and how powerful of a CPU/GPU you have, the performance gains may *only* be a 2x increase in export performance, but in some cases, we saw up to a 4x improvement in export performance. Keep in mind that most of our testing was done with fairly powerful CPUs which meant that our export times were already very respectable even without hardware acceleration. If your system has a significantly more powerful GPU than CPU (which can often be true on gaming computers and many laptops), you are likely to see even larger increases in performance.

Another great part of this feature is that since GPUs are much easier to upgrade than processors (which will usually require at least a new motherboard if not new RAM as well), this gives Premiere Pro users a much easier method to upgrade their system if they wish to improve performance. This may only be for export/encoding at the moment, but we have to imagine that at some point Adobe will also make GPU-accelerated H.264/HEVC decoding available as well.

This is going to be a huge boon for so many Premiere Pro users, and we can't wait for it to be fully launched. If you want to try it out yourself, you can install the beta version of Premiere Pro though the Creative Cloud app in the "Beta apps" category. This does not overwrite your normal install of Premiere Pro, so there is no need to worry about it messing up any of your current projects or settings. If you do end up trying it out, let us know what you think in the comments!

Looking for a Premiere Pro Workstation?

Puget Systems offers a range of poweful and reliable systems that are tailor-made for your unique workflow.

Configure a System!

Labs Consultation Service

Our Labs team is available to provide in-depth hardware recommendations based on your workflow.

Find Out More!
Tags: Premiere Pro, hardware encoding, Hardware Acceleration, Premiere Pro Beta, H.264, HEVC
Joel

hi, why you didn't include a AMD 3900 or 3950X in the test ?

Posted on 2020-04-15 02:04:12

Couple of reasons:

1) We are working from home as much as possible right now, and these are the systems we have available for us to remotely access for this kind of testing.
2) The feature is still in beta, and since things can change before it hits full release, we don't want to spend a ton of time doing testing that may become invalid in a short period of time.

Once this version of Premiere Pro goes full release, we plan on doing expanded testing. We'll probably end up using the best CPU from each platform (Z390, X299, X570, TRX40), then using 3-4 GPUs to see how the CPU and GPU combination changes performance.

Posted on 2020-04-15 02:20:23
R.J. Leong

I can contribute my results from an AMD Ryzen 7 3800X with only 32 GB of DDR4-3200 RAM and a GeForce RTX 2060 SUPER. (A 500 GB Samsung 970 EVO Plus m.2 SSD is used as the media, projects and export drive.)

My average export results are:

4K H.264 150 Mbps: 108 FPS
4K ProRes 422: 111.9 FPS
4K RED: 51.2 FPS

If my humble R7 3800X system does so well with NVENC, then why bother with a mainstream Intel platform these days? You spend a lot of money for that i9-9900K and an RTX 2080 Ti, and it still underperforms my R7 3800X and only an RTX 2060 SUPER in overall exporting. The only area where the i9-9900K/RTX 2080 Ti combo outperforms my R7 3800X/RTX 2060 SUPER combo is in exporting H.264 150 Mbps to H.264 40 Mbps (and that's only because QuickSync is also decoding that video on the Intel mainstream platform). But transcode a production-friendly codec such as ProRes or Cineform, and there is where the 9900K falls short.

And I can analyze the relatively poor results from the i5-7600K system with the GTX 1070 that you had for comparison: Part of the reason for the poor i5 system export performance is that the GTX 1070 itself is a weakling performer for video processing by today's standards. It, and every single Nvidia GPU that came before it and prior to Turing, chokes badly on 4K/59.94 FPS video. (I can confirm that with the poor GPU score averaging well below 20 with all Pascal GPUs, including a GTX 1080 Ti, even on a monster CPU-based PC, and my own testing with a GTX 1060 6 GB card that averaged only a 14.3 GPU score on my i7-7700 reserve PC that did not improve one tiny bit when retested in my R7 3800X main system.)

Hope this teaches a lesson, when the finalized release of 14.2 gets released.

Posted on 2020-04-22 18:02:19
Hwgeek

Another reason for using Intel CPU's got erased, no need for quicksync anymore, so 9900K/10900K won't be better choise over AMD's CPU's.
moreover, I wonder if anyone has tested this on X570 with 5700XT using PCIe Gen4 vs Gen3 in heavy projects.
looking on those results - it seams that 9900K is bottlenecking the RTX 2080Ti, IMO 3950X/3970X will show even better results with 2080Ti.
it seems that "5Ghz" trick isn't good enough anymore and with Zen3.0 in few month(confirmed by AMD yesterday during earnings call) the IPC gap will increase from current 5% to ~20%.
moreover - AMD's old gen CPU's will continue to drop in price because they are leftovers from Epyc production[same chiplets], so AMD will always have large stocks to sell and prices will keep dropping, this is why we could buy 2700X under $150 1 year after launch and that what will happen to 3950X too :-).
P.S: Don't worry about Intel, I am sure that Intel is cooking great products for next gen, it's will be very interesting to see their next "Core" jump like they did ~15 yrs ago :-).

Posted on 2020-04-29 15:50:42
R.J. Leong

And speaking of Intel, forget about the i5-9400F: In addition to lacking QuickSync, the performance results with that CPU are disappointing (judging by the lone system result in the PugetBench for Premiere Pro list): Even with a GeForce RTX 2080 Ti and 32 GB of DDR4-3200 RAM, the Standard preset score is much more like a 3-year-old locked-multiplier 7th-Generation Kaby Lake i7 4-core/8-thread CPU than any modern 6-core or better CPU (despite the 9400F itself having 6 cores but only 6 threads), even though it was running the latest beta of Premiere Pro with NVENC enabled. However, part of the reason for such a low score is that the faster systems all use m.2 PCI-E SSDs for their projects and media drive while middling-performing systems such as that 9400F-based system still use SATA SSDs for the projects and media. The use of SATA instead of m.2 PCI-e really impacted the overall Export score, especially when exporting to 4k ProRes 422 (which all of the tests in PugetBench for Premiere Pro include). And further adding insult to injury, two new budget 4-core/8-thread AMD Ryzen 3 3### series CPUs, the Ryzen 3 3100 and the Ryzen 3 3300X, soundly beat the i5-9400F in both overall live playback and overall exports (and that's with the exact same version of Premiere Pro, the exact same disk setup and the exact same RTX 2080 Ti GPU). That makes the i5-9400F too expensive for what it is at this present time.

So, if you're still going to stick with a SATA SSD for your system's projects and media drive, then don't waste your money on a top-of-the-line consumer GPU regardless of which CPU you have. The slower interface will erase most if not all of the advantage of such a monster GPU. And judging by the export score alone, the RTX 2060 (non-SUPER) - the bottom of the RTX lineup - would be the least ridiculous choice for a newer Turing GPU when used in such a system. As a matter of fact, I'd be inclined to even mention that a non-RTX Turing GPU such as a GTX 1660 would have been a better match for that i5-9400F when equipped with a SATA media drive instead of an m.2 PCI-E media drive. The same warning goes against buying an overly powerful CPU if you're configuring a system with only slow disks for the media drive. In other words, if you have only a single spinning hard disk, then don't bother with a CPU that has more than 2 cores and 4 threads - and while at that, you might as well not install or run Premiere Pro at all on such a feeble system.

Posted on 2020-05-05 15:08:30
Kunal

Hi Hwgeek,

Requesting help if you can.

I am planning on getting a new editing PC.
Below are my specs confused on graphics card.
Will be using premiere pro 2020.
AMD 3900x
Gigabyte sorus x570 pro
32gb XMP memory
2 tb of nvme

Which graphics card should I go for.
My requirements.
Long videos of 8hrs or 6hrs.
Bulk video production
No special effects or colour grading.
Worst case some transitions.

Request you to please suggest a card.

Posted on 2020-05-14 07:35:44
Hwgeek

IMO start from RTX 2060 SUPER and check the articles here to see if there is a benefit for you going above this card.

Posted on 2020-05-14 14:55:18
Kunal

Thanks for the concise reply.

Posted on 2020-05-15 02:11:57
uri bak

Does this option exist in After Effects as well? Via Media Encoder, thank you.

Posted on 2020-04-15 06:51:49

Yes, it looks like it is in the current beta of Media Encoder as well. That said, you probably won't see nearly as much of a performance improvement with After Effects comps. Usually, most of the export time is going to be spent rendering each frame - not actually encoding it to H.264. So while the encoding step itself should be much faster, since that is a much smaller portion of the overall export time you likely won't see as much of a benefit. Exactly how much faster it will be will depend very heavily on your exact composition and how hard it is to process each frame.

Posted on 2020-04-15 16:13:04
Tristan Summers

I always recommend rendering to Pro Res first then compressing with Media Encoder. It is usually faster and the compressed results will be smaller and better quality. Also if you have a large cache, you can preview at 100% (not auto scale) and the full res render will use the cache. Faster processors and buckets of RAM is really the only way to accelerate After Effects, though I am thinking that a Premier timeline then rendering individual sections from AE as required might actually export even faster. Never ever use dynamic link or send direct from AE to AME. it just isn't ever worth it. And highly recommend trying Rendergarden...

Posted on 2020-05-04 13:43:06
lordtux

Nice, if possible can you please compare with AMD GPUs as well?

Posted on 2020-04-15 22:25:16

Yea, we will include a couple in the testing we have planned for when the feature is in the release version of Premiere Pro.

Posted on 2020-04-15 23:44:16
lordtux

Just checking the article again, would be interesting, if possible, compare PCI-4 against PCI-3, but maybe the GPU would need be much more powerfull to saturate the pci lane. And if possible to you check how much bw is being used from SSD and if we can extract more perf with better IO as well.

Posted on 2020-05-08 16:29:56
Turing

https://www.nvidia.com/en-u...

Nvidia released Studio 442.92 driver.

Posted on 2020-04-16 16:09:15
Gleb Volkov

So this will not accelerate preview rendering of effects in the timeline in any way?
I'm asking because you can kind of think of it like an "export", since it does create the preview files. So...any chance?

Posted on 2020-04-16 20:59:23

That is a really good question that I have no idea what the answer is. As far as I know, there is no way to enable hardware acceleration when generating previews so my guess would be "no", but that is just a guess. You might want to ask Adobe directly on the beta forums. This is the post regarding the GPU accelerated encoding feature: https://community.adobe.com...

Posted on 2020-04-23 22:43:18
Gleb Volkov

I guess it's possible to test just by comparing same timeline render on this beta version vs. previous build?
I don't have the beta version, but perhaps if someone decided to test they can post in the comments.

Posted on 2020-04-23 22:55:44

That would probably be the way to do it. I unfortunately can't test that right now (I'm actually on leave right now and just taking an hour or so to catch up on emails/comments), but the beta is available to anyone if you want to download it. It is in the Creative Cloud app under the new "Beta apps" category. It install along side the normal version and has completely different preferences and other files, so it won't mess with your normal install of Premiere Pro at all.

Posted on 2020-04-23 23:00:41
Gleb Volkov

Yeah, I figured, did not mean you specifically at all :)
Thanks for the info!

Posted on 2020-04-23 23:03:53
Turing

https://twitter.com/Blackma...

https://forum.blackmagicdes...

Support for decoding H.265 4:4:4 12-bit on supported NVIDIA GPUs.
Support for encoding H.265 4:4:4 12-bit on supported NVIDIA GPUs.

Posted on 2020-04-21 07:22:49

Hello, I want to share my comparisons and thank this article, even if it is a beta state, it is helping me in my youtube videos.

9:53 min project on h264 template youtube 4k and 1080p

Hardware Desktop:
Ryzen 5 3600 OC@4.3Ghz
Aorus B450 pro WIFI
16GB ddr4 3200mhz
SSD 1TB M.2 Nvme + SSD 2TB Sata
Asrock challenger OC RX 5700 XT

Ryzen 5 3600 OC@4.3ghz + RX 5700 XT 4k Acc. Software = 14:47 min
Ryzen 5 3600 OC@4.3ghz + RX 5700 XT 4k Beta = 07:50 min

Ryzen 5 3600 OC@4.3ghz + RX 5700 XT 1080p Acc. Software = 08:31 min
Ryzen 5 3600 OC@4.3ghz + RX 5700 XT 1080p Beta = 06:15 min

Laptop Eluktronics Mech17 G1R - 1TB SSD M.2 + 32GB DDR4 2400mhz

Intel I7 8750h@undervolt + RTX 2070 Max P 4k Acc. Software = 18:16 min
Intel I7 8750h@undervolt + RTX 2070 Max P 4k Intel Graphic = 09:23 min
Intel I7 8750h@undervolt + RTX 2070 Max P 4k Beta Nvenc = 05:50 min

Intel I7 8750h@undervolt + RTX 2070 Max P 1080p Acc. Software = 08:54 min
Intel I7 8750h@undervolt + RTX 2070 Max P 1080p Intel Graphic = 05:34 min
Intel I7 8750h@undervolt + RTX 2070 Max P 1080p Beta Nvenc = 04:19 min

Posted on 2020-04-27 19:01:47
Rithy Goodbye

Why I7 8750h + RTX 2070 Max faster than R5 3600 + RX 5700xt ?

Posted on 2020-05-04 02:02:18

The reason I have no idea I can only imagine that CUDA is more efficient than openCL or some effects do not work with radeon and have to be done with the cpu. But the laptop was faster.

Posted on 2020-05-04 12:03:35

Today I get the ryzen 9 in case you are interested this is the comparison in render.

https://k62.kn3.net/taringa...

Ryzen 5 3600@4.3ghz + RX 5700 XT 4k Beta = 07:50 minutos
Ryzen 9 3900x + RX 5700 XT 4k Beta = 07:32 minutos

Ryzen 5 3600@4.3ghz + RX 5700 XT 1080p Beta = 06:15 minutos
Ryzen 9 3900x + RX 5700 XT 1080p Beta = 06:01 minutos

Posted on 2020-05-04 21:10:59
R.J. Leong

Ouch. Radeon's VCE still isn't there yet despite GPU generational changes. While Nvidia continues to improve its hardware encoding with each GPU generation, AMD's Radeon Technologies unit still struggles with VCE.

Posted on 2020-05-04 22:30:31

I totally agree, as in my case I am a gamer and reviewing is only my hobby for price performance I buy the rx 5700 xt but I am sure that I will buy an rtx 3070 or 3080 when they come out as long as they do not come out with a very crazy price .

Posted on 2020-05-04 23:02:10
Tigre Marino

Can't wait to see the numbers on Ryzen Mobile APUs... they will make for fantastic editing mobile workstations !! :D

Posted on 2020-04-29 20:50:38
R.J. Leong

I have just seen results from a mobile Ryzen APU without a discrete GPU. The PugetBench results fron that Ryzen mobile APU is very, very disappointing. The systems powered by 3000 series (which is actually a Zen+ architecture rather than the more desirable Zen2 architecture) APUs with just the integrated Radeon Vega GPU delivered some of the very worst scores out of all systems in the results database. For example, two tests with a desktop system with a Ryzen 3 3200G and 16 GB of RAM and no discrete GPU whatsoever averaged a paltry 36 on those benchmarks. And the lone system in the database with a mobile 3000-series Ryzen APU that had just integrated graphics, a Ryzen 7 3700U with its Vega 10 graphics, scored a horrible 39 on the test (although it may have been hobbled by it having only 8 GB total of RAM). In fact, the only system in the entire database that scored worse than the AMD APU is an 11-year-old desktop PC with a first-generation Lynnfield i7 and an old Terascale-based Radeon GPU that doesn't have enough RAM to work properly in any newer version of Premiere Pro.

The other AMD APU-based PCs scored better, but they also have sufficiently powerful discrete GPUs.

One mobile Zen2-powered Ryzen mobile PC, with a Ryzen 7 4800H CPU, did score well into the 100s with the integrated Radeon Navi graphics (but well into the 400s and 500s with the discrete GeForce graphics). But that system could switch between the integrated and the discrete GPU for MPE GPU acceleration because its discrete GPU is a Turing-based GeForce GPU. That is accomplished by tweaking a few settings in the laptop's graphics control panel for the discrete GPU. And the scores from just the integrated graphics are no better than the typical GPU-less Intel system.

Thus, what you wished for, sadly, is nowhere to be seen. Especially since Intel CPU-powered systems with only the integrated Intel HD or UHD or Iris Graphics delivered scores of between 100 and 200 - which are still lousy - on those same benchmark tests.

Posted on 2020-05-04 22:41:55
steventimothy

In the current version of Premiere (14.1 I think?), there's a pretty drastic difference in quality between software and hardware encoding for H.264. I'd be very curious to know if hardware encoding is improved using the GPU.

Posted on 2020-04-29 21:11:26
Ben Haerkens

Works, but not with all settings. When I choose exprot 1080P VBR2 I get the message 'you system does not support hardware acceleration for the current settings' I use I9 9900K with GTX1060 6GB and 32 GB RAM. Does anyone have a solution? Greetings Ben

Posted on 2020-05-04 09:10:32
R.J. Leong

There is no solution here. Adobe's implementation of hardware encoding only supports CBR and VBR1 at this time. Hardware encoding for VBR2 requires a massive redesign of the encoder's code at hardware level.

Posted on 2020-05-04 22:36:26
Ben Haerkens

Okay, that's clear, thanks. Hopefully they will make this possible in the final version. Greetings Ben

Van: Disqus [mailto:notifications@disqus...]
Verzonden: dinsdag 5 mei 2020 00:36
Aan: beremaro@home.nl
Onderwerp: Re: Comment on Premiere Pro (BETA) - H.264 NVIDIA Hardware Encoding Performance

<https: referrer.disqus.com="" juggler="" event.gif?category="email&amp;impression=b1407472-8e57-11ea-b985-3a840ab16cf3&amp;experiment=new_notifications&amp;behavior=open&amp;variant=active&amp;type=notification.post.registered&amp;event=email"> "There is no solution here. Adobe's implementation of hardware encoding only supports CBR and VBR1 at this time. Hardware encoding for VBR2 requires a massive redesign of the encoder's code at hardware level."
<https: blog.disqus.com=""/> Disqus

New reply on Puget Systems <https: disq.us="" url?impression="b1407472-8e57-11ea-b985-3a840ab16cf3&amp;thread=7970857085&amp;forum=929533&amp;url=https%3A%2F%2Fwww.pugetsystems.com%2Flabs%2Farticles%2FPremiere-Pro-BETA---H-264-NVIDIA-Hardware-Encoding-Performance-1723%2F%23comment-4901356138%3AHuQRHScI9S2Ode9RVoxiXc3I3QU&amp;variant=active&amp;experiment=new_notifications&amp;behavior=click&amp;post=4901356138&amp;type=notification.post.registered&amp;event=email">

<https: disq.us="" url?impression="b1407472-8e57-11ea-b985-3a840ab16cf3&amp;thread=7970857085&amp;forum=929533&amp;url=https%3A%2F%2Fdisqus.com%2Fby%2Fbenhaerkens%2F%3Futm_source%3Dreply%26utm_medium%3Demail%26utm_content%3Davatar%3Afxv43uHrT5dcj4KeDPrETGfM_M4&amp;variant=active&amp;experiment=new_notifications&amp;behavior=click&amp;post=4900497899&amp;type=notification.post.registered&amp;event=email">
5:10 a.m., Monday May 4

Posted on 2020-05-05 06:26:29
Tristan Summers

Be good to try to develop a stess test for GPU rendering, as I have had some terrible experiences over the years with Premiere / AME hardware encoding.

Posted on 2020-05-04 13:44:46
MusicIsMath

I have a ryzen 2600x and a 1080ti and I have to say, the gains I got with hardware encoding are even more spectacular than what you guys posted! With no effects and in 1080p, the hw encoding speed is almost tenfold that of software encoding, with lumetri on, it's five-sixfold.

Posted on 2020-05-04 16:11:06
R.J. Leong

That's because the 2600X is not a very powerful CPU to begin with. And although the 1080 Ti is relatively weak for video processing compared to all of the GTX 1600 and RTX 2000 series GPUs, the GTX 1070 that was used in the reported i5-7600K is even weaker.

Posted on 2020-05-04 22:34:37
Igor Baryshev

The reason 1070 was slower wasn't because 1070 is slow, but because 7600K is much slower than chips used in the rest of setups.
1070 has a pretty good encoding throughput, main difference between Pascal 1070 and Turing 2080Ti is quality per bits used.
7600K just couldn't saturate the encoder enough to show its speed.

Posted on 2020-05-19 22:12:36
aaaariel

I Have an i7 8700k with GTX 1070 and still get the option grayed out saying "hardware encoding unavailable" , trying to export 1080p h.264 1VBR or CBR, any idea why ?

Posted on 2020-05-05 21:49:38
R.J. Leong

Are you running the 14.2 or 14.3 beta? You see, none of the current official non-beta versions (including the current 14.1 release) have NVENC or VCE encoding support at all, and that you have the integrated Intel UHD Graphics disabled in your system (as it is by default whenever a discrete GPU is installed).

Posted on 2020-05-05 22:37:43
aaaariel

Hey R.J. Thanks for your response.
I checked and the version is 14.0 version... it's the only one available on my CC panel on Beta section. https://uploads.disquscdn.c...
as you can see there is no option to update only open the app.

Posted on 2020-05-06 23:03:02
R.J. Leong

If so, you may have an outdated version of the Creative Cloud desktop app installed. This can happen if you have the "Always keep Creative Cloud up to date" box unchecked in the app settings. That checkbox does not actually do the way you thought it would (that is, auto-update all of the apps, not just the Creative Cloud app) - but it just updates that interface app itself. Updating all apps automatically is a separate setting within the desktop app. You need to update that app before any newer versions of the Creative Cloud apps appear,

Posted on 2020-05-07 15:03:32
maans le roux

How will this improve performance of Ryzen 4000 on laptops? Similar or better performance to QuickSync?

Posted on 2020-05-06 06:36:40

It should help quite a bit - assuming the laptop has a discrete AMD/NVIDIA GPU. I'm not sure how well it would work with an integrated GPU, however, or if it would be faster than an Intel CPU with Quicksync.

Posted on 2020-05-06 17:01:49
R.J. Leong

It does not quite work that way. The way Adobe's implementation of the hardware encoder is designed, only one or the other GPU may be used for hardware encoding. Adobe does not currently support hardware H.264 or H.265 decoding from anything other than Intel's QuickSync at this time. And with the current beta versions, if a discrete GPU is enabled on a system that has both discrete and integrated graphics enabled, then only the discrete GPU (or the GPU that has been selected for hardware acceleration in the discrete GPU's control panel) is used for hardware encoding. In other words, you may have either the integrated GPU or the discrete GPU - but not both - to do the hardware encoding.

And the results in the overall benchmarking with a 4000-series Ryzen APU with just the integrated graphics was disappointing. In fact, no better than the typical GPU-less (aka IGP-only) Intel PC.

Posted on 2020-05-07 17:09:16
INTC

https://downloadcenter.inte...

"This driver is WDDM 2.7 compliant and ready for the Windows 10 May 2020 Update. It introduces support for Dolby Vision and the new DirectX 12 Shader Model 6.5 compiler on 7th Generation Intel Core processors or higher (Intel HD Graphics 610 or higher)."

Intel released 27.20.100.8187 graphics driver.

Posted on 2020-05-06 10:16:46
R.J. Leong

Following up on this, I have now started to see one user testing several low-end PCs with the new 14.3 beta of Premiere Pro, using various current-generation and newly-released low-end CPUs, 32 GB of DDR4-3200 RAM and a GeForce RTX 2080 Ti and what appears to be a typical SATA SSD storage setup. NVENC is enabled, but none of the Intel CPUs used in the comparison that I am to present have QuickSync enabled or present. For comparison, a 6-core/12-thread first-generation Ryzen 5 1600 and a three-year-old 4-core/8-thread Kaby Lake i7-7700K were included in the testing.

The results are quite surprising: Both of the low-end 9th-Generation (current) Intel CPUs - the i3-9100F, and to a lesser extent the i5-9400F - ended up bottlenecking the RTX 2080 Ti substantially. In fact, the 9100F/RTX 2080 Ti combo perfectly illustrates the opposite extreme of a severely imbalanced PC setup: A GPU that's severely overqualified for the CPU. While the RTX 2080 Ti scored between 66 and 68 in the GPU score in the other cheap PCs in this roundup, the RTX 2080 Ti in the i3-9100F scored well below the others in GPU score - more like what a higher-end PC would have scored using merely an RTX 2060 SUPER. That means that effectively, a cheap PC with such an overqualified GPU would perform no better in anything than that same PC would with a much cheaper, lesser-performing GPU such as a plain, non-SUPER RTX 2060; in other words, the PC owner with such a lopsided GPU-heavy configuration would just be throwing money away for practically zero performance improvement. The opposite example of an imbalanced PC configuration would be a GPU that's so weak compared to the CPU that it would choke the life out of everything that ran on such a misconfigured PC (and yes, that would have been ME flashing a choke sign at the sight of such a severely bottlenecked system).

Three other CPUs - the old i7-7700K, the current Ryzen 5 3400G APU and the new Ryzen 3 3100 - are grouped together in a tier that's distinctly above the i5-9400F (meaning that the overall Standard preset scores in PugetBench for Premiere Pro 0.9 for all three are within 10 points of one another, centered around the 500-point mark). Above that tier is the old Ryzen 5 1600. Finally, the new Ryzen 3 3300X, which would almost pass for everyday video editing chores; in fact, that 3300X, despite having only 4 cores and 8 threads, actually outperformed some older 6-core/12-thread CPUs in this test. And that's the biggest surprise in this round-up; however, it boiled down to the differences in the disk I/O throughput between the older and the newer AMD chipsets used in the systems in the roundup: The 1600 was running on the older X470 chipset while the 3000-series CPUs were running on the newer X570 chipset.

However, when another user tested three of the low-end AMD CPUs in the older version of Premiere Pro, using software-only encoding and with only 16 GB of RAM and an RTX 2080 SUPER, the scores predictably dropped, to the point where the 2.5-year-old 6-core/12-thread CPU regained the edge over the newest 4-core/8-thread CPU. The difference in the relative ranking came down to the CPU processing power, where the older 6-core/12-thread CPU still ruled in this test. Here, the newer X570 chipset was used with all three of the CPUs (although the X570 chipset does not officially support any of the first-generation Zen-based Ryzen CPUs without integrated graphics).

One should note that these tests were designed to determine the feasibility of an older or lower-end PC for video editing duties, using a relatively modern hardware-intensive NLE software, in a pinch. Of course, they would be no match for any modern, higher-end PC for this purpose.

One more note:
These results are exactly why I ended up choosing an AMD Ryzen 7 3800X for my $1000 platform/GPU upgrade build (at the time, the 3800X was priced only $30 higher than that of the 3700X). Had the 3800X been at regular price, then I would have gone with the 3700X instead. With the AMD CPU-based upgrade, I was able to squeeze in 32 GB of DDR4-3200 RAM and a GeForce RTX 2060 SUPER into that low of a budget. Had I gone with Intel instead, I would have been trapped between a rock and a hard place, as none of the Intel CPU options within my CPU budget offered enough of a performance improvement over whichever older 4-core CPU that I was using at the time to justify spending that much money on, while the one Intel CPU option that did would have forced me to cut corners on everything else in that platform/GPU upgrade package.

Posted on 2020-05-06 15:14:52
SmokeThatSkinWagon

Did you guys test exporting in 265/HEVC? Or only h264/AVC?

Wondering if the jump in real world performance is equivalent for both. Thanks!

Posted on 2020-05-15 18:09:19

We haven't yet, but if anything I imagine it would be even more beneficial since HEVC is even harder to process than H.264. Once this feature exists in a non-beta version of Premiere Pro, we plan on doing a more in-depth analysis including HEVC as well as multiple AMD/NVIDIA GPU models.

Posted on 2020-05-15 18:35:50
SmokeThatSkinWagon

Looking forward to it. Thanks for all the fantastic work! ☕

Posted on 2020-05-15 18:38:34
theshakl

I'm interested in this too. VERY rare to hear about h265 exports

Posted on 2020-05-22 23:36:51
Ampere

https://devblogs.nvidia.com...

https://devblogs.nvidia.com...

Will Puget be benchmarking the upcoming Nvidia Ampere Geforce consumer GPUs?

Posted on 2020-05-18 09:22:46

Absolutely! We pretty much always look at anything mod-range and above that is released from Intel/AMD/NVIDIA! Sometimes we can get pre-launch samples that let's us have the content ready to go at launch, while other times we have to purchase retail after launch which delays our testing a bit

Posted on 2020-05-18 16:21:43
Radek .

What about hardware decoding on nvidia NVDEC and AMD UVD? For now, premiere pro supports decoding only on intel QSV( quick sync video)

Posted on 2020-05-18 10:50:10

I don't know of any official word from Adobe regarding when decode support will be added. I would have to imagine it is on their road map, however. Probably once they are satisfied that encoding is completely stable they will start working on that.

Posted on 2020-05-18 16:23:22
Dodi Usman

Hi, it would be interesting to see the figures of a SLI or Crossfire setup. Are there any plans to test with a multiple GPU setup?

Posted on 2020-05-19 03:34:48

Multi-GPU testing is something we have had on our list to re-test. Last time we looked at it several years ago, it wasn't all that great and only helped when using low-end GPUs. Things have changed quite a bit since then, however, so it is probably time for a revisit. I was going to roll our more in-depth testing of the GPU hardware encoding into a complete GPU round-up testing, so that would be a good time to toss that in.

Not 100% sure when that will be, however. We are still in pretty tight lockdown for COVID-19 up here in Washington state and doing GPU testing requires us to be on site to swap the GPUs constantly.

Posted on 2020-05-19 17:34:49
Dodi Usman

Thanks, will be looking forward to the next set of test results. If multiple GPUs help improve the time by a further 2-4x, the extra $$$ will definitely be worth it.

Posted on 2020-05-19 20:00:51
bobd

I'm still seeing the bit-rate being capped pretty drastically, similar to how it was via the older, Quick Sync hardware export. It seems the limits are (roughly) 1080p--10 Mbps, 1440p--40 Mbps, and 4K--80 Mbps. Is this how it is supposed to work?

Posted on 2020-05-19 21:05:21

Interesting, I'll have to look into that. We only tested 40mbps UHD in this post, but I wouldn't be surprised if there were hidden bitrate limits. You might want to post on the Adobe forum or tweet at @AdobePremiere (they seem to be extra active today)

Posted on 2020-05-19 21:29:05
Harley.D

hi, could you test video quanlity at low bitrate? for example, fhd h264 at 4 mbps.
and different different graphic cards?

Posted on 2020-05-20 00:59:38
Vaffshammer

Looks like the updated NVENC in the RTX cards is way better than QuickSync, older NVIDIA-cards and everything from AMD.
Have a look at this quite comprehensive analysis on different encoders and bitrates:
https://www.youtube.com/wat...

Posted on 2020-05-20 16:42:05
theshakl

how come no one is exporting in h265 ??? ha

Posted on 2020-05-22 23:35:28
WIN FILMS

Hey Matt,

With an i9 9900k and rtx 2070

I had the old quicksync enabled with the integrated gpu, to best utilize NVENC, should I change back the intergrated gpu off in the bios?

Posted on 2020-05-23 18:49:42

No, I wouldn't for now. If you have "hardware acceleration" enabled in your render settings, it will attempt to use NVENC first, then fall back to Quicksync if you don't have a supported GPU. However, H.264/HEVC decoding is not GPU accelerated yet, so keeping Quicksync enabled should still help with live playback performance.

Once both encoding and decoding is GPU accelerated, you might as well disable the iGPU since it is less things for Windows Update to decide to play games with, but until then I would keep it to make things as fast as possible in Premiere Pro.

Posted on 2020-05-25 16:52:50
Dwither

This is a great article . I have a said question for anyone who could help . The 2080ti seems to preform great h however thats out of my budget. Since this is new maybe no one will have an answer . What is the point of diminishing returns for the lower end RTX cards. Im sure the 2080ti is the best but what about the 2060 vs 2070 - thanks for any and all help

Posted on 2020-05-28 06:19:22
Martin Richard

I also would like to know the difference between the nvidia stack on same cpu platfform.

Posted on 2020-05-28 19:12:00

We haven't tested the new GPU accelerated encoding with a range of GPUs quite yet (although we are planning on doing so in the next few weeks), but in general it depends heavily on the kind and number of GPU-acclerated effects you are using. This is our latest GPU testing we have done in Premiere Pro: https://www.pugetsystems.co...

Basically, if you only use a could effects like Lumetri Color and maybe some transitions, there isn't much of a difference beyond even a RTX 2060. If you have a ton of accelerated effects, however, you can see some pretty nice performance gains all the way up to a 2080 Ti.

To give you a general idea, however, in our Premiere Pro workstations the most common GPU used is the RTX 2070 Super, followed closely by the RTX 2080 Ti. I bet the RTX 2070 Super customers are those looking to maximize their budget, while those who purchased a 2080 Ti are those who can justify almost any cost if it improves performance even by just a little bit.

Posted on 2020-05-28 19:13:35
Dwither

Thank you

Posted on 2020-05-28 19:18:17
besweeet

One thing I noticed is that videos encoded via NVENC have slightly different colors (brighter in most cases). Unusable this way.

Posted on 2020-05-28 16:40:26