Puget Systems print logo
Read this article at https://www.pugetsystems.com/guides/502
Article Thumbnail

Adobe Premiere Pro CC Professional GPU Acceleration

Written on September 27, 2013 by Matt Bach


Prior to the use of GPU acceleration, the time it took to render movies in software like Adobe Premiere was almost completely based on the speed of the CPU. But as more effects and transitions in Premiere take advantage of GPU acceleration, you now have to take into account the speed of the video card in addition to the speed of the CPU when either configuring a new system or upgrading an old one. 

When Premiere started using GPU acceleration, the process was exclusively done via CUDA which is limited to NVIDIA-based video cards. With Premiere Pro CS6 version 6.0.2 and Premiere Pro CC, Adobe introduced GPU acceleration via OpenCL in addition to CUDA which allows AMD cards to be used. This essentially doubles the number of video card that can be used and is reflected in the Adobe's list of supported cards. Note that even if your GPU is not on this list, one of the major improvements in Premiere Pro CC is the fact that you can now enable GPU acceleration without making any special modifications to Premiere. You simply receive a popup when you enable GPU acceleration saying that your card is not officially supported.

We have looked at GPU acceleration performance in Premiere Pro before in our CS6 GPU Acceleration article, but that article was focused mostly on desktop (or "gaming") video cards. This time, we are going to look almost exclusively at professional workstation cards. And since professionals can often justify larger budgets that home users, we also decided to include dual GPU configurations in our testing.

One thing we want to be clear on before we get into our testing is that for the majority of Premiere Pro CC users, a desktop card is going to give you better performance at a much lower price. The advantage of using a workstation card is largely in how quickly Adobe officially certifies newer workstation cards and drivers. So if you have an issue with Premiere, you are much more likely to get a fix from Adobe if you are using a NVIDIA Quadro K4000 versus a NVIDIA GTX 780 simply because the Quadro card is on the certified hardware list. In fact, in our testing Premiere would crash when we enabled SLI with dual NVIDIA Geforce GTX Titans. Take it out of SLI (which isn't required for full performance anyway), and it works just fine.

So even though a desktop card may give better performance, if you need to operate with as few issues as possible a workstation card is usually a safer choice. There are some advantages to workstation cards such as double precision performance and ECC RAM on the higher-end cards, but for movie encoding those are usually not a very large factor. 

Test Setup

To see how each GPU performs with different chipsets, we used two separate testing platforms consisting of the following hardware:

To thoroughly test both AMD FirePro and NVIDIA Quadro video cards, we used the cards below in both single and dual GPU configurations. To provide a comparison to more standard desktop cards, we also included the NVIDIA GTX Titan which is currently the fastest single GPU video card available.

Model Est. Street Price Model Est. Street Price
Quadro K2000 2GB $430 FirePro W5000 2GB $420
Quadro K4000 3GB $765 FirePro W7000 4GB $670
Quadro K5000 4GB $1700 FirePro W8000 4GB $1430
GeForce GTX Titan 6GB $1000 FirePro W9000 6GB $3400

The latest drivers for each card were used (320.86 for NVIDIA Quadro, 320.49 for NVIDIA GTX Titan, 12.104.2 for AMD FirePro) along with the latest drivers and BIOS versions for the motherboard and other components. All Windows and software updates were applied before starting our testing. With the exception of the NVIDIA GeForce GTX Titan, all of these cards are currently on Adobe's supported card list. Note that SLI/Crossfire does not need to be enabled for dual GPU configurations. In fact, NVIDIA drivers currently do not allow SLI with Quadro cards except on a few very specific prebuilt systems.

To measure the performance of each card, we encoded two different timelines using Premiere Pro CC with GPU acceleration enabled. The first timeline was taken from the H.264 portion of the PPBM6 benchmark for Premiere Pro CS6. This timeline has multiple 1080p clips with various effects and overlays. Not all of the effects are GPU accelerated, however, so the CPU will play a relatively large role in the overall performance.

Since this timeline was not intended for Premiere Pro CC, we had to make a few minor modifications to attain full compatibility with Premiere Pro CC. First, we had to re-import several clips (AA004701.MXF and AA006401.MXF) due to a mismatch in the number of audio streams. Second, we had to change the encoding file slightly from 29.97fps to 25fps in order for the timeline to properly encode. Due to these changes, our results are not directly comparable to other PPBM6 benchmark results that you might find online.

PPBM6 H.264 Timeline PPBM6 H.264 Encoding Settings
PPBM6 H.264 Timeline PPBM6 H.264 Encoding Settings

For our second timeline, we used parts of the 4k movie Secret World and applied various GPU accelerated effects and transitions as well as rotations and scaling. In order to really push the video cards, we set up six 4k clips on top of each other, each with a different effect applied. We chose to encode to H.264 at the same 4k resolution and 24fps as the original clip. While the CPU is certainly still used, this timeline should place a much heavier emphasis on GPU acceleration and should more clearly show the differences between each card in a worst-case scenario.

One thing we want to point out is that there are currently a few GPU accelerated effects that are not compatible with AMD video cards - Fast Blur, Gaussian Blue, Directional Blur, and the Basic 3D effect. In order to keep these from skewing our results, we left them out of our timeline entirely. More information on this is available from Adobe's OpenCL and Premiere Pro CS6 blog post.

Custom 4K H.264 Timeline 4K H.264 Encoding Settings
Custom 4K H.264 Timeline 4K H.264 Encoding Settings

For both of these timelines, a custom AutoIt script was used to time how long it took to encode each movie.

1080p (PPBM6) Results

1080p PPBM6 H.264 encoding benchmark

With the results sorted according to the X79 results, we get a chart that is strikingly similar to what we saw with desktop video cards in our Premiere Pro CS6 GPU Acceleration article. Just like in that article, we hit a very clear wall where no matter how fast the video card, the rendering time simply didn't go down any further. This is due to the fact that the timeline included some effects that are not GPU accelerated so the video card ends up waiting for the non-accelerated effects to finish before continuing. One interesting tangent is that this "wall" was about 70-80 seconds lower than what we saw in our previous article. This is likely due to a combination of improvements in Premiere Pro CC compared to Premiere Pro CS6 and the fact that we are using newer (and faster) CPUs in this article.

With the Z87 Haswell system, we hit the performance wall right after the lowest-end cards - the NVIDIA Quadro K2000 and the AMD FirePro W5000. Interestingly, the FirePro W5000 did worse when we used two cards than when we used just a single card. We don't know why this is, but we verified the result many, many times with multiple driver and even full OS re-installations.

For the X79 Ivy Bridge-E system, there was a bit more headroom but there are definitely diminishing returns almost immediately. You could use an AMD FirePro W8000 over an AMD FirePro W7000, but you will be spending a lot of extra money for about 1-2% decrease in rendering times.

With these diminishing returns, it is hard to really give a recommendation on which card is best for this type of movie rendering. The NVIDIA Quadro K2000 is one of the best cards for the price as it is almost as fast as the higher end cards but is much more affordable. The nice thing about it is that if you need a bit more performance in the future, you can purchase a second one and get performance equivalent to a NVIDIA Quadro K4000 or an AMD FirePro W9000.

4k (Custom) Results

4k Custom H.264 encoding benchmark

When we upped the clip sizes to 4k resolution and focus solely on GPU accelerated effects and transitions, we get a much clearer picture of exactly how fast each of these cards actually is. Keep in mind that this is somewhat unrealistic at the moment since you would normally not have six 4k clips with only GPU accelerated effects, but this was the best way we found to show the maximum performance difference between these cards.

Before getting into specific results, there are a few overall things we want to point out. First, since this timeline was so heavy on GPU accelerated effects, the difference between the Z87 Haswell and the more powerful CPU in the X79 Ivy Bridge-E is much smaller than in our 1080p testing. The exception is with dual AMD FirePro graphics and dual NVIDIA GTX Titan configurations. For those configurations, the additional PCI-E lanes (which allow both cards to run at full x16 speeds rather than being limited to x8 speeds) on the X79 system allowed it to perform much better. Interestingly, the NVIDIA Quadro cards did not seem to care if they were running at x16 or x8 speeds. The exception to this is the FirePro W5000. For some reason, it performed much better on the Z87 system when two cards were installed. Combine this with the odd results we saw in the 1080p test and it makes us think that there is currently some driver or firmware issue with this card.

If you are working with this large and complex of video files, you likely have a bit more of a budget than the typical user. For that reason, we would recommend the AMD FirePro W7000 as an entry level card for this type of movie rendering. The NVIDIA Quadro K4000 is also a good choice, but the FirePro W7000 is both cheaper and slightly faster.

If you need the fastest performance possible, you can cut render times almost in half compared to the Quadro K4000 or FirePro W7000 by using dual NVIDIA Quadro K5000 or dual AMD FirePro W9000 cards, but you are certainly going to be paying a large amount of money for that additional performance. But if you work for Hollywood and are constantly waiting for large movies to render, the additional $2600 for dual Quadro K5000s or $6000 for dual FirePro W9000s might be completely worth it.


One thing we have very clearly learned in our testing is that choosing a video card for Premiere Pro, the choice is going to heavily depend on the type of files you are working with. There simply is no "best" video card for everyone since you will see diminishing returns or even hit a performance wall depending on the size and complexity of the timelines you are working with. If you are only applying a few effects to a 1080p file with simple transitions, a NVIDIA Quadro K2000 likely has more than enough power. If, however, you work with complex 4k video timelines with lots of effects and transitions, you may benefit from using one of the higher end video cards.

In addition, remember that not all effects support GPU acceleration. Especially in the case of third-party effects, be sure to check to see if they support acceleration before deciding what video card you should use. Even if they do support GPU acceleration, keep in mind they most likely do not support multiple cideo cards. In those cases, you should avoid multiple GPU setups and stick with a single GPU configuration.

Most professionals will likely fall into an area where a NVIDIA Quadro K4000 or AMD FirePro W7000 will give them the best performance without hitting a performance cap. The nice thing is that if in the future you do need more power, our 4k encoding test shows that adding a second GPU scales very well with Premiere Pro CC. If you are purchasing a new system, a single more powerful GPU will almost always give you better performance at a slightly lower cost, but having an upgrade path available is always a good thing.

Tags: Premiere Pro CC, GPU Acceleration
Avatar Dilemma

Great review thanks. Is there any chance you could benchmark After Effects through this setup?

Posted on 2013-09-30 10:27:26

Yea, After Effects is on our list of software to benchmark. We are basing the order on popularity of previous articles and how many requests we have, and right now AutoCAD is next in line with either Maya or After Effects being next. Most like we will do After Effects before Maya since the setup is so similar to Premiere.

Posted on 2013-09-30 19:33:30

Just an update to my last status: I worked a bit with After Effects CC yesterday trying to get every render benchmark I could find working. They all either didn't open in AE CC or for some reason did not utilize GPU acceleration (even though they should). I'm still hoping to get them working, but since I only have basic working knowledge of After Effects, we are going to bump that down in priority behind they other software we want to benchmark. So it may be a bit (hopefully only a month or two) before we have any After Effects GPU acceleration benchmarks.

Posted on 2013-10-01 19:39:29
Avatar Dilemma

AE pretty much CPU only - except for 3d render stuff. This seems to be a render test a few people have used in the past.

Posted on 2013-10-01 21:09:54

Yea, I looked at that one, but I had issues with it starting the render properly (I.E, it wouldn't start at all for some reason). I also looked at ToalBenchmark http://www.media-motion.tv/... and AE CUDA Benchmark http://forums.creativecow.n... . TotalBenchmark wouldn't open in CC since it was made in an older version and the CUDA benchmark wasn't using the GPU at all. I'm sure we could work out those issues, but right now its simply a matter of efficiently dividing our time. We have a bunch of other things going on right now, so unfortunately it has to get pushed back a bit.

Of course, if you find anything other After Effects benchmarks that should be compatible with CC benchmarks, please let me know.

Posted on 2013-10-01 21:16:05
Avatar Wolfgang

Great Article, looked quite a while for something like this. Thanks! The only thing i would love more to see is how the GTX780 (Ti or non) performs compared to the Titan. I think the question is quite obvious if you look at the price of those two ;)

Posted on 2013-11-18 12:35:27
Avatar DashV

I would also love to know the answer to that question.

780Ti or Titan?

Posted on 2014-02-15 08:13:08
Avatar Arcraz522

The 780 Ti will out perform the Titan. It has more CUDA cores and is all around a better card. The only advantage the Titan has is that it has 6Gb of VRAM while the 708 Ti only has 3Gb. The Titan Black, however, has the same amount of CUDA cores as the 780 Ti but also has 6Gb of VRAM. The only reason you need 6Gb of VRAM is if you're doing a hell of a lot of rendering. If you're gaming, save yourself $300 and get the 780 Ti. Even then, unless you have multiple displays, 780 Ti is a bit overkill.

Posted on 2014-04-04 15:52:11

The 780 Ti is indeed the better choice for gaming... most of the time. When you get up to 4K resolution the added RAM on the Titan / Titan Black is beneficial.

However, for the specific use case of video editing (and other CUDA related tasks), I would recommend spending the extra money on the Titan Black. That added memory definitely comes in handy when you are editing higher resolution video files, and for other sorts of CUDA processing it has better double precision performance than the 780 Ti and other more mainstream cards. The price difference between those video cards is also a lot smaller in the grand scheme of a several thousand dollar media editing workstation.

Posted on 2014-04-04 16:10:01
Avatar Serg76

Nvidia states the Titan Black will have single-precision performance of 5.1 TeraFlops and double-precision performance of 1.3 Tera Flops. This compares to 4.5 TeraFlops and 1.3 TeraFlops for the GTX Titan and 5 TeraFlops and 210 GigaFlops for the GTX 780 Ti. Nvidia reduces the double-precision speed of its non-Titan cards in the driver, making them run at 1/8 of what they're capable of.

Posted on 2014-08-16 21:42:28
Avatar SoNic67

People keep comparing number of cores because that matters for gaming.
But this is a whole another world, the double-precision world.
All those extra cores mean nothing if they are not working in the editing applications.
What good does the 780Ti 2880 cores when they yield only 210 GiFlop DP (1/24 of SP)? Quadro 4000 has 243GiFlop double.
Titan on other hand can be switched to 1/3 Double Precision mode - that yields 1500 GiFlop in double.

Posted on 2014-10-16 16:47:26

The performance difference between the GeForce and Quadro cards for *double precision* performance is not actually a matter of hardware being / not being utilized, but instead that GeForce cards have intentionally crippled double-precision performance. They want folks doing that level of work to be using professional-grade cards like the Quadro series or even Tesla. The Titan is sort of a hybrid: it is still in the GeForce line, but has some improved support for that sort of workload as well.

Posted on 2014-10-16 16:53:27
Avatar astrix_au

I find that hard to believe that it's all in the driver, just because people can flash a card with a different BIOS and it then says it has more stream processors isn't it only reading the BIOS and not the actual amount of stream processors or / Cuda cores. People do this with the 290/290x and others say this is also the case with 780 780ti or something like that.
That is my opinion on that matter that there is more hardware available but they just choose to disable it, it makes no sense to me since they are selling cheaper and those people aren't going to spend more just for the extra cuda cores.

Posted on 2014-10-23 07:14:13
Avatar Arcraz522

780 Ti vs Titan


780 Ti vs Titan Black


Posted on 2014-04-04 15:54:43
Avatar perfectinvesting

Investments To Make Money - 250% daily for 20 days
InvestmentsToMakeMoney.com Investment To Make Money Network has been established by United State of America's InvestmentsToMakeMoney.com companies. After gaining experience in Forex trading for many years, these companies decided to form an investment firm besides their trading in order to accept investment funds from members and deposit for Forex trading. InvestmentsToMakeMoney.com Investment Network has been legally registered as a private investment firm.

Posted on 2014-06-05 11:16:54
Avatar Andy

Can Premiere Pro access the RAM off of the second card?? I thought it only used the primary video card.

Posted on 2014-08-30 20:57:27

At high resolutions, Premiere Pro definitely appears to benefit from having dual video cards. Check out the 4K chart, and look at the single vs dual cards of the same type. Even the Titan shows a nice boost with a second card, at least on the X79 platform (which is better equipped to handle dual video cards).

Posted on 2014-09-01 04:28:47
Avatar recordlifemedia

Does Adobe Premiere CS6 utilize the Open CL standard supported by the Intel HD 4000? I was able to add CUDA support on my desktop with an old GeForce 220 graphics card but my laptop with an i7 using an Intel HD4000 doesn't seem to be supported.
I've been searching for information on utilizing the HD4000 and even trying the adding graphics card to txt list in the directory for the Open cl txt file like you would for the Cuda file.

Posted on 2015-02-24 01:01:12
Avatar Maxwell Neufeldt


I just did some napkin math trying to sort out whether PPCC is utilizing single or double precision. Doing a correlation between (min)4k-seconds and inverted TFLOPs (1/TFLOPs) for the K4000/5000/Titan/W5000/W7000 I got correlation values of 0.74 for single precision and 0.04 for double precision.

Unless I've made some kind of terrible error (other than ignoring RAM), it certainly looks like PPCC is not using double precision.

Posted on 2016-01-31 20:53:03
Avatar claniusg

Great article. In this article you benchmark 2 quadro k2000s and show a performance gain but I'm under the impression that the k2000 is not SLI compatible. It isn't listed as compatible by nvidia and it doesn't have the sli connector. How did you achieve a performance increase with two of these? I'm putting together an extra editing workstation with old parts and have 2 of these available.

Posted on 2017-03-12 22:19:20

Premiere doesn't utilize SLI (that is really just for gaming) so you can use multiple GPUs without any special configuration. You should just be able to toss in multiple cards and Premiere should just use them automatically.

Keep in mind that this testing is a bit old at this point. Currently, we don't tend to recommend using multiple video cards in Premiere because a single GPU tends to be both cheaper and faster and after a certain point you don't get any faster performance with more GPU power. In your case, however, if you have multiple K2000 cards just sitting around you might as well toss them both in.

Posted on 2017-03-13 18:21:14