by
19. March 2017

I have spent a good part of the last week staring at my stopwatch and progress bars, calculating seconds and frames and typing numbers into excel tables. So now I’m excited to present you my big GTX970 vs. GTX1080 comparison.

But first, let’s take a look at the test setup:

Test system:

  • Intel I7 4930K, 6 x 3,40GHz
  • 32Gb DDR3 Ram
  • Software on Samsung 850 Pro SSD
  • Windows 10 Pro

Test footage:

Unfortunately, I can’t provide the test footage for the URSA Mini and the FS7, as they are part of ongoing projects.

After Effects:

  • Video Copilot “Rain FX” –> You can download the project file from Video Copilot here:

    New Tutorial: Realistic Rain Drop FX!


     

  • Element 3D & Saber title animation
  • Upper left title with alpha

Again, I can unfortunately not provide the Element titles and the Upper Left title as AE projects.

Software:

  • Adobe Premiere Pro CC2017
  • Adobe After Effects CC2017
  • Avid Media Composer 8.8.
  • Blackmagic Da Vinci Resolve 12.5.

GPU:

  • Palit GTX970 Jetstream (driver version: 378,78)
  • Palit GTX1080 Super Jetstream (driver version: 378,78)

Important note:

There are two important things to keep in mind when looking at the results: Accuracy and consistency.

Accuracy: You need to understand that the test results here are not 100% accurate measurements. For some of the tests, I had to use a stopwatch to get any data at all, like rendering in Premiere or Media Composer. I started at the exact moment when I hit render or dropped an effect on a clip, and I stopped when the process was finished and I could use the interface again. This works really good actually, but like I said, those are not completely accurate values.

Consistency: You do not get the same results every time you do the same test. Most of the time, the difference between two tests is just a second, but the longer the test takes, the bigger the difference. So I did run all the test multiple times and calculated an average. I will present all the results to you, but personally, I would leave out at least those results where the difference is only 1 second.

Alright, now you know the setup and what to take to care of. So let’s start with the test.

Playback:

I tested the playback in Adobe premiere pro and In Avid Media Composer. As there is no real data for this, you gotta trust me on this one.

Adobe Premiere Pro playback:

In Adobe Prmeiere Pro, I have tested the footage in it’s native resolution as well as in an HD sequence. This is not a very in-depth test. I simply checked the footage in a sequence with no effects on it.

  • Red Weapon Red Raw, 8192 x 3456, 50 fps, Redcode 9:1, native sequence, unrendered: No realtime playback, extremely choppy
     
  • Red Weapon Red Raw, 8192 x 3456, 50 fps, Redcode 9:1, native sequence, rendered: Nearly realtime playback, a little choppy
     
  • Red Weapon Red Raw, 8192 x 3456, 50 fps, Redcode 9:1, HD sequence, unrendered: Nearly realtime playback, a little choppy
     
  • Red Weapon Red Raw, 8192 x 3456, 50 fps, Redcode 9:1, HD sequence, rendered: Realtime playback
     
     
  • Arri AlexaXT Arri Raw, 2880 x 1620, 25 fps, native sequence, unrendered: No realtime playback, extremely choppy
     
  • Arri AlexaXT Arri Raw, 2880 x 1620, 25 fps, native sequence, rendered: Realtime playback
     
  • Arri AlexaXT Arri Raw, 2880 x 1620, 25 fps, HD sequence, unrendered: No realtime playback, extremely choppy
     
  • Arri AlexaXT Arri Raw, 2880 x 1620, 25 fps, HD sequence, rendered: Realtime playback
     
     
  • Arri AlexaXT Prores 4444, 2048 x 1152, 25 fps, native sequence, unrendered: Realtime playback
     
  • Arri AlexaXT Prores 4444, 2048 x 1152, 25 fps, native sequence, rendered: Realtime playback
     
  • Arri AlexaXT Prores 4444, 2048 x 1152, 25 fps,HD sequence, unrendered: Realtime playback
     
  • Arri AlexaXT Prores 4444, 2048 x 1152, 25 fps,HD sequence, rendered: Realtime playback
     
     
  • Blackmagic URSA Mini Prores 422HQ, 3840 x 2160, 25 fps, native sequence, unrendered: No realtime playback, choppy
     
  • Blackmagic URSA Mini Prores 422HQ, 3840 x 2160, 25 fps, native sequence, rendered: Realtime playback
     
  • Blackmagic URSA Mini Prores 422HQ, 3840 x 2160, 25 fps, HD sequence, unrendered: No realtime playback, choppy
     
  • Blackmagic URSA Mini Prores 422HQ, 3840 x 2160, 25 fps, HD sequence, rendered: Realtime playback
     
     
  • Sony FS7 XAVC-I 100, 4096 x 2160, 23,976 fps, native sequence, unrendered: Realtime playback
     
  • Sony FS7 XAVC-I 100, 4096 x 2160, 23,976 fps, native sequence, rendered: Realtime playback
     
  • Sony FS7 XAVC-I 100, 4096 x 2160, 23,976 fps, HD sequence, unrendered: Realtime playback
     
  • Sony FS7 XAVC-I 100, 4096 x 2160, 23,976 fps, HD sequence, rendered: Realtime playback
     

Avid Media Composer playback:

I have linked all footage via AMA, scaled it down and put it directly into a HD sequence. So there’s no native resolution test here. Also, there’s no Arri Raw test, as Media Composer does not support Arri Raw.

  • Red Weapon Red Raw, 8192 x 3456, 50 fps, Redcode 9:1, HD sequence, unrendered: No realtime playback, extremely choppy
     
  • Red Weapon Red Raw, 8192 x 3456, 50 fps, Redcode 9:1, HD sequence, rendered: Realtime playback
      
     
  • Arri AlexaXT Prores 4444, 2048 x 1152, 25 fps,HD sequence, unrendered: Realtime playback
     
  • Arri AlexaXT Prores 4444, 2048 x 1152, 25 fps,HD sequence, rendered: Realtime playback
     
     
  • Blackmagic URSA Mini Prores 422HQ, 3840 x 2160, 25 fps, HD sequence, unrendered: No realtime playback, choppy
     
  • Blackmagic URSA Mini Prores 422HQ, 3840 x 2160, 25 fps, HD sequence, rendered: Realtime playback
     
     
  • Sony FS7 XAVC-I 100, 4096 x 2160, 23,976 fps, HD sequence, unrendered: Realtime playback
     
  • Sony FS7 XAVC-I 100, 4096 x 2160, 23,976 fps, HD sequence, rendered: Realtime playback

You can see that Premiere Pro, as well as Media Composer, plays all footage in realtime when rendered. Now let#s move on to the next tests:

Adobe Premiere render:

For this test, I have measured the time it takes to render our test footage. I tested every shot in it’s native resolution and also downscaled to HD. Here are the results:

We can see in this test that the GTX970 is nearly as fast as the GTX1080. The GTX1080 shows a slightly better overall performance. When rendering Arri Raw in it’s native resolution, the GTX970 is even faster than the GTX1080, but only by 1 second. In 5 out of 10 tests, there’s no difference between the GTX970 and the GTX1080.

Adobe Premiere warp stabilizer:

For this test, I applied the warp stabilizer to the different footage in it’s native resolution.

This is kind of unexpected, but we can see in this test that the GTX970 is outperforming the GTX1080 in every test. Overall, they are still close together. The biggest difference we see is when stabilizing Arri Raw, where the GTX970 is about 5% faster.

Adobe Premiere export:

I have exported the footage in 5 different formats, and here are the results:

For the Red footage, we see the GTX1080 performing better when exporting to MXF and DNxHD, while the GTX970 is faster when exporting Animation and TIFF. For h264, both cards took the exact same time. The biggest difference is the Animation export, where the GTX970 is about 10% faster than the GTX1080.

With the Arri Raw footage, both GPUs show the exact same performance, except for the Tiff export, where the GTX970 is about 8% faster than the GTX1080.

With the Arri Prores footage, there’s no difference between the GTX970 and the GTX1080.

With the Blackmagic footage, we see both GPUs very close together. The GTX970 is slightly better with DNxHD, Animation and TIFF.

The Sony footage shows no difference between the GTX970 and the GTX1080.

The surprise here is that the GTX970 is faster than the GTX1080 in 6 of the 25 tests. In 17 tests, there is no difference between both GPUs, and only in 2 times is the GTX1080 faster than the GTX970.

 

Premiere Pro LUT and overlay export:

For this test, I have apllied a LUT to the footage with Lumetri, and I have also added a simple TIFF overlay with alpha to the footage. I used the LUTs accordingly to the footage, and for Red and Blackmagic, I used the Arri Alexa default LUT. I also set the sharpening to 100. All footage was scaled down and put in a HD sequence. For reasons I don’t remember, I forgot to test the Arri Prores footage here. Sorry for that.

The GTX1080 is slightly faster than the GTX970. Overall, both GPUs are very close together.

 

Avid Media Composer 8.8. render:

I have linked the footage via AMA, scaled it down and put it into HD sequences.As Avid does not support Arri Raw, this footage is missing from this test.

We see the biggest difference between the two GPUs when rendering the Red Raw footage, where The GTX1080 is about 8% faster. For the other footage, both are very similar.

 

Avid media Composer 8.8. stabilize

I applied the region stabilize effect to the footage, all scaled down and put into HD sequences.

We can again see that the Red Raw footage shows the biggest difference, with the GTX1080 being about 12% faster. 

 

Avid Media Composer 8.8. export:

All footage was scaled down and put into HD sequences. The reason why h264 takes so much longer than the other formats is that h264 was set to multipass.

This is the first time where we can see a constant and big gain in performance with the GTX1080. It’s faster in all of the tests, with the biggest difference being 30,58% faster for the Animation export.

The Arri Prores footage shows very similar results for both GPUs. The GTX970 is performing slightly better.

Both GPUs are very similar with the Blackmagic footage. Again, the GT970 is slightly better when exporting to TIFF.

Not much difference between the GTX970 and the GTX1080 with the Sony footage.

Media Composer shows much better performance when exporting the Red Raw footage with the GTX1080. With the other footage, both GPUs are very similar. You can also see that for the Red Raw footage, even the fastet Avid Media Composer export is still slower than Adobe Premiere.

 

Avid Media Composer trailer timeline:

To add a very real-life example, I have rendered the timeline of my latest finished project. It’s a 80 seconds 25p timeline, 1920 x 1080 with 2 alpha layers on top and 134 effects to render, mostly motion fx and audio transitions.

There is no difference at all between the GTX970 and the GTX1080.

 

Adobe After Effects “Video Copilot Rain FX”

I have rendered the whole 10 seconds timeline.

This was the biggest disappointment to me. I have expected way better performance in After Effects with the GTX1080, but we can see that both GPUs are very similar. While the GTX1080 is slightly better overall, the GTX970 outperforms the GTX1080 when exporting MXF via Media Encoder by about 7%.

Adobe After Effects Element 3D titles:

This is a project I have created some time ago. It uses 8 layers of Element 3D titles flying in one after the other, a lot of saber layers and an animated camera. All settings in Element set to ultra. It looks like this:

The GTX970 is about 7% faster when exporting to MXF via Media Encoder, and this time it’s also 6% faster when exporting to TIFF. For the h264 export, the GTX1080 is about 10% faster than the GTX970. Again, I have expected better results and a clearer overall image with the GTX1080.

Adobe After Effects cornerbug:

I have exported a simple 60 seconds upper left title with a semi-transparent background to DNxHD 185 with alpha channel. it looks like this:

The GTX970 is doing this job 30,88% faster than the GTX1080, calculating 25,03% more fps. 

Adobe After Effects warp stabilizer:

I have applied the warp stabilizer to the footage in it’s native resolution:

Again, both GPUs are very similar. The GTX1080 is performing a tiny bit better.

 

Blackmagic Da Vinci Resolve export:

I have again scaled all the footage down to a HD sequence and exported it in different formats. I have also applied a LUT, sharpened it by setting the blur radius to 0 and added a temporal noise reduction with frames set to 5, motion estimation set to better, motion range set to small and the threshold set to 25. This is also the only I have used a Cine DNG 3:1 from the URSA Mini, because Resolve is the only software that supports this format.

 

This is only the second time in all of the previous test that we see a massive and constant increase of performance with the GTX 1080. The biggest difference is the MXF export, where the GTX1080 is 26,54% faster, calculating 36,27% more fps.

The Arri Raw footage shows even bigger differences. When exporting to h264, the GTX1080 is 38,70% faster, calculating 63,26% more fps than the GTX970.

For Arri Prores, we see the same increase in performance with the GTX1080, being up to 24% faster when exporting to DNxHD.

Again, the GTX1080 performs way better than the GTX970.

With the Sony footage, the GTX1080 is also way faster than the GTX970.

Until this test, I have actually thought that my system is limited by ram and CPU power, so that the GTX1080 can just not speed up things anymore. But Resolve clearly shows that it is capable of using the full GPU power from the GTX1080, while the other software can somehow not fully benefit from the increased GPU power.

Blackmagic Da Vinci Resolve stabilize:

For the last test, i have stabilizes the footage in Resolve:

 

Resolve seems to do an amazing job when it comes to use the full GPU power provided, resulting in way better results for the GTX1080.

 

Conclusion:

It’s hard for me to draw a conclusion here. What we can definitely see is that Da Vinci Resolve is using the GPU power way better than other software. In every test in Resolve, we see a massively increased performance with GTX1080 over the GTX970.

  • GTX 970 is faster in 0 out of 24 tests
  • GTX1080 is faster in 21 out of 24 tests
  • GTX970 and GTX1080 show the same results in 3 out of 24 tests
     

For Adobe Premiere, we get mixed results. Sometimes the GTX970 is faster, and sometimes the GTX1080. Overall, both GPUs are performing very similar and there’s no clear winner here.

  • GTX 970 is faster in 12 out of 44 tests
  • GTX1080 is faster in 9 out of 24 tests
  • GTX970 and GTX1080 show the same results in 23 out of 44 tests
     

For Avid Media Composer goes basically the same as for Adobe Premiere. Although we can see the GTX performing better overall, both GPUs are very similar. The trailer timeline render shows also, that we get the exact same result with both GPUs in a very real-life scenario. It seems that media Composer is taking more advantage of the increased GPU power than Premiere, resulting in bigger differences between the GTX970 and the GTX1080. This is quite surprising to me, as Avid is not exactly known for adapting to new technology and GPUs very quickly.

  • GTX 970 is faster in 7 out of 39 tests
  • GTX1080 is faster in 14 out of 39 tests
  • GTX970 and GTX1080 show the same results in 8 out of 39 tests
     

Adobe After Effects is quite disappointing to me. In some tests, The GTX970 is actually faster than the GTX1080, while in the other tests, both GPUs are performing very similar. We see the biggest gain in performance with the GTX0180 when exporting the Element 3D titles to h264. In this test, the GTX1080 is 9,45% faster and calculates 10,71% more fps. Bu then again, when rendering the very simple upper left title, the GTX1080 is actually 30,88% slower and calculates 25,03% less fps. 

  • GTX 970 is faster in 7 out of 16 tests
  • GTX1080 is faster in 8 out of 16 tests
  • GTX970 and GTX1080 show the same results in 1 out of 16 tests
     

Should you upgrade?

The increase in performance with a new, more powerful GPU depends heavily on your system, the footage you use, the software you use and your general workflow. I work mainly with Avid Media Composer, Adobe Premiere Pro and Adobe After Effects. I do use Da Vinci Resolve for color grading, but not as an NLE. I rarely work with footage bigger than 4K. So personally, I decided that the increase in performance is not worth paying about 550€ for the GTX1080, because the GTX970 is still doing a great job. I’m currently waiting for the release of Skylake X to build a new rig, and I decided to upgrade the GPU then.

You really need to take all theses factors into consideration. The best thing you can do is to test a new GPU on your system, with your software,  footage and your workflow.

If you work with Da Vinci Resolve most of your time, then I would say yes, now is the right time to upgrade. 

 

I hope this can help you guys by deciding if you should upgrade you GPU or not. Tanks for reading, and please leave your thoughts in the comments.

Share on FacebookShare on LinkedInShare on Google+Email this to someone
0 Comments

Leave a Reply

*