Premiere Pro CC 2018 GPU Performance: NVIDIA Titan V 12GBWritten on February 7, 2018 by Matt Bach
The NVIDIA Titan V 12GB is the latest in NVIDIA's Titan line of GPUs and while the $3,000 price tag is pretty steep, it has a number of features that makes it very unique. The two most interesting being full-speed FP64 (double precision) performance and the addition of 640 "Tensor" cores. These special cores are designed specifically for use in machine learning, but both they and the FP64 performance unfortunately has no bearing on performance in software like Premiere Pro.
This doesn't make the Titan V a complete dud for Premiere Pro, however. This card also features HBM2 memory (which is faster than the GDDR5/GDDR5X VRAM found on most cards) and has more CUDA cores than any other recent GTX or Titan GPU. In fact, it has 5120 CUDA Cores while the Titan Xp and GTX 1080 Ti only have 3840 and 3584 CUDA cores respectively. One thing to note is that while the higher number of cores gives the Titan V a theoretical higher performance ceiling than any other GTX or Titan video card, each individual core in the Titan V is actually slower than the cores in cards like the GTX 1080 Ti or Titan Xp. So just like how a CPU with more cores is not always better, there is a chance that the lower frequency of the Titan V holds it back from being a top pick for software like Premiere Pro.
To find out, we will be looking at how this card performs in a number of realistic situations using our newly revamped Premiere Pro testing process. This new process is not vastly different than our old testing, but simply focuses on a wider range of codecs and has more in-depth Live Playback testing. More information can be found in the Test Hardware & Methodology section.
Test Hardware & Methodology
To see how the new Titan V performs in Premiere Pro, we opted to pair a number of different GPUs with the Intel Core i9 7940X CPU which is currently the fastest all-around CPU for Premiere Pro. Our full test platform consists of the following hardware:
|Skylake-X (X299) Test Platform|
|Motherboard:||Gigabyte X299 AORUS Gaming 7
|RAM:||8x DDR4-2666 16GB
|Hard Drive:||Samsung 960 Pro 1TB M.2 PCI-E x4 NVMe SSD|
|OS:||Windows 10 Pro 64-bit|
|Software:||Premiere Pro CC 2018|
To act as comparison points, we will be testing a number of NVIDIA GPUs ranging from the new Titan V all the way down to the GTX 1060.
|Test Video Cards|
|NVIDIA GeForce GTX 1060 6GB||NVIDIA GeForce GTX 1070 8GB|
|NVIDIA GeForce GTX 1080 8GB||NVIDIA GeForce GTX 1080 Ti 11GB|
|NVIDIA Titan Xp 12GB||NVIDIA Titan V 12GB|
For those that read our articles regularly, you will notice that our testing looks quite a bit different since we just recently completely re-vamped our entire testing process for Premiere Pro based on feedback from our customers and readers. The biggest change is that we are now testing a wider range of codecs using media that is either publicly available for download or transcoded from media that is publicly available. This was done so that anyone can repeat our testing in order to both verify our findings and to see how their current computer stacks up to the latest hardware available.
However, in order to keep our testing process from taking weeks or months to complete, we had to drop some of the variety of testing and instead focus on just two aspects of Premiere Pro: Exporting and Live Playback. This isn't to say that the other things we used to test (such as rendering previews and warp stabilize) were not useful, but we simply had to cut something out in order to make time for the wider range of codecs. In fact, we may need to cut a bit more in the future as a single benchmark loop currently takes more than 18 hours to complete.
The footage used in our testing is shown below with links to where you can download it yourself:
|H.264||3840x2160||59.94 fps||Panasonic GH5||60p Snow Handheld||The Angry Video Guy
Panasonic GH5 Sample Footage
|ProRes 422 HQ||3840x2160||24 fps||Ursa Mini 4K||City Train Station||Blackmagic Design
Production Camera 4K Update
|ProRes 4444||3840x2160||59.94 fps||Canon C200||Untitled00024199||4K Shooters
Canon C200 Raw Footage Workflow
|CinemaDNG||4608x2592||24 fps||Ursa Mini 4K||Interior Office||Blackmagic Design
|ARRIRAW||6560x3100||23.976 fps||ALEXA 65||A003C025
(Open Gate spherical)
ALEXA Sample Footage
|23.976 fps||EPIC DRAGON||A016_C001_02073O_001||RED
Sample R3D Files
|29.97 fps||RED ONE MYSTERIUM||A004_C186_011278_001||RED
Sample R3D Files
|23.976 fps||EPIC DRAGON||A007_C115_07181B_001||RED
Sample R3D Files
|23.976 fps||WEAPON 6K||S005_L001_0220LI_001||RED
Sample R3D Files
|23.976 fps||WEAPON 8K S35||S002_C074_02065Z_001||RED
Sample R3D Files
|25 fps||WEAPON 8K S35||B001_C096_0902AP_001||RED
Sample R3D Files
|23.976 fps||EPIC-W 8K S35||S002_C074_02065Z_001||RED
Sample R3D Files
|3940x2160||29.97 fps||Transcoded from RED A004_C186_011278_001|
|6144x3160||23.976 fps||Transcoded from RED A007_C115_07181B_001|
|8192x4320||25 fps||Transcoded from RED B001_C096_0902AP_001|
While this is by no means every codec available, we do feel that this covers a much wider range than our previous testing. In the future we may cut down on the number of RED clips and replace then with something like XAVC-S or AVCHD but we really wanted to see how the different compression levels impact performance.
Our testing was done with three different timelines to simulate different types of workloads.
To be clear, these test sequences are not anything fancy and frankly don't end up with a very nice final product since they are simply copies of the same clip over and over. However, if you wish to replicate our testing we have all the project files and export presets available for download. You should be able to replicate any of our tests by downloading the clips from the original source, making 10 copies of each clip, and re-linking the media in the appropriate project file. Export time was simply recorded from when the "Export" button was clicked until the export completed. Live playback FPS (frames per second) was measured based the number of dropped frames relative to the total number of frames in the sequence. For example, if we dropped 100 frames in the 4K ProRes 422 HQ "Basic" timeline (which has 579 frames), that means we rendered a total of 479 frames over 24 seconds for an overall result of ~20 FPS. For ease of comparison, we also have the raw number of dropped frames listed in the Live Playback section.
Live playback performance is a challenge to accurately test since performance depends not only on the codec and resolution of your media, but also the playback resolution (full, half, quarter, etc.) and the effects you have applied. Since this is the first full round of testing with our new benchmark process, we decided to go all out and benchmark all our test media with not only three different timelines, but also at full, half, and quarter playback resolution. For more information on each test media and timeline, check out the Test Hardware & Methodology section.
Due to the huge amount of data gathered in our testing (over 800 different data points in total!), we are not going to go through everything point by point. If we tested a codec you typically work with, we highly recommend looking at the "Individual Avg. FPS Results" charts or the raw Dropped Frame results by clicking on the "Show Raw Dropped Frames Results" link. However, for a general overview of how each GPU performed we decided to average the results relative to a GTX 1080 8GB video card.
With our new approach to playback testing, we are able to get some very interesting insights into how the video card affects live playback performance. First of all, we have to point out that both the Titan XP and the new Titan V are not great choices if live playback in Premiere Pro is a primary concern. With very few exceptions, the GeForce GTX 1080 Ti was as fast or faster than either of the Titan cards at a much lower cost. In fact, in some cases (4K ProRes 4444 and 6K RED 12:1 with the "Heavy Effects" sequence) the GTX 1080 Ti was not just a little bit better, but was able to play the sequence at 50% higher FPS compared to the Titan cards!
Another interesting discovery was that the lower the playback resolution, the less the GPU model impacts performance. Of course, at a certain point the GPU doesn't matter since you can't get any better than 0 dropped frames, but even when we were dropping a decent number of frames the difference between the GTX 1060 6GB and the GTX 1080 Ti at half or quarter resolution was often less than 1-2 FPS. This suggests that at lower playback resolutions, the GPU takes a back seat to the CPU in terms of importance. This is a very significant finding since with 6K and 8K footage you will likely be dropping to half or quarter playback resolution if you add a decent number of effects in order to maintain near real-time playback. If this is something you will be doing regularly, then you are likely better off with a slightly lower end GPU and spending the cost savings on a more powerful CPU if possible.
Exporting is one of the biggest single time sinks for a Premiere Pro user and is often the go-to metric for measuring performance. For this test, we will be examining 4K, 6K, and 8K projects using all the different codecs listed in the Test Hardware & Methodology section. Added up, this works out to just over 40 different individual tests for each GPU resulting in just under 250 total data points.
Due to the amount of data we collected during our testing, we once again decided to compile the results into a single overall average relative to the GTX 1080 8GB video card. If you are primarily concerned about just one of the codecs or resolutions we tested, we highly recommend checking out the individual results - just click on any of the thumbnails to view the full-sized chart.
With that said, from an overall perspective we saw a small but noticeable performance improvement as we used more and more powerful (and expensive) GPUs. However, keep in mind that this is just an overall average. In some cases - like when using ARRIRAW footage or when exporting 8K projects to 8K H.265 - all the GPUs performed roughly the same. Other times, we saw a huge improvement in export times such as when using DNxHR HQ 8-bit media where the export times were often cut in half going from a GTX 1060 6GB to a Titan V 12GB.
Looking at just Live Playback at full resolution with the "Heavy Effects" timeline and overall Exporting performance, we get a pretty good idea about how the Titan V stacks up against the other video cards. Frankly, it is a bit of a mixed bag - especially considering that the Titan V is more than 4x the cost of the GTX 1080 Ti.
If exporting is your primary bottleneck, then you might be able to justify the cost of the Titan V, but it is only ~6% faster than the Titan Xp or ~8.5% faster than the GTX 1080 Ti. In terms of real numbers, this means that going from a GTX 1080 Ti to a Titan V will save about 5 minutes on a 60 minute render. Most people would not consider this to be worth the roughly $2,300 increase in cost, but for those that do a lot of rendering with a dedicated render workstation or server it very well may.
Moving on to Live Playback, this is unfortunately where the Titan V (and Titan Xp) fall short. In pretty much every test that put a decent load on the GPU, the GTX 1080 Ti was faster than both the Titan Xp and Titan V. We do want to point out again that in our full testing we saw minimal differences between all GPUs at half and quarter playback resolutions, but if you are going for full res playback then the GTX 1080 Ti is clearly a better choice than either Titan card. The Titan V is not exceptionally bad by any stretch, but you can simply get better performance at a much lower cost with a GTX 1080 Ti card instead.
Premiere Pro Workstations
Intel w/ Thunderbolt 3
Related Hardware Analysis Articles:
- Core i7 9700K & i9 9900K Performance
- NVIDIA GeForce RTX 2080 & 2080 Ti Performance
- AMD Threadripper 2990WX & 2950X Performance
- NVIDIA GeForce vs AMD Radeon Vega
- iMac Pro & Mac Pro vs PC Workstation
- Workstation GPU Performance
- GPU Performance: NVIDIA Titan V 12GB