Puget Systems print logo
Read this article at https://www.pugetsystems.com/guides/1565
Article Thumbnail

Does PCIe Gen4 improve GPU performance in video editing apps?

Written on August 27, 2019 by Matt Bach


PCIe Gen4 is finally here, boasting twice the theoretical bandwidth as PCIe Gen3. However, there is currently only a single platform (AMD's x570 chipset) and a handful of video cards (AMD's Radeon 5700 line) that support it. Even though PCIe Gen4 is in its infancy right now and things will likely change over time, this raises the question: does PCIe Gen4 matter for video editing?

Does PCIe Gen4 matter for video editing?

This is actually a very complex question since PCIe speed not only affects things like GPU performance, but it also allows for faster storage and other devices. To keep things simple, for now we are going to strictly look at how PCIe Gen4 affects performance with one of the few Gen4 capable video cards - the AMD Radeon RX 5700 XT 8GB. To provide a point of reference, we will also be including NVIDIA's GeForce RTX 2060 SUPER 8GB using PCIe Gen3 since it has the exact same MSRP as the 5700 XT.

Note that this article uses the same benchmarks as our recent "GPU Roundup: NVIDIA SUPER vs AMD RX 5700 XT" articles for Premiere Pro, After Effects, DaVinci Resolve, and Photoshop. So, if you wish to know how the performance stacks up against any of the higher-end GPUs from AMD and NVIDIA, the scores can be directly compared.

If you would like to skip over our test setup and benchmark sections, feel free to jump right to the Conclusion.

Looking for a Video Editing Workstation?

Puget Systems offers a range of powerful 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!

Test Setup & Methodology

Listed below are the specifications of the systems we will be using for our testing:

Test Platform
CPU AMD Ryzen 9 3900X
CPU Cooler Noctua NH-U12S
Motherboard Gigabyte X570 AORUS ULTRA
RAM 4x DDR4-3000 16GB (64GB total)
Video Card NVIDIA GeForce RTX 2060 SUPER 8GB
AMD Radeon RX 5700 XT 8GB
Hard Drive Samsung 960 Pro 1TB
Software Windows 10 Pro 64-bit (version 1903)

*All the latest drivers, OS updates, BIOS, and firmware applied as of August 25th, 2019

The benchmarks we will be using are the latest release of our public benchmarks. Full details on what is tested as well as links to download and run them yourself are available at:

Raw Benchmark Data

While our benchmarks present various scores based on the performance of each test, we also wanted to provide the individual results. If there is a specific task that is a hindrance to your workflow, examining the raw results for that task is going to be much more applicable than any of the overall scores. Feel free to skip to the next section for our analysis of these results if you have no desire to pour through thousands of numbers.

Premiere Pro Benchmark Results

PCIe Gen4 Premiere Pro Performance Benchmark

Starting off with the Overall Score from our Premiere Pro benchmark, we saw no difference between PCIe Gen3 and Gen4 with the Radeon 5700 XT. However, most of the tests in our Premiere Pro benchmark are not especially taxing on the GPU, so let's take a look at just the results for our "4K Heavy GPU Effects" test sequence:

PCIe Gen4 Premiere Pro Heavy GPU Effects Performance Benchmark

Our "Heavy GPU Effects" sequence is intended to be an unrealistic example the maximum performance difference between different GPUs in Premiere Pro and the best chance for PCIe Gen4 to show a performance advantage. It includes only GPU-accelerated transitions and effects such as Cross Dissolve, Dip to Black, Dip to White, Wipe, Lumetri Color, VR De-Noise, Ultra Key, Sharpen, Gaussian Blur, Basic 3D, VR Digital Glitch, and Directional Blur.

Unfortunately, the results for Gen3 and Gen4 with the 5700 XT are all within 1 FPS of each other which is within our margin of error for this type of test. This means that with the fastest PCIe Gen4 capable GPU currently available, there is no performance gain to be had with PCIe Gen4 in Premiere Pro.

After Effects Benchmark Results

PCIe Gen4 After Effects Performance Benchmark

After Effects uses the GPU even less than Premiere Pro, so the Overall Score for this benchmark really doesn't tell us much. Between Gen3 and Gen4, the 5700X scores were less than 1% apart, and even the RTX 2060 SUPER scores were effectively the same as the 5700 XT. In order to get a closer look at one of the few times we did see a performance gain with PCIe Gen4, let's examine just the RAM Preview result from our "GPU Stress" project:

PCIe Gen4 After Effects GPU Stress Performance Benchmark

Looking at just this one test, we saw a small 3% gain in playback FPS with the Radeon RX 5700 XT using PCIe Gen4 over Gen3. This is a really small difference, and means that at least at the performance level of the 5700 XT, there shouldn't be a realistic situation where PCIe Gen4 matters for GPU performance in After Effects.

DaVinci Resolve Studio Benchmark Results

PCIe Gen4 DaVinci Resolve Studio Performance Benchmark

DaVinci Resolve is a very GPU-focused application which means we should see a much larger performance impact with PCIe Gen4 compared to Premiere Pro or After Effects. And in fact, we did see a larger benefit, although it was only by about 5% overall.

What is interesting is that when we examined the individual results, it was not the OpenFX or TNR tests that tended to show the largest increase in performance - it was our "Basic Grade" tests:

PCIe Gen4 DaVinci Resolve Studio Basic Grade Performance Benchmark

Averaging all our "Basic Grade" tests with the various codecs we use, we saw about an 8% increase in FPS with PCIe Gen4. This isn't a huge increase - and not enough to allow the 5700 XT to close with the RTX 2060 SUPER - but it should be noticeable in the real world.

Dialing down things even finer, the largest performance gain across all the individual tests was surprisingly with ProRes 4444 footage running the "Optimized Media" test:

PCIe Gen4 DaVinci Resolve Studio maximum performance increase

The fact that this test showed the largest performance gains with PCIe Gen4 took us completely by surprise since this is essentially testing straight transcode performance with no grading whatsoever. Typically, the CPU makes a much larger difference than the GPU for this, but for whatever reason the extra bandwidth from PCIe Gen4 was able to increase performance by about 20%.

Photoshop Benchmark Results

PCIe Gen4 Photoshop Performance Benchmark

We often round out our video editing articles with Photoshop since even though it is not a true video editing application (unless you are one of the rare people that uses the video timeline features), it is very commonly used alongside Premiere Pro, After Effects, and similar applications.

Similar to After Effects, faster GPUs really don't make a huge impact in Photoshop in most situations, so from an overall perspective we saw no difference between the 5700 XT with PCIe Gen3 or Gen4. However, our Photoshop benchmark provides a "GPU Score" that is based on the performance of only the GPU-accelerated effects we test:

PCIe Gen4 Photoshop GPU Performance Benchmark

Unfortunately, even though the GPU Score is able to show a performance benefit from faster GPUs, we still saw no real difference between PCIe Gen4 and Gen3 with the Radeon 5700 XT in Photoshop.

Does PCIe Gen4 improve GPU performance in video editing apps?

With the AMD Radeon RX 5700 XT, we saw little to no performance advantage with PCIe Gen4 over Gen3 in most of our video editing benchmarks. The exception was DaVinci Resolve where we saw a 5% performance increase overall, and up to a maximum performance gain of 20% in one specific test. However, it is worth noting that even in the best case, the NVIDIA GeForce RTX 2060 SUPER scored on par or better than the 5700 XT even though it is limited to PCIe Gen3.

Keep in mind that PCIe Gen4 is in its infancy with only one platform (AMD x570) and a handful of mid-range GPUs like the AMD Radeon RX 5700 XT that support it. As higher-end video cards with Gen4 support become available, at some point we should start to see a larger performance benefit. When that will happen is anyone's guess, but there is certainly an argument for getting a PCIe Gen4 capable system strictly from a future-proofing standpoint.

In addition, since Gen4 has twice the bandwidth of Gen3, you can get the same performance from half the number of PCIe lanes. With the growing popularity of PCIe-based NVMe storage (for which Gen4 can be very significant), being able to get the same performance from fewer lanes can at times allow for more flexibility when configuring your system.

At the moment, PCIe Gen4 likely isn't a major concern for video editors from a GPU performance standpoint. Technology is constantly changing and evolving, however, so it will be interesting to see how this changes as more powerful GPUs come out with support for PCIe Gen4.

Looking for a Video Editing 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: Video Editing, Premiere Pro, After Effects, DaVinci Resolve, Photoshop, Intel vs AMD, Xeon W, W-3225, W-3245, W-3265
Jan Dorniak

Too bad current Radeons don't support multi GPU setups. In cases where there are limited lanes - say a multi GPU where some cards work at x8 - is where PCIe Gen4 would really shine. Or in general: if the extra bandwidth doesn't help why not use less lanes? In theory PCIe Gen4 x8 has the same bandwidth as PCIe Gen3 x16.

Is it possible for you to test the 5700 XT at Gen4 x8?

Posted on 2019-08-28 11:30:34
Misha Engel

Programs like Davinci Resolve can support upto 8 GPU's under Linux (NVidia and AMD) and 4 GPU's under Windows(NVidia and AMD) and MacOS(AMD only since 10.14 mojave).

Posted on 2019-08-28 15:04:27

Radeon cards definitely do support multi-GPU since pro apps like Resolve or rendering don't want or need SLI/Crossfire - most of the time, it actually breaks things. For example, I know that two NVIDIA cards in SLI makes it so Resolve only actually sees one of the cards rather than both of them.

The reason we didn't really dive deep into things like PCIe lanes and multi-GPU configurations is frankly because the Radeon RX 5700 XT does so poorly in these applications that we don't want to spend a huge amount of time testing things like x8/x4 or multi-GPU configurations since the results are probably going to be way different with a faster GPU. Once higher-end cards come out, especially NVIDIA cards so we can include CUDA-only rendering engines as well, we will do a much more in-depth analysis.

I'm personally very curious how Resolve will perform in multi-GPU setups with Gen4 since we know that with Gen3, the difference between x16 and x8 is typically only a few percent. That may mean that Gen4 x4 will be viable, but it seems like AMD/Intel are starting to get more PCIe lanes in general so I'm not sure how common using a card at x4 will be once those faster cards come out. But that's the fun of technology, it is always changing!

Posted on 2019-08-28 16:58:38
Jan Dorniak

One one hand - yes, they are getting more lanes. But on the other - there are more and more things which use PCIe lanes. USB 3.1 Gen2, NVMe disks, ethernet, WiFi, the list just gets longer and longer. Maybe it's time to start using B or B+M keys on NVMe SSDs?

SLI/Crossfire - my bad for being misinformed. Doesn't that "single card" show more compute units (or whatever) though? Will be interesting how two chips connected with Infinity Fabric on a single card work out in the new Mac Pro. Assuming AMD gets it in shape.

And you are right - those 5700s disappoint. Greatly. I just hope it's just another case of AMDs software being horrible for the release. Looking at games it does beat the 2060 SUPER in most games. And as much as I have curiosity - it makes sense for you to not bother with those new AMDs, at least for now.

Posted on 2019-08-28 17:37:02

Completely agree on the PCIe devices - for us it seems to be NVMe storage that is the biggest contributor. The pricing is getting really good now that most of the systems we sell have at least one NVMe (usually M.2) drive in it. It isn't that uncommon that a system won't have any SATA drives at all to be honest - although that usually only happens with customer that have NAS storage since SATA is still the way to go for larger capacities.

The last time we did formal testing for NVLINK (SLI) in Resolve, it flat out didn't use the second card: https://www.pugetsystems.co... . It is possible that AMD Crossfire is different, and I'm sure Blackmagic will get it fixed eventually, but my bet is that the new Mac Pro will simply use multiple GPUs as accelerators (no SLI/Crossfire). To be honest, that is how most of the multi-GPU workstations we sell go out the door. Only ML/AI devs or gamers want to use NVLINK/SLI at this point.

Posted on 2019-08-28 17:42:25
Jan Dorniak

Most cards for the new Mac Pro, yes. But not the Radeon Pro Vega II Duo. You get two Pro Vega II connected with Infinity Fabric (the same one which connects the chiplets in the new Ryzens) which then plug in two PCIe Gen3 x16 connectors (the new, Apple-proprietary, MPX connector). It will be interesting how this is handled both in the OS and by 3rd party software.

Posted on 2019-08-28 18:29:29
Michael D. Andreas

It makes total sense to me that simpler grades have more of a speed boost than complex grades and processing like TNR. The video frame data has to be transferred to and from the GPU regardless of how heavy or light the processing. So if the GPU processes the frame less the data transfer interface will be exercised for the larger portion of the total time and you'll see a larger proportional increase.

I understand that Bitcoin mining machines provide low bandwidth interfaces to multiple GPUs, so probably the GPU spends a long time munching on data that takes a relatively short time to transform.

Posted on 2019-11-05 05:01:42

Apologies if you've already done the testing, but have you done any testing to see the differences for Video editing between PCIE3 and 4 as pertaining to NVME drives? I noticed in the results above you were using a PCIE3 drive.

Posted on 2019-11-05 18:16:28

is this all about the games and mining?
Natalie | blogger | easybuilder.pro

Posted on 2019-12-14 03:51:45
Jakub Badełek

Matt, are you planning to test PCIe 4 from storage point of view?

Posted on 2020-03-10 11:41:01

Next time we do storage testing it will be included, but we are not prioritizing that. Going from SATA to even a 3.5GB/s NVMe drive rarely gives a significant benefit, so it is extremely unlikely that the even faster PCIe Gen4 drives will be any better for these kinds of workloads. Especially since it is really just the sequential read/writes that are faster, and not the speed for small files like cache.

Posted on 2020-03-10 15:59:47
Frank Varro

Are there any plans to re-test this now that AMD sorted out their terrible GPU drivers issues? Is it possible the drop in performance(even if negligible) from PCIE3 to PCIE4 was a driver issue? Also are there thoughts on testing the 2060KO? or at this point are you holding off for 3000 series/Big Navi?

Posted on 2020-08-10 00:52:42