Puget Systems print logo

https://www.pugetsystems.com

Read this article at https://www.pugetsystems.com/guides/1012
Article Thumbnail

Premiere Pro CC 2017.1.2 CPU Comparison: Skylake-X vs Threadripper

Written on August 30, 2017 by Matt Bach
Share:

Introduction

When AMD launched their Ryzen CPUs back in March, they became a serious challenger to Intel in many markets including video editing with software like Premiere Pro. Intel fired back in June with their 6-10 core Skylake-X CPUs which made Intel once again the go-to choice for most video editors using Premiere Pro but the dust hasn't even settled and both Intel and AMD are again launching new products. In mid-August, AMD launched their 12 and 16 core Threadripper CPUs that are designed specifically for workstation workloads. On Intel's side, the Core i9 7920X 12 core Skylake-X CPU launched a few days ago (August 28th, 2017) and have another three CPUs with 14, 16, and 18 cores coming in September.

With all these new CPUs, it can be very tough to keep track of what CPU is the best for different applications so today we are going to look specifically at how AMD's Threadripper CPUs compare to the currently available Intel Skylake-X CPUs in Premiere Pro.

There are a wide variety of tasks we could test in Premiere Pro to see how these CPUs perform, but in this article we will specifically be looking at:

  1. Rendering previews
  2. Exporting
  3. Performing a Warp Stabilization
  4. Live playback performance

Our testing includes test footage with resolutions of 4K, 6K, and 8K using six different codecs (more information in the test setup section). In total, we ran nearly 90 unique tests on 6 different CPUs (along with testing both Creator and Game Mode on Threadripper) resulting in more than 600 data points. If you would rather skip over our analysis of the individual benchmarks, feel free to jump right to the conclusion section.

Test Setup

Our test platforms for Skylake-X and Threadripper are listed below, but we did want to point out that the RAM configuration changed a little bit depending on the CPU. For the majority of the CPUs we used DDR4-2666 RAM but since the Core i7 7800X only natively supports DDR4-2400 we tested with that RAM instead.

Before getting into our testing, we want to point out is that while our test platforms are using a single hard drive, that is not actually what we would typically recommend to our customers. We have found in our testing that using at least a two drive configuration with the media cache and scratch files on a secondary drive can make a big impact when it comes to importing footage and tasks like conforming audio. However, since we will not be testing any of these actions in this article we opted to use a single drive simply to cut down on the number of variables.

Most of the media we will be using is available from the Sample R3D Files and transcoded to the various codecs we wanted to test.

4K

EPIC DRAGON
3840x2160
23.976 FPS

6K

WEAPON 6K
6144x3077
23.976 FPS

8K

WEAPON 8K S35
8192 x 3456
50 FPS

To test exporting and rendering previews we used a moderately complex timeline involving multiple clips, Lumetri Color, multicam footage, and some other effects like a logo overlay, Gaussian Blur and Cross Dissolves. If you want a more in-depth look at what our timelines look like, we recorded a short video explaining our test process:

Our 4K VR testing was performed using the "Sample 1 - Ring road motorbike ride" footage from the Autopano Video Benchmarking page. We tested using both some built-in Premiere Pro effects (Lumetri Color, text overlay, and cross dissolve) as well as using the Mettle Skybox 360 VR Tools and Skybox 360/VR Transitions plug-ins to apply a number of effects such as Denoise, Rotate Sphere, Sharpen, and Iris Wipe. We typically try to avoid using plug-ins in our testing, but since Premiere Pro only has basic support for VR at the moment we felt it made sense to also look at the popular Mettle plug-in for VR projects. In addition, Adobe recently acquired Mettle Skybox which means all Creative Cloud customers will be able to use this plugin for free by the end of the year.

Threadripper: Creator Mode vs Game Mode

Before we start comparing Intel and AMD, the first thing we want to do was to ensure that we are testing Threadripper in the proper mode. Since Threadripper is in essence just two Ryzen CPUs combined onto a single device, it can have some performance issues depending on how the application is coded. To be clear, this is not an AMD-only issue and is something we have seen from Intel as well with dual Xeon workstations.

You can read the fine details in AMD's blog about using Threadripper for gaming but the short of it is that some applications simply don't work well with very high core counts - especially when those cores are spread across multiple "CPUs". To try to alleviate these issues, AMD has provided the ability to switch between two modes through the AMD Ryzen Master software:

  1. Creator Mode (default) - All cores are used
  2. Game Mode - Half the cores are disabled which turns Threadripper into essentially a Ryzen CPU with more PCIe lanes and memory bandwidth.

The big downside to this approach is that switching modes requires you to restart your computer so it isn't something you can do on the fly. Since Premiere Pro typically scales decently well with more cores we don't expect Game Mode to give any sort of performance boost but we are curious as to what the difference actually is between the two modes:

Premiere Pro Threadripper Game Mode vs Creator Mode

[+] Show Raw Results

Feel free to examine all the raw data by expanding the "Show Raw Results" link, but the data is pretty straight-forward: Creator Mode is definitely the right choice for Premiere Pro. In fact, it wasn't even close as the default Creator Mode resulted in ~15-25% higher performance. Due to this, we are going to continue our testing without worrying about including any Game Mode results.

Rendering Previews

Now that we know that Creator Mode is definitely the way to go with Threadripper, the first task in Premiere Pro we want to compare Threadripper to Skylake-X is Rendering previews. This is something you never really want to have to do since it interrupts your workflow, but if you do complex editing it is sometimes unavoidable. Because of this, being able to render previews as quickly as possible is often an important part of a Premiere Pro workstation.

Premiere Pro Skylake-X vs Threadripper Render Previews Benchmark

[+] Show Raw Results

Since we are comparing 6 different CPUs across 16 different projects that include a range of resolutions and source codecs, it would take us a long time to go through the results one by one. Because of this, we decided to compile all the results into an overall average for each CPU compared to the Intel Core i7 7800X. As the lowest cost CPU in both the Skylake-X and Threadripper lines, it should be a great comparison point to judge the other CPUs against. If you wish to examine the raw results yourself, you can do so by clicking on the "Show Raw Results" link under the chart.

The Core i7 7800X is quite a bit worse than the other CPUs (likely due in large part to the lack of Turbo Boost 3.0 support), but surprisingly the other CPUs all performed within ~4% of each other. The only real interesting thing to point out is that the Core i9 7920X was actually slower than both the Intel Core i9 7900X and the AMD Threadripper 1950X even though it costs about $200 more. Otherwise, the CPUs largely performed about where you would expect given their price points. The Core i7 7820X was a hair faster than the more expensive TR 1920X, however, which makes Intel a bit stronger overall for this test.

Exporting

Exporting is one of the biggest time sinks for a Premiere Pro user and is often the go-to metric for measuring performance. For this test, we looked at 35 different combinations of source footage and export settings. This includes 4K, 6K, and 8K resolutions along with H.264, DNxHR HQ, ProRes 422 HQ, ProRes 4444, RED, and H.265 codecs. In addition, we will also be looking at VR projects using both built-in effects as well as with the popular Mettle plugin.

Premiere Pro Skylake-X vs Threadripper Exporting Benchmark

[+] Show Raw Results

Just like in the previous section, since we are comparing 6 different CPUs across 35 different projects that use a range of resolutions, source codecs, and export settings, we decided to compile all the results into an overall average for each CPU compared to the Intel Core i7 7800X. If you have the time and will, feel free to examine the raw results yourself by clicking on the "Show Raw Results" link below the chart.

The results for exporting are about as straight-forward as you can get. Whether you go with Intel or AMD, the more money you spend on the CPU, the faster your exports will complete. It is worth noting however, that even though the AMD Threadripper 1950X is technically $1 more than the Intel Core i9 7900X, it was 3% faster on average for exporting so for this test the 1950X is definitely the stronger choice over the Core i9 7900X. However, note that this performance gain was almost entirely when exporting 4K media to 4K DNxHR HQ 8-bit and when exporting 8K media to 8K H.265. So if you primarily export to H.264, the Core i9 7900X should actually faster than the AMD TR 1950X.

Overall, there isn't really a clear winner for this section between Intel and AMD, but given the strong performance of the AMD TR 1950X we are going to give the edge to AMD.

Warp Stabilize

While exporting and rendering previews may be the easiest and most common thing to benchmark in Premiere Pro, we have received a lot of feedback that performing a warp stabilize is another task where high performance is important. We have found that the time it takes to complete a warp stabilize analysis does not vary much on the source codec, but the resolution does make a big difference so we will be testing with a 4K H.264 clip as well as a 8K ProRes 4444 clip.

Since warp stabilize is not well threaded (meaning that it does not take great advantage of multiple CPU cores), we also opted to split our test clip into multiple parts and analyze all of them at the same time in order to force Premiere into making more effective use of all the CPU cores. This is a trick some people use to speed up the time it takes to analyze a single clip, but it is also a great indicator of performance when you have multiple clips that need to be stabilized. With this in mind, we not only timed how long it takes to apply a warp stabilization effect to a single 10 second clip, but also how long it takes if the clip is split into 2, 4, 8, and 16 "sub-clips" that are all analyzed at the same time.

Premiere Pro Skylake-X vs Threadripper Warp Stabilize Benchmark

[+] Show Raw Results

While we were able to get away with an overall average in the last two sections, the results for warp stabilize depends so heavily on the number of clips you are analyzing that we decided to show both the average result with a single clip and with 16 clips. Feel free to examine the raw results if you would like to see the results for 2, 4, and 8 clips, but the results for those fell between the single and 16 clip results about as you would expect so we opted to not muddy the chart with even more data.

Starting with a single clip, all of the Skylake-X CPUs interestingly performed within a percent or two of each other. Due to the lower single-threaded performance of the AMD CPUs, however, Threadripper was about 20% slower than Skylake-X when stabilizing a single clip. With the number of clips increased to 16, the Threadripper CPUs compare a bit more favorably to the Intel Skylake-X CPUs, although they are still around 10% slower than most of the Intel CPUs.

Live Playback

Live playback performance is a challenge for us to accurately test since whether you can play a timeline at full, half, quarter, etc. resolution is highly dependent not only on your source footage but also what effects you have applied to the timeline. To try to keep things universally applicable, we opted to test 10 different projects using 4K, 6K, and 8K footage with multiple codecs across three relatively simple timelines. What we wanted to see was if we would be able to play the timeline at either full or half resolution without dropping any frames - even if it was just one or two at the very start of playback.

Basic

  • 4 clips in series
  • No effects
  • No transitions

Lumetri Color

  • 4 clips via multicam sequence
  • Lumetri Color Correction
  • No transitions

Lumetri & Cross Dissolve

  • 4 clips via multicam sequence
  • Lumetri Color Correction
  • Cross dissolve

Premiere Pro Skylake-X vs Threadripper Live Playback Benchmark

[+] Show Raw Results

For a general overview of how each CPU performed for live playback, we created a scoring system based on how many timelines each CPU was able to run at either full or half resolution. Every successful playback without dropping any frames at full resolution is 1 point while every playback at half resolution is .5 points. However, if there is a codec you typically use that we happened to include in our testing, we highly recommend taking a look at the raw results to see how each CPU performed for that specific codec. 

Using our score-based system, you can get a pretty decent idea of how each CPU might affect live playback performance. Honestly, the results were a lot closer than we expected with only the Core i9 7900X being truly better than the others. One thing we do want to point out, however, is that while the Threadripper CPUs had some issues with 4K ProRes 4444 timelines, they were able to play basic 8K RED timelines at half resolution - a feat that only the Intel Core i9 7920X could match.

Conclusion

Competition is a wonderful thing and one thing is clear: Threadripper is definitely a competitive product. Looking at the overall performance including rendering previews, exporting, warp stabilize, and live playback, we saw the following average performance from each CPU we tested:

Premiere Pro Skylake-X vs Threadripper Overall Benchmark Results

Starting with the AMD Threadripper 1920X 12 core, despite being $200 more expensive than the Intel Core i7 7820X we really have to give the edge to the Core i7 7820X as the overall the better choice for Premiere Pro. The two CPUs are basically identical when exporting and rendering previews, but the Core i7 7820X is much faster for warp stabilize and has better live playback performance as well. Warp stabilize might not be terribly important for many Premiere Pro users, but the improved live playback performance is a big deal and makes the Core i7 7820X a solidly better choice over the AMD Threadripper 1920X for Premiere Pro.

Unfortunately, the Threadripper 1950X 16 core has many of the same issues as the TR 1920X. It matches the Intel Core i9 7900X when rendering previews and is actually 3% faster overall when exporting but it has worse performance for both warp stabilize and live playback. More than anything else, it is the lower live playback performance that keeps the TR 1950X from being a better choice than the Core i9 7900X.

So is Threadripper a bad choice for a Premiere Pro workstation? If you just care about export performance, you probably wouldn't be able to tell the difference between a workstation using AMD Threadripper or one using Intel Skylake-X. However, the key area where Threadripper has issues is with live playback. It isn't really even that it is particularly bad, but Intel simply has a small lead in that area. Live playback is such an essential part of a video editing workstation that this alone keeps us from really recommending Threadripper for Premiere Pro. If you want to support AMD's efforts and help them continue to provide solid competition for Intel that is a completely valid reason to use Threadripper over Skylake-X, but otherwise Intel just has the stronger offering for Premiere Pro at this time.

Tags: Premiere Pro, Skylake-X, Threadripper
jacob m

Was looking through the extended results section and I think the titles for the bars have been reversed cause you're conclusion was that creator mode was 15-20% better but all the red bars were the longest and were labelled as game mode

Posted on 2017-09-01 00:43:59
Ready

What about the fact that the TR has more lanes to support 4-way SLI / four 16x PCIe? Or conversely, that we'll likely never see Thunderbolt supported on an AMD board... Decisions decisions...

Posted on 2017-09-01 20:49:13
Toby Hoffman

This is all well and good for those only using Premiere as a solo tool but in my editing work I am often using a combination of Premiere, Photoshop, After Effects and Media encoder all running at the same time using Adobe Dynamic Link and for this Threadripper will be of more use to me! TR 1920x all the way.

Posted on 2017-09-06 01:06:50
Theodosios

Good test, but not fair. No one buys threadripper with ddr2666 mhz. This is as if you had lowered all Intels from 4,5 Ghz to 4Ghz just to equal Threadripper. But you didn't. You underclocked only Threadripper's infinity fabric instead. Clever, but I don't buy it. Threadripper under 3200 mhz at least, ideally 3600, leaves all intels far behind. Pci's on Threadripper's platform are directly connected to the cpu, on Intel's platform to the south pole. Pci x8 and pci x16 have no difference in performance at all for multi vga card configurations. Last but not least, you are a seller and at the same time a tester. Since you sell only Intels, the manipulation is obvious.... Thats not fair...

Posted on 2017-09-13 18:01:20

Hey Theodosios, just a couple notes on your comment:

We tested with DDR4-2666 because that is what AMD has officially certified for these CPUs: https://www.amd.com/en/prod... . You certainly can use higher speed RAM (and many people who build their own PC will do so) but we consider stability to be one of our top priorities even above performance. Keep in mind that our customers are often not hardware enthusiasts - they are video editors, photographers, day traders, engineers, etc who simply need a workstation that will let them get their work done without having to worry about bluescreens or fiddling around in the BIOS. So for us, sticking to the officially supported memory speed is extremely important. Also, using higher frequency RAM that what is officially supported technically voids your CPU warranty which is something we as a business have to take seriously.

As for the PCI-E factor, both AMD and Intel CPUs have a number of lanes connected directly to the CPU. Threadripper has 64 lanes through the CPU for everything including GPUs, M.2, USB, etc. The Skylake-X CPUs in this article have either 28 or 44 lanes through the CPU, plus up to another 24 lanes (depending on the motherboard) through the chipset. Typically the CPU lanes are used for GPUs and other PCI-E devices while the chipset lanes are used for things like USB, M.2, etc.

As for us selling Intel, that is true at the moment but we definitely do not allow what we currently sell to influence our testing. In fact, it is exactly the opposite! The whole reason we do this kind of testing is to determine what products are truly the best choice for the software our customers use. For a number of years, Intel was simply the best choice for the workloads of our customers, so we haven't had a reason to carry AMD but we actually are planning on offering Threadripper very soon which we are pretty excited about. Not for Premiere Pro since it isn't the best choice, but we will be listing it at least on our Rendering workstations to start ( https://www.pugetsystems.co... ) and will be available for our consultants to sell whenever it makes sense. We just finished all our software testing about a week ago ( https://www.pugetsystems.co... )and are currently in the process of qualifying all the different components (motherboard, CPUs, coolers, etc.) to ensure that they meet our quality standards for things like stability, noise, etc. Once that is done, you will see it listed on our website for the applications where it is a good fit.

Posted on 2017-09-13 18:32:43

7820x 28 pcie lanes. Do the x299 mobos use pcie lanes for m. 2 drives like x99? In which case 1 graphics card 1 raid card and one nvme drive max it out

Posted on 2017-09-23 10:08:55

On every X299 motherboard I can think of, things like M.2 or U.2 NVMe drives are connected via PCI-E lanes on the chipset rather than the lanes on the CPU. X99 tended to be a mix since there are only 8 PCI-E 3.0 lanes on that chipset, but X299 can have up to 24 PCI-E 3.0 lanes on the chipset. Some of those lanes are needed for things like USB, audio, LAN, etc, but there are plenty left over for NVME drives. Having three PCI-E 3.0 x4 M.2 slots is relatively common on many ATX boards and most (if not all) all these slots use the chipset lanes.

Many boards also have at least one PCI-E 3.0 x4 slot as a chipset slot which is nice when you have something like a RAID controller. I don't think there is a major performance difference between CPU and chipset lanes, but that lets you reserve all the CPU lanes for video cards.

Posted on 2017-09-24 22:15:41