Puget Systems print logo

https://www.pugetsystems.com

Read this article at https://www.pugetsystems.com/guides/849
Article Thumbnail

Adobe Premiere Pro CC 2015.3 CPU Comparison

Written on September 6, 2016 by Matt Bach
Share:

Introduction

When configuring a workstation for Premiere Pro, there is a huge variety of components that you need to choose from. Everything from what CPU and video card to the number and speed of drives you need should be taken into consideration. In this article, we will be tackling the question of what CPU you need by looking at six different CPU options that you might consider in a workstation. These CPUs cover the highest end model from the standard Core i7 line, three of the "High End Desktop" Intel CPUs, and a pair of dual Xeon configurations with up to 28 physical cores.

In this article, we will be focusing on two primary tasks: exporting and rendering previews. If you want to skip over our individual benchmark results and simply view our conclusions, feel free to jump ahead to the conclusion section.

Test Setup

To benchmark Premiere Pro, we used the following configurations:

These test configurations include three different platforms along with six different CPU models. For Premiere Pro, we typically would recommend a CPU choice that has a moderate number of cores (between 6 and 12 cores) along with a fairly high frequency. However, to ensure that we are thorough in our testing we will also be including a Core i7 6700K (which only has four cores and is limited to 64GB max, but has extremely good single-threaded performance) and a high core count dual CPU setup. While our multi core performance testing has made it very clear that there likely is no benefit to using a dual CPU configuration with a high number of cores, this will let us show how much performance you may be giving up if you happen to already have a high core count dual CPU workstation or if you need/want a dual CPU workstation due to other applications you use in your workflow. 

To help with consistency - and since the benchmarks we performed ran for several days - we programmed a custom script using AutoIt to start Premiere Pro, load the relevant project, time how long it takes to export the timeline with the appropriate settings or generate previews, close Premiere Pro to clear any data from the system RAM, then loop while changing the project file.

The files we will be testing with came from a variety of sources:

1080P H.264/CineForm
4K H.264/CineForm
Provided by: Jerry Berg
Barnacules Nerdgasm - YouTube
ProRes 4K Grant Petty
Blackmagic Design Forum (available for public download)
4K RED RAW Provided by: Mike Pecci
Director & Photographer
6K RED RAW Neumann Films
RED Dragon Test Shot (available for public download)

In order to make our testing as accurate as possible, we used relatively simple timelines for our testing. In the past, we've loaded on the accelerated effects to show the maximum difference between configurations, but we found that this was not representative of typical real-world performance. Our test timelines consisted of:

  • 4-5 clips arranged in series to make a 60 second timeline
  • A basic transition applied between each clip
  • Lumetri color correction effect applied to each clip
  • Vector-based logo graphic added to the bottom corner of the footage

Exporting to 1080p

Premiere Pro 6950x 6900K 6850K Benchmark

Individual Export Results

While the individual results varyied from test to test, compared to the testing we have done for other applications like After Effects, there is a remarkable amount of consistency. Nearly across the board, the Intel Core i7 6950X was the top performer, with the only exception being "ProRes 4K > H.264 1080p" where the dual Xeon E5-2643 V3 was about 7% faster. However, for every other test the i7 6950X was anywhere from the same performance as the dual E5-2643 V4 to almost 25% faster. 

Overall, we would say that there is little reason to us a dual Xeon configuration for exporting to 1080p in Premiere Pro. If you only work with ProRes footage there may be an argument for it, but since PP 2015 is the first version with native Quicktime support there is no way to know if this is simply a fluke or something that will be true in future versions of Premiere as well. 

Overall, it would appear that using a "High End" Core i7 CPU makes the most sense for exporting to 1080p in Premiere Pro. There is certainly diminishing returns - where the i7 6850K is ~25% faster than the i7 6700K, the i7 6900K is only ~15% faster than the i7 6850K, and the i7 6950X is only ~4% faster than the i7 6900k. So while the i7 6950X is the overall best for this round of testing, we would say that the i7 6900K is probably the best value.

Exporting to 4K

Premiere Pro CPU benchmark 4K

Individual Export Results



 

We always like to see consistent results, so we were very pleased with what we saw when exporting to H.264 4K. Compared to our 1080p testing, the exact results changed a bit (the higher core count CPUs performed a bit better) but it is good to see that the best overall CPU for exporting to 1080p is the same as the best CPU for exporting to 4K.

As we noted, however, there are some differences - although those differences all mean that a more expensive CPU will actually give you an even greater increase in performance. While the i7 6850K was still roughly 25% faster than the i7 6700K, the i7 6900K is now about 20% faster than the i7 6850K (it was only 15% faster when exporting to 1080p). In addition, the i7 6950X is a respectable 10% faster than the i7 6900K instead of the meager 4% we saw in the last section.

While higher core count CPUs did better in this round of testing, there still is little use to a dual Xeon configuration. The dual Xeon E5-2690 V4 was about 10% faster when exporting RED 4K to H.264 4K, but that is the only time a dual CPU configuration was faster than a single CPU. And since you would be paying between two and three times the cost for a E5-2690 V4 system that is a bit better for one thing, but much worse for everything else, that is likely not a good option for most users - even if they primarily work with RED 4K footage.

Render Previews

Premiere Pro Preview Benchmark

Individual Rendering Results

Our final round of testing - rendering previews - is not only a good test for previews, but also an indication of how well each CPU is able to scrub the timeline live. We still have not found a reliable way to test scrubbing directly (recording the amount of dropped frames is widely inconsistent between runs), but this is still a decent way to measure the performance difference when scrubbing in addition to how fast each CPU can render previews.

While the end results are pretty similar to what we saw when exporting to 1080p or 4K, the higher end Core i7 CPUs actually saw slightly smaller amounts of performance gains in this testing than they did when rendering. There is still a decent increase using the i7 6850K compared to the i7 6700K (about 17%), but we only saw an additional 15% when upgrading to a i7 6900K, and only another 3.5% when upgrading to a i7 6950X.

Similar to the rest of our testing, the dual CPU configurations were overall much worse than the best single CPU option, although there was still a couple of times that they were faster. This time, the dual Xeon E5-2643 V4 was a bit faster when generating previews for RED 4K and RED 6K footage, but only by 3-4%.

Conclusion

It is fairly rare that the results of our testing is consistent across the different tasks we benchmark, but Premiere Pro appears to be one of those programs where it thankfully is the case. In fact, the results are close enough that we are comfortable averaging all the results together to give you a "overall performance" chart that we feel should be relatively accurate:

Premiere Pro Corei7 Xeon Benchmark

If you only work with one type of footage that we happened to test, we highly recommend looking at that specific testing result to determine which CPU would be best for you. However, based on our testing overall, we can make a general recommendation for what CPU you should use in a Premiere Pro workstation:

  Recommended CPUs for Premiere Pro CC 2015.3
Entry: Intel Core i7 6850K 3.6GHz (3.7-4.0GHz Turbo) Six Core
Best Value: Intel Core i7 6900K 3.2GHz (3.5-4.GHz Turbo) Eight Core
7-24% faster than i7 6850K (16% average)
Best Performance: Intel Core i7 6950X 3.0GHz (3.4-4.0GHz Turbo) Ten Core
0-15% faster than i7 6900K (5% average)

For an entry level Premiere Pro workstation, we would recommend using an Intel Core i7 6850K. It gives great performance when you are on a bit of a budget, but doesn't sacrifice a huge amount of performance compared to the higher end options (only 16-22% on average). If your budget allows, the Intel Core i7 6900K is an excellent value giving you up to 24% higher performance compared to the i7 6850K depending on the specific task. This difference in performance is most obvious when exporting to higher resolutions like 4K, however, so if you only work with 1080p footage the difference in performance will be closer to ~10% rather than 24%. For the best possible performance, the Intel Core i7 6950X is an excellent choice. This CPU was the fastest (or a very close second) in nearly all of our tests and while it is only 5% faster on average compared to the i7 6900K, if you need the best, this is it.

You will notice that we didn't even include a dual CPU configuration in our list of recommended CPUs. Frankly, Premiere Pro simply is not capable of taking advantage of the larger number of cores available in a dual CPU configuration. At best, they were slightly faster than the single CPU options, but mostly they performed either the same or slower. Some of that is likely due to Adobe's recent focus on GPU acceleration (which should yield much higher performance gains for the majority of users), but whatever the reason, dual CPUs simply don't make sense for the majority of users.

One last thing we will note is that these results are only completely accurate for Premiere Pro 2015.3. We've seen fairly wide shifts in what CPU is best for Premiere in the past, so if you are still using CS6 or a different older version, this testing may not be be representative of what you would see.

Tags: Adobe, Premiere Pro, CPU, Processor
Ray Gralak

If Premiere Pro is not NUMA aware (Non-uniform memory access), using cores across dual CPUs can considerably slow down an application. So, have you considered trying a single 2690 V4 to see how performance stacks up? The 2690 V4 can spin all 14 cores up to 3.2 GHz, which should give it the equivalent performance of a 4.48 GHz 10-core CPU. I know that's not the only factor (memory and disk performance need to be balanced too), but it would be interesting to see if NUMA might be hurting the dual Xeon performance in this case. Also, it's also important to have adequate cooling on the Xeons or they might not reach full rated speed across all cores. It may be worthwhile to confirm that the CPU cooler you are using is keeping the core temps on the Xeon low enough.

Posted on 2016-09-08 22:47:13

As a part of our multi-core testing article (https://www.pugetsystems.co... we actually tested with NUMA both on and off. We didn't bring it up in that article since most of our readers don't know what NUMA is, but we didn't see a major difference in performance so it may be that Premiere isn't NUMA capable. As for the E5-2690 V4, that is a good idea - there may be a performance gain over the Core i7 6950X. The E5-2690 V4 does have the same all core Turbo as the i7 6950X along with four more cores, but the i7 has Turbo Boost 3.0 capability which allows one of the cores to run at 4GHz. If I get a chance, I will try to test the E5-2690 V4 but I would guess it to be pretty similar or a hair slower than the i7 6950X due to that Turbo Boost 3.0 capability. Not sure when I will have the chance to do so, but it would be good to confirm my guess.

As for cooling, that shouldn't be a problem with our systems. All of our Premiere Pro workstations use the Noctua NH-U12DX i4 (https://www.pugetsystems.co... and the system cooling is designed to keep the CPU under ~80C even under extreme loads. Last time I happened to check, I believe I was seeing about 55C on the CPUs when running Premiere so cooling shouldn't be a problem at all.

Edit: OK, you made me curious. I'm going to get some testing going on the E5-2690 V4 - hopefully I will have results ready on Monday that I can pass on.

Posted on 2016-09-08 23:40:59
Ray Gralak

OK, great! I look forward to seeing the results. If it wouldn't be too much trouble I think the best apples to apples comparison might be obtained if you installed the 2690V4 in the same type of X99 motherboard that you used for the 6950x (in case the Xeon chipset might somehow affect the results).

Posted on 2016-09-09 16:25:03

Hey Ray, I got the results for the E5-2690 V4 (I did test in the same X99 system as I used for the other Haswell-E CPUs). Real quick though, I was wrong that it has the same all-core Turbo as the 6950X. The 2690 V4 is 3.2GHz with all cores, the 6950X is 3.4GHz.

For the results, it was around 8% slower than the 6950X when exporting to 1080p, which makes it a bit slower than even the 6900K actually. Exporting to 4K was actually pretty good, beating the 6950X by about 4% on average. Generating previews, however, was about 3% slower than the 6950X - so most of the gains when exporting to 4K are lost when generating previews. So on average, the E5-2690 V4 is 150.7% faster than a Core i7 6700K while the Core i7 6950X is 152.6% faster. That's not a big difference, but considering the 6950X is cheaper and has a much higher single core frequency (which makes it faster for all the random things you do in Premiere and most other programs), there isn't much reason to use a E5-2690 V4 instead.

The one time you might consider it is if you work exclusively with RED 4K footage and only export to 4K. For that combination, the E5-2690 V4 was faster for both exporting and rendering previews. If you export to 1080p or work with almost any other footage type we tested, however, the performance is either extremely similar to the 6950X or slower either rendering previews or exporting drops so the 6950X is just a better choice overall.

Hopefully that answers your question. In pretty much every case, the single E5-2690 V4 was faster than dual Xeon E5-2690 V4, but it just isn't good enough to beat the Core i7 6950X.

Posted on 2016-09-12 16:49:31
Ray Gralak

Hi Matt,

Sorry for the delayed reply. I wonder if the faster systems in your test are hitting a wall, much like using a high-end video card at too low of a resolution, You can't really measure the performance of that video card until the screen resolution is increased. Likewise, something else in the system might be preventing the many cores of a CPU from being used. One possibility is the Xeon was throttling back because it is hitting its thermal limit, but I think you assured me that wasn't happening.

Another possibility, which I mentioned above, is that memory and disk performance are not balanced. It could be that the limiting factor is memory bandwidth. If the Adobe Premiere is memory bandwidth limited during parts of your testing then adding more cores won't improve performance, so the Xeon might not be able to attain its full potential. Theoretically, a well designed NUMA-aware application should have roughly double the memory bandwidth as an i7-based system. I use some applications that scale very well with dual Xeons, so I know it is possible.

Likewise, the disk drive could also be a limiting factor. Most of the high-end Adobe Premiere setups I have seen have large RAID arrays instead of a single SSD. In fact the highest performance systems have separate SSD or hard drive RAID arrays for media, scratch drives, and output drives. To maximize performance parallel disk paths usually are used.

So, using a single SSD, as you were doing, might have been the limiting factor in some of the tests. Since it is only a SATA drive, all disk I/O had to be serialized through that single SATA interface. That might put the 14-core Xeon at more of a disadvantage if all cores are thrashing, trying to read/write at the same time on the same drive. A PCie 4x NVME drive would allow up to 4 simultaneous threads. Of course, one of those drives each for source, scratch, and output, would be ideal!

-Ray

Posted on 2016-10-04 12:49:19
Ryan Thompson

Thank you so much for making a detailed white paper for Video Production. We NEED and love you Puget Systems :*

Posted on 2016-09-29 02:41:07
gabemcg

Have you had a chance to test the 6800K? Priced right between the 6700K and 6850K I wonder if the performance is in-line?

Posted on 2016-11-07 02:43:58
matinciel

I would be glad to know that also ;-)

Posted on 2016-11-16 17:25:04
godisafairytale

I was about to ask the same thing! Very curious about 6700k performance.

Posted on 2016-12-05 14:46:30
TomChief

Have you guys done any testing at all, or plan to, with DaVinci Resolve?

If so, do you have any back of a stamp style advice for building a DaVinci Resolve system?

Thanks,

Tom

Posted on 2016-11-07 12:45:30

We have not yet tested Resolve, but I believe it is on the list of applications our Labs folks want to tackle in the not-too-distant future.

Posted on 2016-11-07 19:18:30
Lightprism

There are a lot of Resolve users that would be extremely interested in your testing of the dual Xeon vs i76950. Lots of confused hardware decisions being debated over on the official Blackmagic Resolve user forum. The current August 2016 Configuration Guide still recommends Dual Xeon's. Blackmagic has not stated what the basis of this configuration guide was - don't know if any actual benchmarking of then current CPU/motherboard was performed.
Because a lot of folks using Resolve were on the Mac and Apple hardware related to the needs of the video industry have lagged, many folks are in the process of moving to Windows based workstations. Big unanswered questions on how Resolve uses CPU frequency speed versus number of cores/multi-threading. Current "guess" is i7 8-10 core single CPU for short 1080 to 4k projects ... dual Xeon for long high resolution projects - mostly to get more PCIe slots/lanes ... and Resolve encoding seems to use the cores for encoding. And Resolve definitely uses GPU so if at all possible use at least 2 good nVidia GPU's such as the GTX1080 or even Titan Pascal in your testing if at all possible.

There will be lots of very interested Resolve users. And the Resolve base is growing since it's become a pretty good NLE as well as the great grading tool it has long been. But it does require some solid hardware to run efficiently.

Posted on 2017-01-17 15:04:16

I believe we will be doing more in-depth Resolve testing later this year, and input like you provided is greatly appreciated for that! We did do a *little* testing, specifically looking at PCI-Express scaling (x8 vs x16) with single and dual video cards. Interestingly, you can also compare the results on the two graphs to eachother and see that the benefit to having two Titan X cards vs one ranged from about 20% to a full 100% (doubling of performance). Here is a link to that section of the article:

https://www.pugetsystems.co...

Posted on 2017-01-17 18:55:17
Lightprism

Thank you William for the Resolve info. Please let folks on the Resolve Forum know when you've done your benchmarking as there are a ton of us out there hungry for better testing. Resolve is a very hardware intensive piece of software. And the only widespread benchmarking "Candle Test" is really setup for GPU decision making.
Good benchmarking like you folks perform is going to be tough as the range of work done by Resolve Users requires totally different hardware. Examples: Short 30 sec spot commercials/1080p/heavy effects laden projects require fastest CPU ... versus the feature length 2 hr narrative/documentary with voluminous terabytes of 4k footage require a lot more external raid storage and thus usually more PCIe slots than an X99 motherboard that handles the fastes Core i7 CPU's - those folks are stuck with slower dual Xeon motherboards to get the needed PCIe slot real estate for multiple nVidia GPU cards AND raid controllers.

So you folks have your job cut out for you on doing the quality testing you are known for. Even the maker/owner of Resolve has so far not invested the resources into testing the various user scenario needs and published a too broad configuration that has been outdated understandably. And they still have a large user base who is jugging along with the 4 year old and limited from the start Apple trashcan systems and prior MacPro's. But even the switchover to Windows has built steam over the last year or so and thus the need for some good fuller testing like Puget performs. So thank you again for that commitment.

Posted on 2017-02-15 23:03:43

I found this site and excellent reviews and comments very timely, am in process of upgrading my 2 yr old system running PPro CC 2017.
After days of researching other sites and Vids on You Tube, this is the Mother Lode and font of all useful knowledge. I spent 15 years as a Telecine operator (Bosch FDL60) and grader in the old analogue days. Loving the Digital domain, been with Premiere Pro since V4.2! Who can remember a Firewire capture card costing a fortune?
Still interested to see article as final word on optimum RAM, 16/32 g?
Cheers to all the experts here!

Posted on 2016-12-14 11:02:45
Jeff Stubbers

Hi Sergei, the RAM and other specification recommendations can be found at:
https://www.pugetsystems.co...

Posted on 2016-12-14 14:35:00

Hi Jeff,
that is useful, but a more in depth comparative study on amount versus performance gains, also type and brand of RAM, ram frequency ratio to CPU frequency would be also useful. An area that has been largely skimmed over on most test sites. Thanks again.

Posted on 2016-12-14 17:01:33

We tested different speeds of memory a while back, and found that outside of a few niche applications the performance impact of different speeds / latencies of memory was minimal. As such, we prefer to focus on memory that matches Intel's recommendations for their CPUs and is from brands which we have found to be the most reliable over the years. The factor that then affects performance most is the *amount* of memory, which is something we do test in certain applications - like what Jeff noted.

Posted on 2016-12-14 18:25:38

Thanks William,
that's a good summation, maybe I overlooked it, but it seemed absent from your referenced article.
If I understood you correctly, I should not waste my money by ordering 3200ghtz RAM instead of 2400g.

Posted on 2016-12-15 06:16:33

That is definitely our stance. The last time we tested memory speeds was a couple of years ago now, so it is *possible* that things have changed - but the basic results from the last two such tests we did were pretty much the same. A couple of really niche applications (usually things related to data compression / decompression and encryption) benefited somewhat, but most other programs saw at most a % or two difference. Given the added cost and lower reliability that we have measured on "high performance" (higher speed) memory, we recommend sticking with what Intel rates their CPUs for. Currently that is 2133MHz for the Xeon E3 and Core i3 / i5 / i7 (up to 6700K), and 2400MHz for the i7 6800K and above as well as the Xeon E5 series.

Posted on 2016-12-15 17:31:14
siposh

Thank you, Puget Systems and Matt for these useful tests, it's one of the only site where you actually look at so many angles of performance.
I have one question though regarding UI and day-to-day use of Premiere.
I see much slower response time and general "snappiness" on my Dual Xeon 2670 (with 980Ti, OS on NVMe, Cache on SSD, footage on HDD) than just recently bought sister's PC with i5 7600K Kaby Lake. I work on feature length documentaries and I would much more appreciate general response time in Premiere than a few minutes saved at exports. Does it really depend so much on CPU clock speed for the UI?

I mostly worth with XAVC-I from FS7 and XAVC-S from A7sII. If I transcode to DNxHR it renders frames faster, but still far slower than on i5 Kaby Lake PC (looking at the render time in CTRl+SHIFT+F11 cuda render system info)). Also, playback gets totally laggy if I use overlaying layers; eg. an interview on layer 1 and b-roll on layer 2 without actually disabling clips in the layer 1.

So, all in all, how much difference in very complex sequences and UI reponse time have you seen among different builds, from lower clock xeons to high clock i7s? I generally want to scrub through foorage on very complex sequences without dropped frames.

I know Premiere 2015 had some issues with XAVC-I format, but 2017 seems to have resolved that. Now I really want to edit fast cuts withoutalways rendering the timeline.

Thanks a lot for anyone chipping in with some experience.

Posted on 2017-02-18 09:57:43

When you talk about general response time, do you mostly mean live playback performance without previews? Unfortunately, for that kind of thing (and most other things in Premiere Pro) your dual Xeon is not really ideal at the moment since Premiere doesn't scale well enough to take advantage of that many cores. It used to, but with all the work being put into GPU acceleration (among other things) it simply isn't the optimal configuration anymore. In fact, I'm not terribly surprised that the i5-7600K is outperforming your system in many areas since the 2670 (v3 Im guessing) have pretty low operating frequencies.

In general, a single CPU with around 8-10 cores and a relatively high frequency (like the i7-6900K or i7-6950X) should give you the best all-around performance in Premiere. This should apply for both exporting and things like live playback. There are a few times that having more cores will help with export times, but like you are seeing you have to give up quite a bit in other areas.

If you use any GPU accelerated effects (Lumetri color correction being a big one), upgrading to the newer Pascal GPUs (like the GTX 1080) would probably also help you with performance. But if there was one thing to change on your system it would be to move to a single 8-10 core CPU. That's not an easy change since it will require both a new CPU and motherboard (and a re-installation of the OS) but it is what it is.

Posted on 2017-02-20 18:42:59
siposh

Yes, live playback without rendering previews, as well as zooming in and out of seqeunces, drag&drop of files from one sequence to another, rearranging sequence tabs ...
Thank you for your response! I really appreciate it! I hope I can ask some more questions in the future. I will probably try to migrate to a new PC in a while (but still coworkers could use the 2x2670v3 for other tasks); I don't know if maybe waiting for Intel's Coffee Lake would be a good idea since Ryzen is coming and Intel will have some competition. Also, it's a pity that my ZOTAC 980Ti is already outdated.

Posted on 2017-02-21 20:59:03

As many of you know, Cinegy Cinescore is the closest thing we have to a standardized test. Here are my results, putting my system up against those widely touted Xeons with surprising results: https://uploads.disquscdn.c...

Posted on 2017-02-23 06:38:31

Interesting, but not sure how many people could be expected to know about a benchmark that is less than a month old:

https://www.cinegy.com/inde... (dated Feb 3rd, 2017)

Posted on 2017-02-23 06:41:04

By standardized I don't mean universally adopted, but rather, finally a test that respects the most critical factor when it comes to comparing real-world rendering speeds: identical source footage. No test has employed that strategy until this one.

Posted on 2017-02-23 06:43:46