Puget Systems print logo

https://www.pugetsystems.com

Read this article at https://www.pugetsystems.com/guides/1528
Article Thumbnail

Does RAM speed affect video editing performance?

Written on July 22, 2019 by Matt Bach
Share:

Introduction

In our Puget Systems workstations, we typically stick to the RAM speed that is officially supported by a processor since it generally gives the best mix of both performance and stability. However, just like with more traditional CPU overclocking, there are certainly times that an application can benefit from having higher speed memory. This is not always going to be the case, but the difficult part is knowing whether the applications you use every day are ones that will benefit or not.

Today, we will be looking at a number of applications used by video editors to discover whether or not higher speed RAM gives a tangible bump in performance.

Does RAM speed affect video editing performance?

While we are going to primarily be looking at pure performance in this article, keep in mind that stability is a significant concern for any workstation. Getting an extra X% performance gain is always nice, but if it causes even minor system stabilities, that increase in performance can easily be wiped out with a single program crash or bluescreen.

However, whether or not higher frequency RAM is worth the potential stability tradeoff really depends on exactly how much the performance gain is and how much risk you are willing to take. Our testing and articles are primarily geared towards our customers who are overwhelmingly not the kind of people who want to tinker with their system - but we understand that our testing is also widely used as a resource for people who build their own computers.

To help answer whether higher frequency RAM is worth it, we will be looking at a range of applications including Premiere Pro, After Effects, Neat Video, and Photoshop with a number of CPUs and RAM speeds ranging from DDR4-2666 to DDR4-3600. In addition, we will also include one set of testing with 128GB of RAM since we have been curious if the newer 32GB RAM modules are any slower than the more standard 16GB modules.

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 workstations that are tailor-made for your unique workflow. Our goal is to provide the most effective and reliable system possible so you can concentrate on your work and not worry about your computer.

Configure a System!

Test Setup & Methodology

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

AMD Ryzen Test Platform
CPU AMD Ryzen 9 3900X
Official RAM Support

2x DIMM: DDR4-3200
4x single rank DIMM: DDR4-2933
4x dual rank DIMM: DDR4-2667

CPU Cooler AMD Wraith PRISM
Motherboard Gigabyte X570 AORUS ULTRA
Intel 9th Gen Test Platform
CPU Intel Core i9 9900K
Official RAM Support DDR4-2666
CPU Cooler Noctua NH-U12S
Motherboard Gigabyte Z390 Designare
AMD Threadripper Test Platform
CPU AMD TR 2990WX (DLM on)
Official RAM Support

Single Rank:
1DPC, 4x4: DDR4-2933
1DPC, 4x8: DDR4-2666
2DPC, 8x8: DDR4-2133

Dual Rank:
1DPC, 4x4: DDR4-2933
1DPC, 4x8: DDR4-2666
2DPC, 8x8: DDR4-1866

CPU Cooler Corsair Hydro Series H80i v2
Motherboard Gigabyte X399 AORUS Xtreme
Intel X-series Test Platform
CPU Intel Core i9 9980XE
Official RAM Support DDR4-2666
CPU Cooler Noctua NH-U12DX i4
Motherboard Gigabyte X299 Designare EX

*All the latest drivers, OS updates, BIOS, and firmware applied as of July 2nd, 2019

Note that most of our testing will be done with 4 sticks of dual rank RAM (higher capacity modules are almost always dual rank). This means that for each platform - both Intel and AMD - we will technically be overclocking once we get above DDR4-2666. In some cases, like Threadripper with 8 sticks, we would technically be overclocking with anything higher than DDR4-1866. However, since we don't have the time or resources to test every possible combination of RAM count/frequency/timing, we are going to make things easier on us and just stick with DDR4-2666 as the bottom end of our testing.

Something else to make note of is that we are only using 4x8GB for the DDR4-3600 testing. There are 16GB DDR4-3600 modules available from few manufacturers, but Crucial was able to generously provide the RAM for this testing and they currently only offer 8GB modules at this speed. This shouldn't affect performance since our benchmarks don't need more than 32GB of RAM, but there is the possibility that higher capacity DDR4-3600 RAM may show higher performance in our benchmarks.

The benchmarks we will be using are the latest release of our public benchmarks, plus NeatBench 5. Full details on the benchmarks and links to download and run them yourself are available at:

Raw Benchmark Results

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 our overall scores. Feel free to skip to the next sections for our analysis of these results if you rather get a wider view of how each type of RAM performs.

Premiere Pro Benchmark Results

Starting off with Premiere Pro, the results are quite different than what we expected. Oddly enough, we actually saw slightly worse performance with the higher frequency RAM when using the Intel Core i9 9900K and Core i9 9980XE. The 9900K in particular saw up to a 7% performance drop with DDR4-3000 RAM, but recovered a bit with 3200 and 3600 RAM.

This is a very strange result that we honestly don't have a good explanation for. We ran our tests several times to make sure it wasn't an issue with a single run, but the results were consistent - DDR4-2666 actually was consistently slightly faster in Premiere Pro on Intel CPUs. However, one thing to keep in mind is that the current version of our Premiere Pro benchmark is still fairly new and we are not quite sure yet what the standard deviation is. There are some tests like 8K H.264 live playback where the performance appears to vary quite widely, while others are very consistent. This is part of the difficulty of testing real-world performance: it rarely is as smooth as a synthetic benchmark.

With AMD, the results are a bit more as we expected. On Threadripper, the RAM speed didn't affect performance much, but the new Ryzen 3900X did see a small gain in performance as we increased the speed of the RAM. However, it was only by about 4% going from DDR4-2666 CL16 to DDR4-3600 CL16.

After Effects Benchmark Results

Our After Effects benchmark tends to have a margin of error right around 2%, and there were only a few isolated cases where we saw a performance delta larger than that. While you could cherry-pick individual results from the tests, overall it appears that RAM speed does not have a significant impact on performance in After Effects regardless of the CPU you use.

Something we do want to point out here is that in After Effects, the amount of RAM you have can make a huge difference since more RAM means you can have more frames cached in RAM Preview (which means less frames that have to be re-rendered). Our benchmark, however, is designed so that people with 16GB of RAM can successfully run it. So, while you want as much RAM as possible if you use After Effects, this is not reflected in our benchmark which primarily looks at render performance during RAM Preview and Exporting.

Neat Video NeatBench 5 Benchmark Results

NeatBench 5 (which is the stand-alone benchmark for the popular noise reduction plugin Neat Video) easily saw the largest performance gains with higher frequency RAM out of all the applications we tested. The Core i9 9980XE still didn't see a benefit, but the Threadripper 2990WX did see a small ~4% gain when we got up to DDR4-3200 and DDR4-3600 RAM.

The real winners here were the Intel Core i9 9900K and AMD Ryzen 9 3900X. Increasing in RAM speed from DDR4-2666 to DDR4-3000, both of these CPUs saw a ~4% gain in performance. Going up to DDR4-3200 and DDR4-3600, however, net about a 7% and 12% gain over DDR4-2666 respectively.

This is a pretty significant increase, although it is interesting that both the Core i9 9900K and Ryzen 9 3900X saw almost identical boost to performance with higher speed RAM.

Photoshop Benchmark Results

Photoshop may not be a true video editing application (unless you are one of those people that uses its rarely used video timeline features), but it is commonly used alongside Premiere Pro, After Effects, and similar apps so we decided to throw it into the mix.

While not quite to the same degree as Neat Video, Photoshop can definitely see some benefit from higher frequency RAM with select CPUs. The Intel Core i9 9980XE didn't see much of an increase in performance, but the AMD Threadripper 2990WX and Intel Core i9 9900K saw up to a 5% performance gain going from DDR4-2666 to DDR4-3600. The AMD Ryzen 3900X saw a slightly larger increase, with performance gains up to a 7.5% with DDR4-3600 RAM.

Does RAM speed affect video editing performance?

In our testing, we found that whether higher frequency RAM improves performance or not depends heavily on both the application and the CPU used. Some (Neat Video) showed a fairly substantial benefit going up to DDR4-3600 with both the Core i9 9900K and Ryzen 9 3900X processors. Others (Photoshop) also saw a benefit, but not to as significant of a degree. At the same time, both Premiere Pro and After Effects really didn't see a significant performance advantage with higher frequency RAM, regardless of the CPU.

While we primarily want to focus on pure performance in this article, a key consideration is also overall system stability and reliability. Using RAM that is beyond what is officially supported by your CPU is technically overclocking and carries many of the same risks as more "traditional" CPU overclocking.

This absolutely does not mean that higher frequency RAM will always be less stable, but it is going to depend highly on the quality of your RAM, motherboard, power supply, and other components in your system. However, even with some of the highest quality parts you can currently purchase (Crucial RAM, Gigabyte motherboard, EVGA PSU, etc.), we did see a noticeable increase in the number of times our benchmarks crashed as we used higher and higher frequency RAM. It is always hard to know what is to blame (our benchmark itself, early platform bugs, a single weak component, etc.), which is why we did not want to focus on the details of stability in this post.

Overall, our recommendation for most users is to stick with the RAM speed that is officially supported by your CPU in order to maximize the stability of your system. If you are looking to get every ounce of performance, however, there are some applications (Photoshop and NeatBench from what we tested) that can to potentially get up to ~13% higher performance with DDR4-3600 RAM. Just keep in mind that this performance is definitely not "free" - just like CPU overclocking, it is possible that it may cause more problems than the extra performance will solve and may require a bit of tinkering in the BIOS to get it stable over the life of your system.

Looking for a Video Editing Workstation?

Puget Systems offers a range of workstations that are tailor-made for your unique workflow. Our goal is to provide the most effective and reliable system possible so you can concentrate on your work and not worry about your computer.

Configure a System!

Tags: 3900X, 9900K, 2990WX, 9980XE, RAM, Memory, DDR4-2666, DDR4-3000, DDR4-3200, DDR4-3600
Batt Mach

I find it interesting that ram speed can give more performance than ram capacity. I might take that into consideration.

Posted on 2019-07-22 22:10:16

Hey evil twin! I would say that in most cases, having enough RAM is much more important than the speed of the RAM. The problem with showing that in benchmarks is that if you don't have enough RAM, that often just causes the program to crash. Or in the case of something like After Effects, having more RAM means more frames can be stored in RAM Preview, which means the system has to re-render frames less often.

Our benchmarks tend to be made so they will run on systems with either 16GB or 32GB of RAM (depending on the benchmark), so in terms of performance, having more than that won't affect anything. A good way to think of it is like traditional storage. If you have a 2TB drive with only 500GB of data on it, you probably would rather have a super fast 1TB drive than a slower 2TB drive. But if you have 1.5TB of data, it doesn't matter how fast a 1TB drive is since you simply can't fit everything you need onto that drive.

Posted on 2019-07-22 22:14:52
yezhacker

So, the sweet spot for the Ryzen platforms, both X399 and X570, is the 3200 frequency. This showed up that Ryzens are way more "sensitive" to higher frequencies than Intels. Maybe I'm wrong but on the Z390 platform, the 3000 frequency gives the worst performance among all.
"More RAM vs higher frequency? It depends..." as everything on life XD

By the way, I notice an issue with your tables: when I make them bigger I barely see what's written. I had the same issue with the tables of your latest AE benchmarks last Friday. I'm using Android, but don't have the same problem with older articles. Hope you can check that, and thanks for sharing this!

Posted on 2019-07-22 22:55:34

At least with the applications we tested, I agree that 3200 is probably a good sweet spot. There are definitely cases where even higher will be faster (Neat Video is one example, and I know AMD has published charts with others), but strictly talking about video editing apps, that definitely seems to be the case from what we saw. Even then, however, the gain on average was only ~2% over DDR4-3000, which is what AMD supports if you want to use four sticks (ignoring the single vs dual rank nonsense). I know we are trying to figure out what RAM speed to use on the Ryzen systems we will be offering, but I honestly am not sure what we are going to end up deciding. I think we are leaning towards 3000MHz as a middle ground between stability and performance, but a lot of it is just taking a guess at the risk vs reward.

I know, our tables are kind of getting out of hand. I've been thinking about just not including them at all since people really just care about the overall scores... but I figure "why not?". They should be really high resolution, though, so you just need to zoom in. Still a huge pain to actually find specific results, but we do have a plan we are hoping to implement in the next 6 months or so that will really make things easier.

Posted on 2019-07-22 23:03:59
yezhacker

No, I didn't mean "eww, these tables are a mess of data I don't care." I really enjoy reading your tables, what I mean is precisely their resolution, when I zoom them in, everything appears pixelated. That only happens with your last articles!

Posted on 2019-07-22 23:10:31

Oh, I think I know - anything that is a "image gallery", if you click on it, opens up another image gallery... Doesn't make sense, but it is a little holdover from when we implemented the new gallery system. We are getting that fixed in the next week or so I hope.

Posted on 2019-07-22 23:17:34
David Varela

Heya Matt, in that sweet spot of 3200Mhz but in terms of latency: 3200 CL14 is better than 3200CL16 but does RAM latency has any true benefits or noticiable impact in Davinci Resolve? Thanks

Posted on 2019-12-12 14:42:00
Randy

I did notice that AMD lists 3200MHz for the new Ryzens.

Posted on 2019-07-23 18:14:11

Yeah, if you are using just two memory modules (regardless of whether they are single- or dual-rank) AMD's official supported memory speed is 3200MHz. It drops a bit if you use four modules, though :/

See: https://www.tomshardware.co... (about half-way down the page)

Posted on 2019-07-23 18:32:53
Randy Fellmy

I'm curious... I know the rated memory speed for the i9-9900K is 2666MHz, so just looking at that... why would the results for 64GB of DDR4-2666 CL16 be worse than or equal to the results for the same quantity and clock with CL19 modules? That seems completely counter-intuitive. Is there something about the i9-9900K (or the Z390 Designare) that gets tripped up by faster timings? Is some sort of thermal throttling skewing the results?

Posted on 2019-07-23 06:20:56

I think it is just margin of error. Real-world testing like this is going to have a lot more variance between runs than something like a straight bandwidth benchmark, so unless we ran each test a dozen times and took the average (which would take months), we aren't going to get perfect results.

Generally, I consider anything within 2% of each other to be the same, or at least close enough that it doesn't matter. 5% is where you can start to be confident that at least you know that one is generally faster than the other, and 10% is where you would start to notice it in everyday workloads.

Posted on 2019-07-23 17:12:18
Jig Serencio Navasquez

Hi Matt! What would be the difference from RAM 2666 compared to 2400?(for AE ryzen) because Im currently using 2400 32gb, Im thinking upgrading it to 64gb 2400 so that I dont need to buy 4sticks of higher speed. thanks!

Posted on 2019-07-26 00:50:00

We didn't test that low, but given that we saw only a minor difference in After Effects with the higher RAM speed, I suspect it will only be a percent or two at most. Nothing you would ever notice in the real world, so I wouldn't worry about upgrading to 2666. Simply upgrading to 64GB is going to be a way more significant upgrade.

Posted on 2019-07-29 02:21:51
Serj Kasparoff

Oh, that's funny : )
Exact the same question about RAM speed difference for video editors was in my attention for the whole previous week! And I was hoping to find the answer here, but no luck at that moment... and some days after you posted it : ))
Thank you, as always great info!
But what about DaVinci Resolve? Why did you decide not to include it in the test?
For sure, DaVinci traditionally is more for color grading, but the year after year we see it aims to be the only software package for video post-production. And as we know DaVinci and Premiere differently use computer resources... So it stays unclear to me - could it be that for Resolve RAM speed is more important than for Premiere?

Posted on 2019-07-27 21:34:46

Including Resolve would have almost doubled the testing time, so we kept it out (for now). All our benchmarks are automated, but for Resolve we have to switch the GPU up to Titan RTX 24GB cards in order to get enough VRAM for our 8K tests. That would mean another whole round of swapping our RAM on each platform which was honestly a pain since we often had to clear the CMOS in order for the systems to boot with the different RAM.

That said, I did do some spot checks and it appeared to be pretty close to our Premiere Pro and After Effects results - only a very minor difference in performance with higher frequency RAM.

Posted on 2019-07-29 02:25:27
Serj Kasparoff

"Including Resolve would have almost doubled the testing time" - I guessed that, maybe it's not worthy : )
"a very minor difference in performance with higher frequency RAM" - it's a general answer for me, thank you once again!

Posted on 2019-07-29 07:47:05
Dan McDermott

It would be nice to see more info on why Puget only offers Crucial DDR4-2666 RAM in their computer build drop down menus (for 16GB & 32GB). The Ryzen 7 3800X CPU that you offer allows much higher RAM speeds. I have heard it is because the 2666 offers better stability but I can't find anyone else but Puget who has run into that issue. Now I know this article was about Threadripper, but what about Ryzen 7 CPUs?

Posted on 2019-11-30 16:42:19

You are right that it is about stability and reliability. Those are huge factors for us since it is our firm belief that a small gain in performance is not worth a small increase in the chance that the system or application may crash. It is the same reason we don't do overclocking to be honest. I think part of the reason you haven't heard much about the issues with higher frequency RAM is because there are very few system builders like us (especially in the workstation market) who are as transparent as we are. Take Boxx for example: they are more willing to push things a bit than we are since they offer CPU overclocking, yet their Ryzen systems all use DDR4-2666. If they aren't willing to use higher frequency RAM when they are OK with overclocking, I think that shows that this is not something that we are alone on.

I think a lot of it is that there is a big difference between the different "groups" on the internet when it comes to things like this. On the one hand, you have people like us who have thousands of systems going through our door which allows us to get a ton of hands on experience with a huge number of systems. Not to mention the fact that we have the resources to do a bunch of hardware testing to really drill into things like stability (where you need to do a ton of repeating testing to even start to measure). You don't often hear much from this group since they tend to keep any information they glean to themselves as it is an advantage they have over their competitors. We are really odd how we freely publish most of what we learn.

On the other hand, there are the people who build their own machines and love to tinker and talk about it. This group is often pushing the envelope and going way beyond spec with CPU overclocking, high frequency memory, etc. There is absolutely nothing wrong with that, but I think they tend to be more willing to forgive any small issues with the systems they build. If they have a bit of instability at first, but they tweaked the RAM voltage a bit and now it works fine, they are probably completely fine with that. But if that same situation happened to one of our customers, it would be a huge deal since they are almost never the kind of people who want to have to go into the BIOS and tweak things.

In our most recent articles, we have actually switched up to DDR4-2933 for the most recent Ryzen and Threadripper systems. We aren't using that speed of RAM yet in our systems, but it should be safe and within spec for these platforms. What we are waiting on is for 2933MHz RAM that meets JDEC specs to be available. Even then, we are going to take it pretty slow to make sure we don't start getting an increase in stability issues.

Posted on 2019-12-02 18:43:13