Puget Systems print logo

https://www.pugetsystems.com

Read this article at https://www.pugetsystems.com/guides/1671
Article Thumbnail

AMD Threadripper 3990X: Does Windows 10 for Workstations improve Adobe CC performance?

Written on February 17, 2020 by Matt Bach
Share:

Introduction

With the launch of AMD's new Threadripper 3990X 64 Core processor, there was a concern by some reviewers that the standard version of Windows 10 Pro was limiting the performance of this powerful CPU. This was largely led by Anandtech's review where they explained that Windows 10 Pro can only handle 64 threads within a single "Processor Group", and since the 3990X has 128 threads when SMT (Simultaneous Multithreading) is enabled, it has to create two processor groups to manage this CPU. In their review, they stated that to get around this issue, you either need to use Windows 10 Pro for Workstations (or Enterprise) or disable SMT, which in a few isolated cases resulted in up to an 8.3x increase in performance.

On the other hand, Tom's Hardware put up a post about how they did not see any benefit to using Windows 10 Enterprise, and that AMD themselves told them that "Higher editions/versions of Windows 10 confer no additional performance or compatibility benefits to the processor".

Due to the concern about Windows 10 Pro holding back the Threadripper 3990X, we decided to do additional testing with the 3990X and a number of other processors to see if using Windows 10 Pro for Workstations or disabling SMT/HT can result in significant performance gains. We cannot speak for many of the tests that Anandtech and Tom's Hardware use, but in this article, we are going to focus on one of our specialties: Adobe Creative Cloud.

AMD Threadripper 3990X 64 Core Windows 10 Pro for Workstations and disabling SMT

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

Looking for a Content Creation 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

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

AMD Ryzen Test Platform
CPU

AMD Ryzen 9 3950X ($749)

CPU Cooler Noctua NH-U12S
Motherboard Gigabyte X570 AORUS ULTRA
RAM 4x DDR4-2933 16GB (64GB total)
Intel 9th Gen Test Platform
CPU

Intel Core i9 9900K ($499)

CPU Cooler Noctua NH-U12S
Motherboard Gigabyte Z390 Designare
RAM 4x DDR4-2666 16GB (64GB total)
AMD Threadripper 3rd Gen Test Platform
CPU AMD TR 3990X ($3,990)
AMD TR 3960X ($1,399)
CPU Cooler Noctua NH-U12S TR4-SP3
Motherboard Gigabyte TRX40 AORUS PRO WIFI
RAM 4x DDR4-2933 16GB (64GB total)
Intel X-10000 Series Test Platform
CPU Intel Core i9 10980XE ($979)
CPU Cooler Noctua NH-U12DX i4
Motherboard Gigabyte X299 Designare EX
RAM 4x DDR4-2933 16GB (64GB total)
Shared Hardware/Software
Video Card NVIDIA GeForce RTX 2080 Ti 11GB
Hard Drive Samsung 960 Pro 1TB
Software Windows 10 Pro 64-bit (version 1909)

Windows 10 Pro for Workstations (1909)

Adobe Premiere Pro 2020 (Ver. 14.0.1)
Adobe After Effects 2020 (Ver. 17.0.2)
Adobe Photoshop 2020 (Ver. 11.0.2)
Adobe Lightroom Classic 2020 (Ver. 19.1)

PugetBench for Premiere Pro V0.86 BETA
PugetBench for After Effects V0.82 BETA
PugetBench for Photoshop V0.8 BETA​​​​​​​
PugetBench for Lightroom Classic V0.85 BETA​​​​​​​

*All the latest drivers, OS updates, BIOS, and firmware applied as of January 20th, 2019

A few notes regarding the hardware and software used in our testing:

First, since many of the platforms we are testing support different RAM speeds depending on how many sticks you use and even whether those sticks are single or dual rank, we have decided to standardize on using four 16GB sticks of whatever the "middle" supported RAM speed is (rounding up when applicable). This means that the latest Ryzen, Threadripper, and X-series platforms are using DDR4-2933 while the Intel 9th Gen is using DDR4-2666.

The second thing to note is that we are using Noctua air coolers on all of our test platforms. We do plan on upgrading to the Noctua NH-U14S TR4-SP3 for the Threadripper CPUs in the future as that is what we will be using in our workstations, but in our air-conditioned Labs environment on open-air testbeds, we did not encounter any significant performance degradation with the NH-U12S TR4-SP3 and wanted to limit the number of variables between this set of testing and the testing we did right when the 3990X launched.

Benchmark Results

While our benchmark presents various scores based on the performance of each test, we also like to provide the individual results for you to examine. 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 the scores that our benchmark calculated.

Feel free to skip to the next sections for our analysis of these results to get a wider view of how each the different OS and SMT/HT configurations perform in Premiere Pro, After Effects, Photoshop, and Lightroom Classic.

Premiere Pro Benchmark Analysis

Threadripper 3990X Windows 10 Pro for Workstations and SMT Disabled performance in Premiere Pro

Out of the Adobe Creative Cloud applications we are testing, Premiere Pro benefits the most from higher core count CPUs which means that it is the most likely to be impacted by using Windows 10 Pro for Workstations. However, on the platforms we tested, there was no significant increase (or decrease) in performance versus the standard Windows 10 Pro version. All the results were within a few percent, which is well within the margin of error.

Disabling SMT (AMD) or HT (Intel), however, did result in some performance change - but not for the better. On most of the CPUs we tested, we saw roughly a 10% drop in performance with HT/SMT off. The 3990X was the odd man out since it technically saw a 2% increase in performance, but that is within the margin of error for real-world tests like this. In other words, disabling SMT/HT at best provides no benefit in Premiere Pro, and at worst can drop performance by ~10%.

After Effects Benchmark Analysis

Threadripper 3990X Windows 10 Pro for Workstations and SMT Disabled performance in After Effects

Moving on to After Effects, Windows 10 Pro for Workstations again provides no solid benefit over the normal version of Windows 10 Pro.

Disabling SMT/HT did provide a small ~5% boost in some cases, but in others, it resulted in a 7% drop in performance. 5-7% is starting to get beyond the margin of error for this benchmark, but considering the fact that it was inconstant in terms of helping or hurting, we would recommend keeping HT/SMT enabled for After Effects.

Photoshop Benchmark Analysis

Threadripper 3990X Windows 10 Pro for Workstations and SMT Disabled performance in Photoshop

In Photoshop, we once again saw no benefit to using Windows 10 Pro for Workstations. Disabling SMT/HT also resulted in no significant change in performance with the exception of the Intel Core i9 9900K where we saw a 7% drop in performance. With the higher core count CPUs (including the Threadripper 3990X), the difference between SMT/HT being on or off was only a few percent which is nowhere near enough to be meaningful.

Lightroom Classic Benchmark Analysis

Threadripper 3990X Windows 10 Pro for Workstations and SMT Disabled performance in Lightroom Classic

Lightroom Classic is an application we looked forward to testing since there are known issues with SMT/HT that we have discussed in our support guide: Hyperthreading & SMT causing low performance in Lightroom Classic. The crux of the issue is that in some tasks (primarily exporting and generating previews), we noticed up to a 50% performance gain with SMT/HT disabled on certain CPUs. On the other hand, having SMT/HT enabled resulted in better performance for many of the "active" tasks in Lightroom Classic.

In this case, Windows 10 Pro for Workstation made effectively no difference so there is no concern about Lightroom Classic having issues with multiple processor groups in Windows 10 Pro. There were, of course, some small discrepancies in the results, but everything was well within the margin of error.

Disabling SMT/HT did result in some decent performance gains with a few of the CPUs (most notably the Core i9 10980XE), but the 3990X in particular only saw a small 3% gain in performance which is again within the margin of error. In fact, the benefits of disabling SMT/HT even on the passive tasks like exporting was much smaller than what we saw in previous versions of Lightroom Classic, so there is currently little reason to disable SMT/HT unless you export huge numbers of images.

Does Windows 10 Pro for Workstations or disabling SMT/HT improve Adobe CC Performance?

If you are using the AMD Threadripper 3990X (or any other CPU for that matter), we found no reason to upgrade to Windows 10 Pro for Workstations for the Adobe Creative Cloud applications we tested. Neither Premiere Pro, After Effects, Photoshop, or Lightroom Classic showed any performance gain, so you may as well save your money and stick with the normal version of Windows 10 Pro.

Disabling Simultaneous Multithreading (AMD) or Hyperthreading (Intel) also rarely improved performance, and often resulted in a small loss of performance. In some cases - like exporting in Lightroom Classic - it can improve performance, but that is not unique to the 3990X and many other CPUs (like the Core i9 10980XE) actually benefit to a more significant degree.

We don't want to discount Anandtech's findings since there very well may be some applications that benefit from using Windows 10 Pro for Workstations with the AMD Threadripper 3990X, and we know for a fact that there are some applications that benefit from disabling SMT/HT. But as far as Creative Cloud goes, we found no evidence that there is any reason to upgrade to Windows 10 Pro for Workstations.

Looking for a Content Creation 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: Intel 9th Gen, Intel X-series, AMD Ryzen 3rd Gen, AMD Threadripper 2nd Gen, Intel X-10000, Threadripper 3990X, Premiere Pro, After Effects, Photoshop, Lightroom CLassic

I think you don't have enough ram per processor to test the effect properly. The point is that that number of processors hit throttling under five circumstances, but i suspect the low ram is the bottleneck. Obviously mostly single threaded apps aren't really relevant, but i think showing processors and threads maxed out is more relevant to premiere, and maybe Photoshop. 64gb ram really only enough for 16 threads isn't it?

Posted on 2020-02-23 09:39:28

The amount of RAM you need for an application is not impacted by the number of cores you have, at least not in any of these apps. Our benchmarks are all designed to need less than 32GB of RAM, so having "only" 64GB is more than enough to not be an issue. In real workflows, you can need more than 64GB in some cases, but that is due to things like having multiple apps/projects open, really long compositions, and other things that do not directly affect performance (at least not much).

The only exception is After Effects, and that is only when using Aerender to spawn multiple render threads (which is not "stock" behavior). In that one instance, however, the Threadripper CPUs actually do pretty poorly anyways and the amount of RAM you need completely depends on your comp and how many render threads your system can handle (which also depends on VRAM capacity).

Posted on 2020-02-23 15:09:59

I have to disagree with you for video apps that require min 1GB per frame at HD for each processor to load the frame and process it. GB per thread is vital in any multi process video based work flow, but particularly in after effects, which it seems will never be truly multi process aware. I don't think you can get away from needing the whole frame in ram to process it. Is like to see tests with more ram to see where you actually hit the bottlenecks.

Posted on 2020-02-23 16:51:51

After Effects does load the frame into RAM for RAM Preview, but it still has no bearing on the number of CPU cores (again, outside of using Aerender). All it means is that the more RAM you have, the more frames yo can have have cached. Resolution and bit depth determine how many frames can be cached, as we go over here: https://www.pugetsystems.co...

The reason it doesn't matter for our testing is that even at 4k 16-bit, our test comps are not long enough to need more than 32GB of RAM. That is by design since testing a longer comp isn't going to change the performance in terms of FPS, and this way we (and anyone else with at least 32GB of RAM) can benchmark their system without the RAM capacity being an issue.

Also, even if it was a problem, it wouldn't result in lower performance - AE just stops the RAM Preview when it runs out of RAM. In that case, out benchmark would throw an error and not complete.

RAM is shared across the entire CPU - it isn't like each cores needs it's own frame buffer like multiple GPUs do with VRAM. The only time more cores means you need more RAM is when you are running renders in parallel. Aerender is the only Adobe app I know of that can do that, and even outside Adobe (Avid, Resolve, etc.) that is not the case.

You are welcome to ask Adobe directly if you want, but we talk fairly regularly with many of the dev teams (including to help develop our benchmarks), and they have never expressed even a tiny bit of concern over our testing configurations.

Posted on 2020-02-23 18:19:38
Nick Lam

Because we don't see a difference, what is this telling us about how these applications are dealing with two groups of 64 cores? Does this mean they are group aware?

At any given time, did you see all 128 threads being used simultaneously?

Posted on 2020-02-28 02:45:51

I think two things are at play:

1 - From everything I've seen and the testing we've done here on other applications, it doesn't look like there is actually any practical performance difference between Windows 10 Pro and Windows 10 Pro for Workstations. The initial report that there was a difference seems pretty suspect at this point to me, I think they were just mistaken.

2 - Regardless of that, Adobe applications don't tend to max out the thread count on these high core processors. We've known that for a long time, and it's actually true of most applications out there. Only a handful of things really hit all of the cores in a processor like the 3990X.

Posted on 2020-03-02 19:32:51
Dragon

The upside of all this is that I am still happy with my decision to build a 9900K system last year. It is holding up well for CC in spite of the onslaught of cores.

Posted on 2020-03-04 18:36:16

Yea, more cores is definitely not as useful as a lot of marketing makes it out to be. Sure, if you are doing ray traced rendering that scales well with more cores, but video editing? There are definitely diminishing returns well before you get into those 20+ core range.

Posted on 2020-03-04 19:04:50
Dragon

Yup, I do the occasional ray trace in Chief Architect and debated more cores at the time of the build, but the 9900K was faster for just about everything else. The system has a good AIO, so it holds 4.9 GHz on all cores and stays below 70 deg fully loaded. Even the ray tracing was a huge jump from my previous 3770. Now I think I will wait until more of the software catches up with the core jump. Probably just about in time for Intel 7nm. :-).

Posted on 2020-03-04 20:11:19
Dune Haggar

what's your recommendation for a person using adobe illustrator (complex art that pushes Illustrator to its limits), photoshop, indesign, blender and after effects as main apps?
I am guessing mainly high single core speed? do you have any benchmarks with illustrator?

Posted on 2020-03-06 06:56:05

Yep, you pretty much want the best single core speed - so Ryzen or Intel 9th Gen right now. Blender is the only one that may benefit from more cores, but that is only if you are using the rendering engine in CPU mode instead of GPU.

We don't have any Illustrator benchmarks at the moment. We have considered it, but there hasn't been much of a demand from what we have seen. Photoshop and Illustrator also both behave very similarly, so we've never had an issue using our Photoshop testing as a stand-in for Illustrator performance.

Posted on 2020-03-06 16:19:51