Puget Systems print logo

https://www.pugetsystems.com

Read this article at https://www.pugetsystems.com/guides/1524
Article Thumbnail

First Look at AMD Ryzen 3rd Gen CPUs for Rendering

Written on July 7, 2019 by William George
Share:

Introduction

AMD launched their new line of mainstream Ryzen processors today, along with a new motherboard chipset called X570, but we were only provided with a sample of the low-end Ryzen 5 3600 ahead of time. We have ordered the Ryzen 7 3800X and Ryzen 9 3900X and should be testing them soon, but until then we can at least look at how the overall architecture is doing with the example we do have.

AMD Ryzen 3000 Processors and X570 Chipset

While this is technically the third generation of Ryzen CPUs from AMD, this is actually the first set of processors which use their new Zen 2 architecture. The first and second gen Ryzen chips were built on Zen and Zen+ architectures, respectively, and the main difference there was a die shrink from 14 to 12nm. The actual "floor plan" for the chips didn't change much, so performance gains were limited to clock speed increases that were facilitated by the smaller die size.

With Zen 2, and the new Ryzen 3000-series processors, there has been a move to a multi-chip module design with the I/O components separated onto their own die. The compute sections of the CPU - nicknamed "chiplets" - also received another die shrink down to 7nm. Between those and other architectural changes, Zen 2 based processors are expected to have higher clock speeds, improved performance per clock, and higher core counts. That combination should give users great performance from these new AMD chips!

Test Hardware

For this preview we are going to just look at the Ryzen 5 3600 sample AMD sent us and compare it with one previous-gen Ryzen 5 processor and one Intel Core i5 model that is similar in price and specifications. Our full review with additional Ryzen 3rd Gen chips will include a few other processors for a better overall comparison, but that will have to wait till AMD sends us the CPUs we've ordered.

Another topic which we will investigate in more depth in the coming weeks is memory speed since this new generation of processors supports higher RAM frequencies. The exact speeds AMD officially supports are a bit tricky to find, and they vary depending on factors like how many memory modules you are using, but I did want to include a small taste of this testing here in our preview article. As such, the Ryzen 5 3600 was run through these benchmarks with both 2666MHz and 3200MHz memory and both results are included below. The older AMD processor, as well as the Intel counterpart, officially do *not* support that speed - so we tested them with only 2666MHz.

AMD Test Platform
CPU AMD Ryzen 5 3600
AMD Ryzen 5 2600X
CPU Cooler AMD Wraith PRISM
Motherboard MSI MPG X570 Gaming Edge WiFi
RAM 4x DDR4-2666 16GB (64GB total)
4x DDR4-3200 16GB (64GB total)
Video Card NVIDIA GeForce RTX 2080 Ti 11GB
Hard Drive Samsung 960 Pro 1TB
Software Windows 10 Pro 64-bit (version 1903)
Cinebench R20
V-Ray Next Benchmark
Intel Test Platform
CPU Intel Core i5 9600K
CPU Cooler Noctua NH-U12S
Motherboard Gigabyte Z390 Designare
RAM 4x DDR4-2666 16GB (64GB total)
Video Card NVIDIA GeForce RTX 2080 Ti 11GB
Hard Drive Samsung 960 Pro 1TB
Software Windows 10 Pro 64-bit (version 1903)
Cinebench R20
V-Ray Next Benchmark

Benchmark Details

Since the focus of this preview is on CPU based rendering performance, we used two benchmarks: Maxon's Cinebench R20 and Chaos Group's V-Ray Next Benchmark. These are the latest in a long line of benchmarks from both companies, and they use their latest engines to see how quickly different systems can render still images. Both include a test that involves all available CPU cores, and the V-Ray Next Benchmark also includes a GPU-rendering mode which can utilize the CPU as well. Please keep in mind that if you are working with animations, rather than just still images, there will also be processing time between each frame which may behave differently than the heavily-threaded rendering itself... so single-threaded performance, which we are not looking at here, will matter too.

As always, each test was run 2-3 times per hardware combination and the best result is included below.

Results: Cinebench R20

First up, here are the performance numbers for Cinebench R20's CPU test - which uses all available cores to render a single scene:

Cinebench R20 CPU Test AMD Ryzen 5 3600 vs 2600X vs Intel Core i5 9600K

Results: V-Ray Next

The V-Ray Next Benchmark test includes a few different modes, so the results are spread across three graphs in this gallery:

Analysis

AMD's Ryzen 5 3600 performs really well in both of these benchmarks, outpacing the previous generation 2600X by about 15% and beating Intel's Core i5 9600K by over 30%. This isn't quite a fair comparison to Intel in one sense: all of these are six-core processors, but both of AMD's Ryzen 5 chips have SMT while the Core i5 here lacks Intel's version of that technology (Hyperthreading). That is definitely hurting Intel's performance here, but at the same time, both the 3600 and 2600X are less expensive than the 9600K. Intel doesn't have any 6-core chips in the 9th Gen Core series with HT to include as a more "fair" comparison either, and while we could have gone back and tested the older Core i7 8700K to find that 6-core / 12-thread feature combination it was even more expensive.

Looking just at the 3600 now, and comparing its performance with 2666MHz and 3200MHz memory, there is less of a difference than I expected. Both Cinebench R20 and V-Ray Next GPU mode showed no significant difference - the 2666MHz memory actually scored a little better in those test, but the results are well within the margin of error. V-Ray Next's CPU mode did have the 3200MHz memory come in about 2% faster, which is bordering on margin of error too... but from the way, a lot of AMD fans talk about the importance of memory speed I had expected a bigger difference. We'll see if this changes when looking at a wider range of CPUs and RAM soon.

Conclusion

CPU Rendering Performance Comparison Between AMD Ryzen 5 3600, 2600X, and Intel Core i5 9600K

AMD has come out swinging with the launch of the third generation of Ryzen processors, and in well-threaded applications like rendering it looks like they will topple Intel's mainstream Core series. This is very similar to the situation we saw last year with AMD taking the high-end desktop (HEDT) multi-threaded performance crown from Intel with their second generation Threadripper processors - but this time they seem to have done it with better performance at the same core count rather than by simply having more cores than Intel (which was Threadripper's trick). Admittedly, this preview with a single Ryzen 5 3600 is limited in scope... but if the higher end models follow suit, AMD is poised to do very well in many applications. Hopefully, the X570 platform proves stable, enabling its use in our workstations!

Tags: AMD, AMD Ryzen 3rd Gen, Intel vs AMD, Cinema4D, Cinebench, V-Ray, Processor, CPU, Rendering

30% higher performance and it costs less, and more secure? Where am I, is this reality?

As always, thank you for the reviews.

Posted on 2019-07-07 13:47:13
Leroy Brown

Zen 2 is looking good. Is it possible to include Capture One when you get around to photography benchmarks?

Posted on 2019-07-07 19:27:22

Capture One is on our list of potential software to test, but we primarily look at software that we specifically target for our workstations. Capture One may be one of those at some point, but not for a while I think.

Our decision to target software like that isn't just a matter of performance testing. We also have to develop a relationship with the developer so that we have a line of communication to help fix any customer issues. Not to mention training our entire service staff (sales and support) in how the software works, common issues, workflows, etc. It is a huge amount of work, which is why we only target a limited number of applications directly.

Posted on 2019-07-08 15:24:54
Hahaha

Capture One Pro is a popular software among photographers instead of LR CC and we dont see any benchmark or test results. I hope to see some testing with Capture One Pro soon.

Posted on 2019-08-11 07:28:30
Hugh

I would second that, although Capture1 is less resource intensive than Lightroom from my personal editing observations

Posted on 2019-08-12 11:22:11
Johnny

Sooo keen to see the crème de la crème 3800x compared 9900k after effects performance..

Posted on 2019-07-07 23:12:19
Nino Skupnjak

If you have 3900x, can you please test it in Reality Capture? :)

Thanks

Posted on 2019-07-08 14:09:21

We just got a 3900X and have a 3700X on the way - and yes, RealityCapture is on the docket for testing :)

Posted on 2019-07-11 05:51:20
Nino Skupnjak

Thank you, I'm looking forward to it :)

Posted on 2019-07-11 08:52:03
Neo Morpheus

And yet, no OEM have a system in stock for me to buy. I have to wonder if Intel is back to its old tricks of paying to have AMD away from the mainstream...

Posted on 2019-07-08 14:32:00

I can't speak for other companies, but we aren't offering these CPUs yet because we haven't completed our qualification process yet. I posted this as a reply to your comment on our other article, but for the benefit of anyone else reading:

We don't list any products in our line until they have completed our full qualification process. We still haven't tested the higher-end models, not to mention evaluating different motherboards to determine which is the most stable with the feature set our customers need.

We don't just toss hardware together into our systems and hope it works - a huge amount of effort goes into each product launch to make sure our customers have the best experience possible. Most of our customers are not tech enthusiasts, they are video editors, VFX artists, animators, etc. who want a rock solid system and don't want to have to tinker with anything.

Performance testing is just the tip of the iceberg for us and we haven't even completed that yet for these CPUs.

Posted on 2019-07-08 15:19:54
MarketAndChurch

And thank you for taking the time to do all of that.

Posted on 2019-07-16 08:33:47
M2018

Thanks for the article. Some impressive numbers there.

Not sure if it's going to make a huge difference for these tests, but I suggest you use higher speed RAMs like 3200MHz (AMD's official rating for Ryzen 3000's) going forward. With RAM prices so low even for 3200/CL16, there's really no reason to test with lowly 2666MHz/CL19.

Posted on 2019-07-08 20:29:07

I actually included both 2666MHz and 3200MHz memory in my tests above :) [for the Ryzen 5 3600 only, not the other two CPUs]

We are planning to do even more in-depth memory comparisons in the coming weeks, but selecting what speed of memory we eventually offer with these Ryzen processors will be tricky. AMD doesn't just give a maximum supported memory speed: they give multiple speeds, based on how many modules are installed and whether you are using single- or dual-rank modules. I believe the full capacity (4 slots + dual-rank) maximum supported speed is all the way down at 2666MHz, but AMD isn't very forthcoming with those details.

That is why we want to do more comprehensive testing: to answer the question of "how much does memory speed really matter to these processors?" If it turns out that faster memory only increases real-world applications by a couple percent, then we might as well stick with the lower speeds that we know are fully supported by AMD and which have also proven to be highly reliable over the years. On the other hand, if a moderately faster memory (like 3200MHz) is a lot faster... then we'll have to decide internally whether we are okay breaking AMD's officially supported specs and running the risk of higher failure rates in order to gain that performance for our customers. Stay tuned in the coming weeks to see how that all ends up playing out :)

Posted on 2019-07-08 20:34:50
M2018

Oop, my apologies! I was looking at the other article on video editing which was run at 2666MHz only.

Posted on 2019-07-08 20:51:02
justintime

So you used 3200Mhz in all 4 slots. Did that mean it only ran at 2666MHz - and so gives identical result? I wonder if it improves if you use 2 sticks of 3200.

Posted on 2019-07-08 21:13:43

I was using 3200MHz memory in all 4 slots, and it was set to run at the full 3200MHz. That is beyond AMD's officially supported speed for these Ryzen 3rd Gen processors, though... which historically is something we have avoided doing here at Puget, since we value reliability and support from hardware manufacturers over having the flashiest sounding specs.

You can always set memory speeds and timings beyond what the CPU manufacturer supports, or even what the RAM modules you are using are rated for - the only limit is what the motherboard's BIOS allows. However, if you push memory above manufacturer specs then it can void warranties. Even more importantly, it can lead to system instability (blue screens, application crashes, data corruption, etc). In extreme cases, it may not even boot up with improper settings.

Tom's Hardware listed AMD's official supported memory speeds for these new CPUs, which you can see if you scroll down about half-way on this page: https://www.tomshardware.co...

Depending on the number of modules you have installed, and whether they are single- or dual-rank, it can be as high as 3200MHz or as low as 2667MHz (the same as 2666, which is the way I am more accustomed to writing it, just rounded up instead of down).

Posted on 2019-07-08 21:21:51
David Varela

heya, for davinci resolve a 3200 CL14 vs CL16 has ANY impact? thanks!

Posted on 2019-07-20 09:02:01
Roman

Hi, please, would be possible to test new Ryzen 3900x in Phoenix FD for 3ds max? I'm really curious it it can beat i9 9900k or even 9920x

Posted on 2019-07-09 08:39:12

Unfortunately, at this time we do not have testing built up for 3ds Max itself or any plug-ins for it. We do want to add 3ds Max and Maya to our benchmarking, but even when that happens it is unlikely that plug-ins will be included. If you know of any existing, public benchmarks for this sort of software, though, I would be interested in hearing about them. It seems like fluid dynamics and other simulations are an area that is under-served in terms of real-world testing.

Posted on 2019-07-09 16:52:35

Very impressive results! I am looking forward to seeing how the motherboards shake up in qualification testing.

Posted on 2019-07-09 18:14:07
Brian Mack

I may have missed it but do these tests include patches for Spectre, Meltdown, etal that affect only Intel's speculative execution?

Posted on 2019-07-10 07:26:52

The systems used in this test were all running Windows 10 Pro 1903 and were fully patched / updated, so I assume all applicable fixes for those vulnerabilities were installed.

Posted on 2019-07-10 16:31:34
Brian Mack

Thx! Can I look forward to reading an article on how SolidWorks runs on these new Ryzen? Intel was (?is?) the single threaded king but with AMDs improved IPC and the aforementioned patches, I wonder if the king has been dethroned or at least truly challenged.

Posted on 2019-07-11 05:49:21

Yes, we are working on SW 2019 SP3 testing on these new Ryzen chips and a few of Intel's most direct competitors.

Posted on 2019-07-11 20:03:45
Nicholas Lemley

Hey William! I know that this is probably a stretch considering the scope of the GPUs that you all typically focus on for the software, but is there any shot that we could see some benchmarks of the RX 5700 reference cards in Solidworks compared to AMD's Polaris architecture cards and even the Radeon VII? I admittedly am trying to do too much with a single system with too many constraints, but I am trying to make as educated of a decision as possible about which of the currently offered AMD cards are a solid choice for a myriad of productivity softwares. If this is something that is or could be in the pipeline then it would be awesome to know so I can hold off on purchasing until I've seen good data. Attempting to linearly correlate different benchmarks to find a loose position in rankings isn't something I feel confident in.

Posted on 2019-07-21 16:25:50

Hmm, I don't think it is likely that we'll do any testing of GPUs for SW outside of models that are supported (or should eventually be supported) by Dassault. I know AMD talked about having "Pro" drivers for the Radeon VII, but as I understand it that is not something which will lead to them being certified for applications like SW... and indeed, the only AMD cards listed as certified for SW are FirePro and Radeon Pro models :(

https://www.techpowerup.com...

Part of the reason for this testing limitation is that we focus on configurations which we could sell to our customers, and in the case of SW you cannot enable certain graphics features if the video card is not officially supported. That would make them less that ideal for our users, and also impacts the ability of our benchmark to run properly.

Posted on 2019-07-22 17:46:26
CJ

In previous articles you noticed that 2700x is worst performing in AE cc19 than 9600k. In this article 2600x ahead of 20% 9600k. Reds money brought to 2019?

Posted on 2019-07-10 11:44:34

This article is about CPU based rendering in Cinema 4D and V-Ray Next, not After Effects. Did you perhaps mean to post this question on a different article? I'm also not sure what your question at the end there means... "Reds money"?

Posted on 2019-07-10 16:32:37
LondonArcher

Do you have a standard Lightroom benchmark? I'd like o see something like a standard set of 100 RAW Files and timed how long it takes to export them to JPEG, with these files available online so we can download and test them against our own system.

Posted on 2019-07-10 15:09:28

We do not currently have a public benchmark for Lightroom. We do have internal tests we use for that application, which Matt built and utilizes for his articles about that software... and I believe he is working on a public version, but it is not yet ready. If you want to see the public benchmarks we do have, or just to know where to look for a Lightroom benchmark in the future, check out this article list: http://puget.systems/go/ben...

Posted on 2019-07-10 16:36:42
Global_Saffer

Looking forward to a review with Metashape, Pix4D and Linux!

Posted on 2019-07-12 05:30:12

Metashape is on the list, but may be a while (RealityCapture will be sooner). Pix4D is *not* currently on my plan because it seems to do better on higher core count processors. I do intend to check it on the 16-core 3950X when that eventually comes out.

Linux testing is not my area of expertise, but I believe Dr. Don Kinghorn is working on that :)

Posted on 2019-07-12 18:06:40
DevGuy

Thanks for this timely write up. It's articles like this that make Puget Systems stand out. Professionals, content creators, etc. generally don't want to deal with the myriad of hardware choices and that's partly why many end up buying substandard relatively proprietary and overpriced hardware from Apple where the corporate design ethos is form over function. It's nice to see companies like Puget Systems focus on function and productivity.

Posted on 2019-07-13 11:50:38
Matt

Thanks for the article.
Could you guys please do Adobe Premiere Pro Warp Stabilizer tests? That is the #1 time waster as a real estate videographer. It's my understanding that Intel is optimized better for Adobe and that Warp Stabilizer isn't optimized to benefit from more cores. Does ryzen 3000 possibly have any improvements in this department over a 9900k? I'm doubtful it will, but I sure hope it does. It was my understanding I'm better off buying a cpu with the highest single core speed as possible.

Posted on 2019-07-13 17:14:43

Warp stabilizer is a part of our After Effects benchmarks, and from the testing I've done, relative performance between CPUs is nearly identical for that effect in both applications. So if warp stabilize is a big factor for you, just take a look at the raw results for that test in out After Effects article when it comes out.

Sneak peak though, the 9900K is about 5% faster than the 3900X for warp stabilize. Not a whole lot, but if that is your #1 concern then the 9900K is still better since warp stabilize is single threaded.

Posted on 2019-07-15 16:42:10
MarketAndChurch

Can't wait for your 3900x (and later in September, 3950x) + Navi 5700 XT on X570 reviews. I use after effects and Davinci Resolv, 64GB of RAM, and am interested in both Gigabyte's X570 Xtreme and MSI's X570 Creation, as well as the 3950x. I'd also love it if you could compare Ryzen 9's top two CPU's against both Intel and AMD's HEDT offerings. Hopefully you'll be able to a few of those tests out - on the 3900x + 5700 XT - before September. Keep up the good work.

Posted on 2019-07-16 08:39:48
Jig Serencio Navasquez

Hi Im really Waiting for the 3900x bench for AE please release it soon :D

Posted on 2019-07-17 01:35:47

We have almost all the testing done, just a matter of going through the thousands of data points, formatting it to actually be useful, and putting together the posts. I'm aiming for end of this week for the video editing focused articles, but that may slip to early next week depending on a few other projects we have going on right now.

Posted on 2019-07-17 01:38:03
Michael Du

Im really Waiting for the 3700x bench for Solidworks. please release it soon,Thxs

Posted on 2019-07-27 12:29:50

SW article just went up - a little later than I had intended, due to other stuff going on, but hopefully you find it helpful :)

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

Posted on 2019-08-28 19:03:37