Puget Systems print logo

https://www.pugetsystems.com

Read this article at https://www.pugetsystems.com/guides/1045
Article Thumbnail

Arnold for Maya CPU Comparison: New 14, 16, and 18-core Skylake-X Processors

Written on September 25, 2017 by William George
Share:

Introduction

Arnold is the render engine included with Maya 2017, and is also available as a plug-in for several other design applications. It is made by SolidAngle and produces beautiful 3D renditions of scenes and visual effects. When rendering those graphics, the performance of the central processor (CPU) in a computer is of paramount importance. CPU-based rendering generally scales well with both clock speed and core count, but those specifications cannot be directly compared across different brand or generations of processor. Here at Puget Systems we do real-world testing to ensure we provide our customers with the right computer for their needs.

Intel has just released a trio of new Core X series processors, with higher core counts than anything they have offered in this line to date: the Core i9 7940X, 7960X, and 7980XE - with 14, 16, and 18 cores respectively. We ran these chips through a few tests in Maya 2017, using the Arnold render engine, to see how they perform. We had recently put out an article on this same software that covered some additional processors as well, and at the time AMD's Threadripper CPUs were found to be the fastest single-CPU option. This time we will focus just on the Core X series in comparison to Threadripper, to see if the new models can take the performance title back for Intel.

Test Setup

To see how these different CPUs perform in Arnold, we ran it within Maya 2017 on the following configurations:

The main focus here is on the three new Core X (formerly code named Skylake X) processors, and specifically how they compare to AMD's Threadripper models. In our last round of testing we found that the 16-core 1950X took the lead in single-CPU performance with Keyshot, as well as many other CPU-based rendering engines. Dual CPUs are often a great way to increase performance as well, though they cost substantially more and are outside the scope of this particular article. Here are some details about how we conducted our testing, but if you just want to skip straight to the results then feel free to scroll past this section.

The results presented below are from Arnold for Maya, which comes standard with Maya 2017. That program is available from Autodesk for a 30-day trial, as well as in several subscription options for long-term use. Within Maya we opened a test file called from Solid Angle called "model village", and then navigated to the Arnold render window. The scene was rendered at two resolutions: 640x480 and 1280x960. No threads were reserved for the UI or other applications, so that we could measure the true performance of each CPU without artificial limitations.

Village render scene for Arnold from Solid Angle

It is also worth pointing out the speed of memory we used in these tests. We prefer to use the speed of memory that each CPU is rated for, according to its manufacturer. For the current crop of Core X and Threadripper processors that is DDR4-2666. These platforms can technically be run with even faster RAM modules, but that is pushing the memory controller built into the CPU past its rated speed - overclocking it, effectively. That may lead to slightly increased performance but we have also found it to lead to stability issues and higher rates of memory failure. For that reason we stick with the manufacturer specs when it comes to selecting RAM for our systems.

Benchmark Results

Here are the results, in seconds, for the various CPUs we tested in Arnold for Maya at 640x480. Lower times are better:

Arnold for Maya 640x480 Render Results With New Skylake X Processors

And here are the results for the same scene but rendered at 1280x960:

Arnold for Maya 1280x960 Render Results With New Skylake X Processors

Both sets of results are nearly identical in terms of relative CPU performance. Rendering in Arnold is a heavily threaded application, so there is a clear spread between the different processors based on core count. As shown in our last round of testing, AMD's 12-core 1920X ties Intel's 10-core i9 7900X. This time around, though, the 16-core 1950X is bested by the three new models from Intel - with the fastest 7980XE leading the AMD chip by over 20%. It also costs twice as much, but if you look at that additional $1000 in the scope of the whole system price (rather than focusing on the price of the CPUs themselves) it may be worthwhile for the increased performance.

It is also worth noting that AMD and Intel both have server-class processors with even more cores as well: up to 32 on AMD's EPYC and up to 28 from Intel's latest Xeon Platinum line. With their focus on the multi-CPU server segment those may not come into play for Arnold, or other CPU-based renderers, but if a manufacturer puts out a single-socket workstation motherboard that is compatible with either of those platforms then they may be worth a look.

Conclusion

As shown above, the tables have turned since our last Arnold rendering test. Intel took back the lead, but their new Core X processors are also substantially more expensive than AMD's Threadripper. If you want the absolute fastest single-CPU rendering speeds then go for the i9 7980XE - but if you want a great value, the 1950X has very strong performance while costing $1000 less.

Tags: CPU, Rendering, Arnold, Maya, 2017, MtoA, Solid, Angle, Performance, Processor, Intel, Core, i9, Skylake, AMD, Threadripper
Igioz

good to know, IPC is a intel throne yet, but price is an AMD match

Posted on 2017-09-28 11:12:43
Luca Pupulin

Hi,
Do you think of testing Maya 2017/2018 as a whole,now that finally the SPECapc for Maya has been updated?

Cheers,
Luca

P.S. : you accidentally wrote "Cinebench R15" as the software used in a couple of tests (here, keyshot,vray cpu comparison ..) ;-)

Posted on 2017-09-29 12:08:33

Thanks for the catch on those! I've fixed them :)

As for more general testing of Maya, I would like to do that! We have a lot of testing in the works right now, though, especially with the launch of so many AMD and Intel processors this year. We've had to do a lot of development on our own tools and procedures for testing many applications as well, but if there is a good Maya benchmark - one that tests real world usage cases, and in such a way that it will show accurate performance differences between systems - I would be open to that. I wasn't aware SPEC made a Maya test, though... it isn't listed on their website: https://www.spec.org/benchm... (and many of the ones they do list for other programs are outdated). Maybe that is an outdated list? I'll do some more digging when I can.

Posted on 2017-09-29 16:47:38
nichimen

Hi William,
I can guess you are pretty busy at testing different applications in different scenarios... 😉
Maya (and 3D packages in general) is quite a big and complex piece of software and ,as such,I know it's not easy to test;
Maya 2016 and up can take better advantage of multicore system and can leverage gpu power more than in previous releases; it could be interesting adding a second SSD to act as a cache drive for bifrost fluid simulation to see if it can speed it up even further,but this should be done by "hand";
Anyway the SPECapc for Maya 2017 is quite comprehensive (I've used it already).
This is the link:

https://www.spec.org/gwpg/a...

I saw the last Maya benchmark you performed was on Maya 2014 with SPECapc for Maya 2012

Cheers,
Luca

Posted on 2017-09-29 21:18:26
Luca Pupulin

Hi William,
I can guess you are pretty busy at testing different applications in different scenarios... 😉
Maya (well..a 3D package in general) is quite a big and complex piece of software and,as such,I know it's not easy to test;
Maya 2016 and up can take better advantage of multicore system and can leverage GPU power more than previous releases;It could be interesting adding a second SSD to act as a cache drive for bifrost fluid simulation to see if it can speed it up even further,(but this should be done by "hand");
Anyway the SPECapc for Maya 2017 is quite comprehensive (I've used it already).
This is the link

https://www.spec.org/gwpg/a...

I saw the last Maya benchmark you performed was on Maya 2014 with SPECapc for Maya 2012

Cheers,
Luca

Posted on 2017-09-29 21:48:23

Thank you for that link, and for sharing your experience! I will check it out :)

Posted on 2017-10-02 17:28:54
Luca Pupulin

Hi,
glad to be helpful,somehow.. 😉

As a side note...
I'd like to point out that this benchmark (as well as all SPECbenches for other graphics/CAD programs) makes a viewport screengrab of every single test and save it along with the score.
Mere numbers are a good sign of performance of course,but for a 3D artist a good representation of the scene is also vital;
I believe this is an important point to consider/compare when you are testing GeForce and Quadro cards in Maya;
I think you should have better performance/visual representation when using Quadro cards with OpenGL and better one with GeForce using DirectX.
I've been using Quadro cards on Linux for quite some time now..but I'm curious about that...

Cheers,

Posted on 2017-10-03 17:06:26

Well, bad news: I just went to download the SPECapc for Maya 2017, to see how it works and try it out on a system... and ran into a roadblock. It is free for individuals and most companies, but for "sellers of computers and related products" (which we are) the cost is $2500. Considering that we already have a license for Maya itself, and that this is the price for a single benchmark application, I am not sure we can justify that steep price tag. I will talk with some folks here and see, but we may well have to develop our own way of testing in this situation.

Posted on 2017-10-03 22:37:38
Luca Pupulin

Hi,
I noticed that it is actually a bit pricey...
anyway,If you are going to develop your own procedures of testing,
I'll be glad to share my knowledge with you.

Posted on 2017-10-06 17:09:02
CobaltGamer

I will be getting i9-7940X (I mean once I get enough money) because out of most the benchmarks it had better performance than TR 1950X and I need the single threaded performance of Intel and the big multi-threaded performance (but it is really overpriced though)

Posted on 2017-09-30 22:34:20
CobaltGamer

Will you ever do Blender? That would be really helpful

Posted on 2017-09-30 22:34:43

Maybe, but as far as I know it is not currently on our list of upcoming software. We try to look at what our customers tell us they are using their computers for, and then base our target software on the most common applications. I'm not sure where Blender stacks up in that list at the moment.

Posted on 2017-10-02 17:27:56
Eduardo Souza

This benchmarks focused on 3D rendering are great! Can't thank you guys enough for taking the effort.
Is Houdini in the list of upcoming software? It would be awesome to have a comparison of sim, and mantra rendering times on these new CPUs.

Posted on 2017-10-04 18:10:00

Houdini is on our list, but there are a few software packages we want to tackle first. If I had to guess, I would say sometime early 2018 is when we will (hopefully) get to it. One thing that is often a big hurdle for us is getting our hands on real-world project to use in our testing. If you have some projects we can use that highlight a specific pain point you have, definitely let us know! You can email us at labs@pugetsystems.com if you have anything you think would be useful.

Posted on 2017-10-04 18:41:31
Eduardo Souza

Cool! I'll separate some projects, and see what I can send!
Also, whenever you feel like tackling Houdini, Odforce.net is a forum of Houdini enthusiasts, that I'm sure will have stuff to share and suggestion to give on how to benchmark it.

Posted on 2017-10-05 01:51:27
glassworks

Thank you for Maya Arnold real-world benchmarks!

Posted on 2018-08-11 15:23:52