Puget Systems print logo

https://www.pugetsystems.com

Read this article at https://www.pugetsystems.com/guides/509
Article Thumbnail

AutoDesk Maya 2014 Professional GPU Acceleration

Written on October 15, 2013 by Matt Bach
Share:

Introduction

In a previous article exploring GPU acceleration for Maya 2013, we looked at the performance of both desktop and workstation video cards. Almost all of the cards we tested were certified by AutoDesk but we found that workstation cards were by far a better choice than desktop cards.

While the AMD FirePro cards we tested are still current, NVIDIA has since released an entirely new line of Quadro cards. In addition, Intel has also released new CPUs that might affect our results. So in order to keep our benchmarks up-to-date, we want to benchmark Maya 2014 with the latest NVIDIA Quadro and AMD FirePro cards on the latest Intel CPUs and chipsets.

To benchmark these cards, we will be using the SPECapc benchmark for Maya 2012 on Maya 2014. This benchmark is a full system benchmark that tests 3D and 2D graphic performance in addition to CPU performance in multiple scenes with several model types including shaded, textured, and wireframe models.

Since visual aids are a great way to get a feel for a benchmark, here is single benchmark run using FRAPS that we recorded in our AutoDesk Maya 2013 GPU Acceleration article to help you get a feel for exactly what is tested by this benchmark:

Test Setup

To make sure that the chipset and CPU did not affect our results, we used two separate testing platforms consisting of the following hardware:


To test a wide spectrum of GPUs, we tested the following cards. At the moment, neither AMD nor NVIDIA has a specific driver or plugins available for Maya 2014 so just the base driver was used (320.86 for NVIDIA Quadro, 12.104.2 for AMD FirePro). Note that Maya 2014 does not currently support multiple GPUs, so only a single card was used in all of our testing.

NVIDIA AMD
Model Est. Street Price Model Est. Street Price
Quadro K600 1GB $195 FirePro W5000 2GB $420
Quadro K2000 2GB $430 FirePro W7000 4GB $670
Quadro K4000 3GB $765 FirePro W8000 4GB $1430
Quadro K5000 4GB $1700 FirePro W9000 6GB $3400
GeForce GTX Titan 6GB $1000


Maya was configured according to the SPECapc readme.txt with two minor changes in order to allow the benchmark to run on Maya 2014. These changes were a simple edit to the bat file (correcting the directory change to reflect the proper location) and manually copying the benchmark script to the Maya2014\scripts\others folder. Other than those two minor changes, nothing was needed to allow the SPECapc Maya 2012 benchmark to run on Maya 2014.

Results

Maya 2014 Graphics Benchmark

Individual test results
 

From an overall graphics standpoint, we found that AMD FirePro cards perform much better in Maya 2014 than their NVIDIA Quadro counterpoints. In fact, the only NVIDIA card that overall out-performed AMD was the Quadro K5000 which just barely beat the FirePro W5000. On an individual test basis, the Quadro K5000 was the top performing card in both the Shaded HQ and Textured HQ tests, but the other Quadro cards were all beaten by the AMD FirePro cards.

One interesting thing we found is that while the performance difference between the NVIDIA Quadro cards is pretty large, the AMD FirePro cards all received very similar results. So while it would be worth it to get a Quadro K2000 over a Quadro K600, justifying a FirePro W7000 over a FirePro W5000 is much more difficult. In fact, the FirePro W8000 actually scored worse than the FirePro W7000 in our testing.

While this article is about GPU acceleration, we also found some very interesting results in the CPU portion of the benchmark that we feel deserves to be mentioned:

Maya 2014 CPU Benchmark

The interesting point here is that all of the NVIDIA cards received higher CPU scores than the AMD cards. This deviation mostly occurred in the physics portion of the benchmark, but even the Render test performed better with NVIDIA. So while our testing shows that AMD FirePro is much better from a graphics standpoint, if you perform CPU-intensive tasks in Maya you may be better off with an NVIDIA card due to the higher CPU performance.

Conclusion

Our benchmarks clearly show that the AMD FirePro cards are the top performers in Maya 2014, but the results are muddied a bit by the fact that you get better CPU performance when using an NVIDIA card. Overall, however, we would still recommend using an AMD FirePro card over a NVIDIA Quadro card for Maya 2014.

When it comes to choosing which AMD FirePro card to use, we recommend either the W5000 or W7000 depending on your budget. While there are higher-end FirePro cards available, The W8000 actually performs worse than the W7000, and the W9000 simply does not give enough performance advantage to justify its cost.

Maya Workstations

Tags: Maya 2014, GPU Acceleration
Matt G

Why does the Titan fall so short? In AutoCad it seems to be amazing how fast it is... Driver issue? Test issue?

Posted on 2013-10-29 00:34:51
tricecold

Titan still is a consumer card I suppose, but from what I saw about card performances between applications totally depends on the chip manufacturer and software developers collaboration. I can get 20+fps with 10 million polygon in maya with an AMD HD 7950 and inside HOUDINI same card was 1+ fps in Houdini 12.5 + corruptions and glitches, and now in Houdini 13 it is a lot faster with the same card. So I guess the safe bet is to go with certified cards outlined at the softwares website.

Posted on 2013-11-22 17:36:14
clausbohm

One question is why was the saber tooth MB producing better results?

Posted on 2013-11-14 04:10:13
tricecold

This test is accurate, Firepro and consumer AMD cards works really good in Maya, as an FX TD for 3 years and a Generalist for 10 years , and lately using Houdini,here is my input,

For MAYA, as a freelancer an AMD HD 7950 with 3GB ram is perfect, I have this on my home machine and it runs better than its NVIDIA competitor, like GTX 760 or 770.

For HOUDINI, definitely NVIDIA, AMD drivers suck in Houdini, new release Houdini 13 is a ton better with AMD in terms of speed, but you still get corrupt text, etc.

If you use NUKE for comping together with Mari for texturing I suggest you go with an NVIDIA so you can take advantage of CUDA.

and lastly a request, please add Houdini 13 to these kind of tests, as it is now pretty much the mainstream software for FX artists. This was added to the previous article, so I added here also

Posted on 2013-11-22 17:32:18
Bern

So for Maya do you think an Asus Sabertooth X79 would be good?

Posted on 2014-03-20 14:05:39
tricecold

I don't think motherboards have any kind of effect on compatibility issues.

Posted on 2014-03-20 14:10:15
Bern

Thanks

Posted on 2014-03-21 00:51:50
Michelena

Any tests with crossfire pro? 2 W7000 maybe?

Posted on 2013-12-31 10:28:04

Unfortunately, Maya does not currently support multiple video cards in Crossfire or otherwise. So there is no performance gain at all by using two cards.

Posted on 2013-12-31 18:31:17
Michelena

https://m.youtube.com/watch...

This is the only evidence I've found of crossfire pro improving performance in Maya. But it's a bit old, different version of Maya and GPU.

I already own a W7000, was thinking in ordering a second to try it out. If it doesn't work, I can always return it. Unless you guys are 100% sure it doesn't work.

BTW, thanks for the article, I read the first version of it about a year ago. After reading it, I decided to go for the W7000 and it did improve compared to my GTX 670.

Posted on 2014-01-01 00:54:52

That appears to be a clip of just viewing an existing 3D model - which is actually a lot like what is going inside a video game (as an example) and thus I wouldn't be surprised if it benefited from something like Crossfire... just as games do. The testing Matt did in this article focused on Maya's ability to use the GPU to help improve performance in content creation, though: rendering, shading, etc. That is where we saw no performance difference... and that is what is more demanding, really. From our test results it looks like the W7000 you have should really perform well there.

Posted on 2014-01-01 01:04:11
Michelena

I'm a 3d animator. I spend 95% of the time playing and scrubbing the timeline. The remaining 5% it's only to export the final animation. I don't model, light, render, create shaders or mess with any of the dynamics engines. I usually work with very heavy models and playback is about 12 fps. Wondering if current generation of GPUs work as the previous, as shown in that youtube clip.

If it does, I'm getting two more W7000. That should be able to play animations in "real time" 24fps(final output). No more playblasts/previews =)

Posted on 2014-01-01 06:01:32

Hmm, given your situation it might be good to see if you can find any reviews focusing more on playback performance rather than GPU acceleration. We didn't look at that in this test (so far as I am aware - correct me if I am wrong, Matt) so we can't answer whether additional cards would help with your situation. It is also possible that a single, faster card might help more for that... but again, finding someone with direct experience in the same sort of situation you are in would be best.

Posted on 2014-01-01 19:44:33
Guest

From the Test Setup section of the article:

"Maya 2014 does not currently support multiple GPUs, so only a single card was used in all of our testing."

Posted on 2013-12-31 18:35:02
Scott B

This, and it's 2013 counterpart are fantastic write ups, thank you! I had previously been at the mercy of the 'Gaming cards are better and cheaper' crowd without seeing software specific data. I wasn't really in the market for a new video card but given the stability and performance improvements that even a FirePro v3900 would appear to give over my current, occasionally unstable GTX 470 I'm tempted to make the jump. If anyone could help with the following though that'd be great:
A typical project involves 40% Maya, 40% Photoshop, 10% Mudbox and 10% Lightroom - the (2013) benchmarks suggest a v3900 would be a decent improvement for Maya but I'm struggling to find anything on what the consequence would be for Photoshop? Any help would be greatly appreciated!

Posted on 2014-01-26 20:17:15

We haven't explicitly tested FirePro cards in Photoshop, but in general we found that a mid-range or higher card of any type is sufficient. There is little difference between GeForce, Quadro, and Radeon cards once you reach a minimal performance range:

http://www.pugetsystems.com...

Lightroom I believe uses the GPU even less, so I would likely make your decision based on Maya performance. And there is no need to get an older 3000-series card; just jump up to something like the FirePro W5000 and you should be in good shape :)

Posted on 2014-01-26 21:43:47
Scott B

That's great, thank you. I'll certainly have a think about the W5000 - it's a shame to see the UK price(£345/US$572) so much higher than the US but I guess that's nothing new! I presume a v4900 wouldn't really be worth it?

Posted on 2014-01-27 18:12:46

I would almost always avoid older generation cards, especially when looking at pro-grade cards. They don't update them terribly often, and when they do there is usually a good set of performance and feature differences.

Posted on 2014-01-27 19:02:34
Bonobolov3

This article really convinced me that I need a FirePro W5000, but they seem very hard to get. Are they being phased out or what is going on? You can't even find it when you search for it on pcpartpicker. Any other suggestions for graphic cards that will be equivalent to the w5000?

Posted on 2014-03-08 23:04:30

I don't think they are going away - or at least if they are, I have not heard or seen any evidence to that effect. Here is one supplier that lists 196 of them in stock:

http://www.provantage.com/s...

Posted on 2014-03-09 05:36:14
Bonobolov3

Wonderful! Thank you so much for the article and the reply. I cannot wait to get this card.

Posted on 2014-03-09 15:54:33
Bern

In terms of using a W7000 with a Sabertooth X79, how would the results be expected to stand up to what is shown here?

Also, one other question. I'm not overly well versed on specs and spec tests and was wondering what kind of difference there actually is between a score of 4.08 and 4.97? Thanks

Posted on 2014-03-20 06:42:34

The motherboard itself doesn't matter, it is the CPU and the video card primarily that contribute to the score. The motherboard does determine which CPUs you can use, though, with the Sabertooth X79 putting you in the Ivy Bridge E series. The results above were carried out on two CPUs, and one (a Core i7 4960X, shown as red in the graphs) was on the X79 platform. Interestingly that platform scored lower for the graphics based tests than the other we tested (the Z87 platform), but did better on the CPU-based testing. The latter result makes sense, since the 4960X is a six-core CPU, but it seems a little odd that the graphics performance was lower.

Posted on 2014-03-20 16:29:58
Bern

Okay, thanks. I'm trying to build a computer for the first time so this is all very new to me. If I understand things correctly than an i7-49030k CPU with a W7000 should do well with Maya?

Posted on 2014-03-21 00:54:56

Yes, that combo should do very nicely :)

Posted on 2014-03-21 00:58:53
Bern

Excellent, thanks again.

Posted on 2014-03-21 01:48:35
tricecold

Hi William, any chance on doing a similar test on Houdini, I am a convertee from Maya for 2 years now into Houdini, and it seems to me that there is a wild performance differences between softwares on same mesh or meshes with same cards on same systems. This could potentially improve the results not based on hardware power but also software compatibility.

Thanks

Posted on 2014-04-15 10:18:40

That would be up to Matt and our Labs department, but I know that one of the problems they run into with this sort of software is that benchmarks often don't exist. There needs to be some relatively easy way to run a repeatable set of tests to gauge performance, without needing to be an expert at using the software. If you know if anything like that for Houdini, please feel free to leave details here in the comments and they can take a look.

Posted on 2014-04-15 16:21:12
Giacomo Fratini

I would like to buy a new graphic card, but i haven't got enough money for a professional gpu (and I also have to work with other programs like Premeiere Pro cc, After effects cc ecc..) Do you think a gtx 760 will work fine with Maya or 3ds Max even if not perfect?

Posted on 2014-05-15 13:24:23
SoNic67

Quadro "new feature drivers" allow to dedicate cores just for graphics (instead of reserving some for CUDA). I wonder if that would make a difference.

Posted on 2014-06-10 15:25:53
Lee

It's confusing! Honestly doing a "pure graphic" test for Maya seems to me a bit out of place since Maya mainly uses CPU power.

You said NVIDIA cards generate more overall power in rendering than FirePro cards, but you didn't mention how much more. For if the difference is considerable I would be more inclined to get an NVIDIA card for redering, the main power-hungry task that Maya performs.

I assume the "pure graphic" test took into account only the viewport performance of which the importance comes second to rendering.

Posted on 2014-08-16 21:03:19
eodeo

Please try to avoid SPECheat test or at least don't rely solely on it. It is
designed specifically to cheat in favor of workstation GPUs by using obsolete runtime
that shows nothing of the modern DirectX implementation used across all 3 f the
Autodesk's major 3d DCC apps- Autocad, Max and Maya.

Posted on 2015-02-04 09:52:24

We would absolutely love to test using other benchmarks in addition SPECapc, and if you have any suggestions on what to use please let us know! Maya is actually a bit difficult for us because we are not as experienced using it as we would like to be, so doing something like developing our own benchmark is just a bit out of our reach right now. Not to say we aren't working on it (we definitely are) but any suggestions you might have would be extremely useful for us.

Posted on 2015-02-04 19:33:24
altiris28048

Thanks for this, in terms of rendering, is it better to use the CPU or the GPU? I am building a new PC for autodesk maya and I want to know whether I should spend more on the GPU or on the CPU...which one would be better at rendering?

Posted on 2015-05-03 22:44:08
Phil

Realtime rendering (what you see as you're working) primarily uses the GPU, while final rendering (for export) primarily uses the CPU (and loads of system memory). It would really depend on your workflow in that case. If you spend most of your day in the viewport then you'll see more gains from the GPU, but if you're constantly testing the final render of a scene then perhaps the other way around. Of course you can always batch render overnight if that works into your workflow.

In any case, a very good CPU is always beneficial and nowhere near as expensive as a good GPU.

Posted on 2015-12-01 12:51:24
Mohammed

I want custom build PC
the follow are hard ware configuration.
Please send me a suggestion:

CPU Core i7 - 5960X
Mother board: ASUS 99X Deluxe
32 GB RAM Crossaire
Vengeance - Corsair

Please give me some hints.
Thanks
Regards
Zahoor

Posted on 2015-05-05 07:51:38

If you contact us by email we can help recommend a system configuration from our website that would fit your needs. Please note that we are only able to sell / ship within the US, Canada, Mexico, UK, and Australia at this time. If you reside in one of those countries, please feel free to email our consultants at sales@pugetsystems.com

Posted on 2015-05-05 17:21:03
Phil

So Firepros offer better viewport performances, but Quadros team up better with your CPU? Hmm, that's interesting. I'm looking for a Maya 2016 specific setup to run on both Windows and Linux, but primarily Linux Fedora for work. But I take it, for rigging and character FX such as Hair & Cloth, which would fall into the more cpu related computing along with the physics category, Quadros would have the upper hand?

I was considering going for a Sabertooth motherboard with a Xeon 1650v3 (x99 lga2011-3) but I saw that the 6700K (z170 lga1151) gets actual better benchmarks for Maya than other processors I'm comparing it with e5-1650, e5-1620v3, 5820k (as well as saves some dollars on cpu and motherboard). It loses in multi-threading when matched up against an hexa core, which is normal, but in single core it achieves better results, namely with it's higher base clock speeds, this remains important for Maya in rigging and character fx such as hair & cloth. The new parallel rig evaluation along with the gpu overide are welcomed additions, beneficing from the multi-threads and the gpu to handle deformers, but as some rigs don't cooperate with the parallel evaluation, single core performance remains very important. At least for my situation where I'm not too focused on rendering.

I strayed off there a little, but back to GPUs, I was wondering which line you would recommend for use on Linux; I've read stories saying one is better and then vice versa. I was considering the W5100 vs the K2200 and I could be willing to go up to the W7100.

Thanks in advance,

Posted on 2015-11-23 02:52:43

Honestly, if it is a toss-up (or even somewhat close) I would go for a Quadro. The driver support is simply better (for both Linux and Windows) and NVIDIA as a whole makes much more reliable cards than AMD.

Posted on 2015-11-23 19:41:23
Phil

Thanks for your input, I'll keep this in mind

Posted on 2015-11-26 03:33:25
Luca Pupulin


It's time for another bunch of tests 😉
Even if SPECapc for Maya 2012 and the one included in the SPECviewperf 12.1 are a bit outdated 'cause they don't take into account the new feature set introduced in later versions of Maya,such as viewport 2.0 (VP 2.0), parallel rig evaluation,gpu override and more..would be nice to see an updated test and even better the same benchmark on both Windows and Linux.
Benchmarking Houdini too,would be great,but I know this is a more complicated task to perform.

Posted on 2017-03-16 21:31:56