Are quadros and firepros still important?

[Ognyan]

Ever since 2010 when nvidia and ati decided to cripple gaming drivers in favour of ¨¨pro gpus¨¨ i started using both firepros and quadros.Yes,performance for me was excelent being able to move 30-40 millinos of polys in mayas viewport but in a week or two i am going to build my next workstation and just paused to think.I havent used gaming gpus in maya since a long time and today we have a new maya vieport,new gaming gpus.

Are they capable in handling a large ammount of polygons(good fps) in maya like the pro ones or are the drivers still not up to the task?

Currently i have a k4200 and can push 40m polys in maya at 35-40fps.Was planing on upgrading to a p5000 but im not sure if something like a gtx1080 will manage to do the same job?

[Ognyan]

Bump for the people using gaming gpus.

[Nadja.Bueckmann · Accepted solution]

Hi @Ognyan,

I can´t telll you which gpu to buy but we have a hardware recommendation for all our products.

For Maya 2017 you can find it here: Maya Certified Graphics Hardware 

I think it´s interesting to look it up for older Maya Versions as well, as that can give you some information which are the latest ones as you probably will buy a card you want to use for later Maya versions as well.

Hope that helps,

Best, Nadja

[Ognyan]

Yeah,I have seen the hardware recomendations.In the beginig I remember that gaming gpus were not supported and by todays stnadart they appear as certified.Thats why I was wandering if gaming gpus drivers are finally the level of qudaros/firepros?

I dont have a gaming gpu to answer this for my self,so if someone pushes a lot of polys/animation in maya using a modern gaming gpu,I woud be very interested in knowing how is his experience.

[Ognyan]

Found a review between quadro and gtx gpu but is based on max 2017.From what i see there is no difference in performance between quadros and gaming gpus there.

I know max is not like maya but if im not mistaking max has gone to direct3d and i think maya is using it too so if im not mistaking,performance shoud be similar.

So coud someone with maya and a gaming gpu please tell me if he is getting good framerates on the viewport with lots of polygons?I suspect most people on the forums have gpus in their computers 

Here is the max review if someone is interested:

https://www.pugetsystems.com/labs/articles/AutoDesk-3ds-Max-2017-Quadro-GPU-Performance-822/

https://www.pugetsystems.com/labs/articles/AutoDesk-3ds-Max-2017-GeForce-GPU-Performance-816/

[Ognyan]

Bump on this thread.I asked a friend to try a scene of 13 milion polys on his maya(he is not a advanced user so he just moved in the vieport) and it looks like his gtx 1080 was below my k4200.Ive found a interesting thread here too

http://www.tomshardware.co.uk/answers/id-3331673/gtx-1070-performance-maya-buying-card.html

Which say that a gtx1070 is no good in maya 2016 sp6.Im still searching for info on if gaming gpu are good for smooth multi million poly workflow on maya...

I see people claiming gaming gpus are equal to quadros in maya but if im not mistaking they are placing those claims on tests like spec view and that is just a heavy scene in which the camera rotates around a heavy mesh and that is all.That is waaaay different than selectin the faces of that mesh,extruding it and all type of manipulation made to it.

So as im still puzzled,i ask humbly,people with modern gaming gpus,how is your maya performance?Is it worth it to go to gtx or not.

[Anonymous]

I don't have any scenes of that size, so possibly if you could share a scene that you typically see, I could give you the rundown on a 970 and a 1070 in viewport performance, using Legacy GL and the new GL Core. I can't test DirectX as I'm on Linux.

I can always replicate a cube until I get into the mid millions, but I'm not sure if that would be the best case test against a production scene.

[mspeer]

Hi!

There is just 1 question:

Which card do you need to be able to make your projects (in time) ?

No one in this forum can answer this. No one knows what scenes, settings and tools you typically use and no other one has the same computer like you or uses the same workflow, but this can have a big impact on the overall result.

Keep in mind that the "Pro" cards are solely developed for professional solutions, so in general they provide better performance and feature support for some applications (like Maya). From my personal experience i can say that the difference in performance can vary from drastic to nearly non existent, based on graphic card, scene and current task.

In the end you need a card that fits your needs and not the needs of someone else, so you have to do tests on your own if you want a result that can help you with this decision.

[Ognyan]

Mike,that woud be very helpfull.I will upload a scene in a day or two of a diablo type gate(13million triangles).

Cheers guys 

[Ognyan]

Here are 2 scenes(they are the same )of a mesh that is going to zbrush.(13million triangles)

One of the scene is the low poly but i have all of the meshes ¨¨temporal smoothed¨¨(the 3 key)

And the other scene is your normal final smooth version.

Its a maya 2012 x64 file(if you get a error it will be because i didnt provided the image plane reference)

https://ufile.io/c269f                      

Only 32mb rar.

Please try to navigate the scene and edit a litle to see if its smooth on your gpus model.

----------------------------------------

In my case on a quadro k4200

Scene ¨¨smoothed¨¨

Orbiting the camera in the Default quality rendering viewport i get 57fps.
Orbiting the camera in the viewport 2.0 i get 90fps

Scene ¨¨temporal smooth¨¨
Orbiting the camera in the Default quality rendering viewport i get 20fps.
Orbiting the camera in the viewport 2.0 i get 89-92fps

Editing the meshes in the two scenes (and in the old and new vieport) i get instant results(extruding,vertex selection/moving)

I hope someone with windows can do the test too.

My suspicions are that normal gtx do navigating the scenes just as smooth as quadros but when it comes to manipulating the heavy meshes it shoud lag.

We will see now.

Please dont redistribute the file.

Thanks guys 

[Ognyan]

By the way, I encourage more people to do the test(navigate the scene and edit some polys)so that we can have some final base idea where the gtx and quadros compete if( they even do and save us some cash investment in the future).

Cheers guys!

[Anonymous]

Hey this is a cool test! 

#----------------------------------------------------------------------------------------------------#

In my case on a 980 TI 

Scene ¨¨smoothed¨¨

Orbiting the camera in the Default quality rendering viewport: 7 - 10fps.
Orbiting the camera in the viewport 2.0 (AA on and set to defaults of 2 samples) I get 230fps

    Animated bricks moving (static camera): 214 - 230fps

    Animated bricks moving (while orbiting camera real-time): 198 - 218fps

     With screen-space AO(16samples):

             Orbiting the camera: 75 - 90fps

             Animated bricks moving (static camera): 60-65fps

             Animated bricks moving (while orbiting camera real-time): 56 - 61fps

In VP2.0 I'm getting instantaneous feedback on all modelling operations

#----------------------------------------------------------------------------------------------------# 

Scene ¨¨temporal smooth¨¨
Orbiting the camera in the Default quality rendering viewport: 4.6 - 5.1fps.

Orbiting the camera in the viewport 2.0 (AA on and set to defaults of 2 samples) I get 235 - 250fps

    Animated bricks moving (static camera): 215 - 225fps

    Animated bricks moving (while orbiting camera real-time): 196 - 205fps

     With screen-space AO(16samples):

             Orbiting the camera: 75 - 90fps

             Animated bricks moving (static camera): 60-65fps

            Animated bricks moving (while orbiting camera real-time): 60fps

In VP2.0 I'm getting instantaneous feedback on all modelling operations

 #----------------------------------------------------------------------------------------------------#

It looks like default viewport, the performance of my 980ti is much slower than your Quadro, but in VP2 it does just fine! 

[Anonymous]

Nice results! I'm going to be running the test probably tomorrow, but I expect my 1070 to be probably a bit underneath the 980 Ti. I could be wonderfully surprised, however.

That being said, when you ran the tests with AO, did you have AA on as well? AO seems to be a VP2 killer here on Linux, but I haven't done a lot of stuff besides Mari with my 1070, as my 970 is usually driving display. And in general applying both simultaneously can tax a system more than you'd think. This is obviously depending on the scene, like an arm with AO/AA would do just fine while a city block might suffer.

[Anonymous]

Nice! I'm excited to see!

I should have made that clearer - AA was always on for all the VP2 tests, even when I turned AO on. 

[Ognyan]

Here is some more info from another user from youtube.Not sure how is poly manipulacion as he didnt try it.

Quadro K5000 Maya 2017

Smoothed 

  Viewport 2.0: 125fps; 

  Legacy 15fps

Temporal 

  Viewport 2.0: 120fps 

  Legacy: 46fps

GTX 1080 Maya 2017

Smoothed 

   Viewport 2.0: 125fps; 

   Legacy 4.8fps

Temporal 

   Viewport 2.0: 125fps 

   Legacy: 33fps

Goof info guys!James,i supose that 980ti is in windows using the dx11 shader option?For now it looks like openGL/CL or what ever maya uses there is cripled on gtx?If we go with a gtx vp 2.0 is a must?Just hope it supports most of the basic stuff.Im sooooo glad to know that with a gtx 980ti on that heavy scene in vp2 you are achiving instant modeling results and no lag.In the end i might save on money.Lets see what other users report.

Cheers 

Curious to know why his gtx1080 is slower? than the 980ti.Max 125fps,is it a monitor refresh rate or something that is limiting the situation?James,what cpu do you use and what is your monitor refresh rate set to?(or did you disabled vsinc on the gpu?)

[mspeer]

Hi!

I still miss a lot of information.

1. One you recognized on your own, Monitor Refresh Rate, Sync Settings and of course Resolution.

2. CPU, Mainboard (GPU connection speed)..., RAM , VRAM (Size and Speed)?

3. What happens to frame rate if all objects are selected (far more complex than simple smooth shaded)?

4. What happens to frame rate if some or all objects are transparent and please use a better Transparency Algorithm for Viewport 2.0 than the default one?

5. What if you have many large textures and transparent objects and all of them selected and moving at the same time?

6. What is with FX/Volumes like Particles, Maya fluids and Bifrost?

7. What frame rate can be achieved with animated deformed meshes?

8. What frame rate can be achieved with animated "realtime displaced" meshes?

9. Tests with Hardware Rendering or Software Rendering for renderer with GPU support like Mental Ray?

Until now i can't see any applicable information for an overall decision.

BTW, don't expect the cards to behave linear, for example the frame rates for an easy scene with low quality settings and a very complex scene with high quality settings can be for a gaming card 150 fps / 15 fps and for a pro card 90 fps / 50fps.

[Anonymous]

I just want to say in advance, I agree with everything @mspeer has said so far. This is a very static scene, but if that's what you deal with then by all means ask about it.

Also, for some reason I was not able to get a frame-rate out of Legacy viewports, so I don't have that. Only general feel. For both cards, Legacy Default (didn't test High Quality) was nearly unusable. Camera would have considerable jumps when viewing the whole scene, maybe 7-20fps? The 1070 fared slightly better, but only by a small margin. Frame rate would improve the closer you got to the model or individual pieces. The only different result here was the 1070 in the Smoothed scene, that one had a visible improvement to frame-rate, but was still choppy. That being said, general modeling and manipulations was good enough. With the camera positioned to see the whole scene, moving objects wouldn't be buttery smooth, but enough to not rip your hair out. Getting up close to the model made that a non-issue. General modeling was pretty decent as well, but beveling had a bit of lag to it. Extruding was fine.

Modeling/Moving objects in VP2 was buttery smooth for both cards. Didn't have to get close up or anything, just worked.

i7-6850K (3.6-3.8GHz)

GTX 970 SC 3.5GB

GTX 1070 FE 8GB

BenQ BL2710PT 2.5K Screen (2650x1440) 60Hz over Dual-link DVI-D.

CentOS 7 (1611)

Driver 375.26

Maya 2016 Ext 2 SP2

OpenGL - Core (Compatibility)

* Maya was full screen with the Time Slider, Range Slider, Command line, Help line, Shelf, Menubar, and Channel Box open. So viewport was very large.

Results (All ranges are between having the whole scene centered in the viewport to getting up close to small groups or individual pieces of the scene):

GTX 970

Temporal:

    VP2: 110fps - 145fps (Mostly around 128fps - 135fps)

    VP2+AO: 45fps Full scene, 70fps close up

Smoothed:

    VP2: 110fps - 145fps (Mostly around 120fps - 130fps)

    VP2+AO: 45fps Full scene, 70fps close up

GTX 1070 FE

Temporal:

    VP2: 125fps - 145fps (solid 130fps - 135fps)

    VP2+AO: 80fps Full scene, 95fps - 105fps Medium, 120fps close up

Smoothed:

    VP2: Solid 130fps

    VP2+AO: 83fps Full scene, 94fps Medium, 120fps close up

Side question, @Anonymous: What monitor do you have, and what is it's rated refresh rate?

[Anonymous]

For small scenes and static scenes, you would not see any noticable difference, it's when you work with large scenes.
That's when you notice, the quadro cards push more geometry very easily.
Another thing, not everything works on a gamers card in OpenGL mode, in DX11 mode, there are even more disadvantages.

My advice, if you mainly work with small scenes ( product rendering or low poly ) a gamerscard will do just fine, but if you work with large scenes
and you want to use an AA of 16 , AO of 32 , DOF, 32 bit output, So basicly - use the viewport as your render engine. You can't do without a Quadro or Firepro. ( Nothing beats a 1 sec render ) - Mostly used for cartoon or cheap rendering.

Also keep in mind : Mudbox is still opengl and a quadro card will push more geometry, ( I believe Z-brush has an DX11 mode ?????)
Not sure about Z-brush though ....

[Ognyan]

Good info guys and thank you.I shoud have probably stated what i use maya first, but the fingers typing were faster than my exact thought.

-low poly modeling(game stuff)

-high poly modeling for zbrush(as the example scene of 13m polys)

-many times i even tend to import heavy(decimated)meshes from zbrus in to maya as i like the normal map-----but will probably change to mighty bake as its multithreaded unlike mayas solution.

-camera mapping

-and sometimes i like doing more complex model design only in maya(think a marine star craft suit) so it coud get heavy before i start cleaning the scene and than end with a 100k/200k textured model with 4x2k maps and rig and composit the thing.

-general riging and some animation retargeting

-unreal engine 3(will move to 4 in the future but not now as i have to finish the started game project)

And i plan on starting on using cuda stuff like octane or redshift(im aware what they want)

The new workstatition will be a atx motherboard 2011-3 

cpu  e5 1660 v4  8 core   with 128gb ecc

or a i7 6950x with regular 128gb ram.

2x ssd 

Looking at this thread and the results i suspect that as my tasks evolve over time i will probably need to find that sweet spot.I have the feeling that something like a gtx980 - 1080 will work great but i will probably reach the limit in x time.Ive been speaking with a reseller here and was looking at the new quadro p4000 gpu.A striped down version of the gtx 1070 but probably just as powerfull as a gtx 980 in gaming power(yeah i use game editors too and tend to jump a lot between apps).I can get 2 for 800€ each >one for maya and the other one to help with the octane render.

Or get 2x1080 for the same purpose but with the potential to reach a wall sometime in time due to changing workflows and stuff.

What is your opinion on this?

[Ognyan]

Zbrush is cpu bound only.

--------------------------------

And a small bump before the thread disappear to page 2 and i will be pulling the trigger in 3-4 days on this.

Cheers and thank you so much.