MayaCarBench: real world performance in Autodesk Maya

Discussion in 'Benchmark Mayhem' started by NormanBates, Oct 24, 2009.

  1. NormanBates

    NormanBates Master Guru

    Messages:
    376
    Likes Received:
    0
    GPU:
    ATI HD3870 / ATI HD4670
    Hello World

    here I am again, worrying, as always, about the best way to run Maya as smoothly as possible

    but there's something different now: standard benchmarks won't cut it anymore

    it's not just a matter of whether or not the mods we currently have (for both ati and nvidia cards) are really working, but there's also the question of whether a real FireGL/Quadro makes any difference in the real world anymore: they certainly do much better in the usual spec benchmarks, but it doesn't seem easy to feel the difference in the real world

    this guy has done a great job of analysing performance in 3dsmax:
    http://area.autodesk.com/forum/auto...quadro-benchmarks-40actual-tests41/post-0/#p0

    but I use maya, so, after textbench and bikebench, here comes the new and greatly improved MayaCarBench

    It's a scene containing 11 models from gfx-3d-model.blogspot.com: 10 cars and an aircraft; with 1.3 million triangles, and some nice textures, it's definitely able to bring my rig to its knees, both with the mod and with the standard catalyst drivers

    here it is:
    http://uploading.com/files/8c25edd2/MayaCarBench.zip/

    if you want to use it:
    * download zip
    * extract it to a local drive, it will create a project subdir tree
    * set 1280x1024 resolution
    * open maya
    * if you don't see it already, get the fps indicator: display > heads up display > frame rate
    * file > project > set > look for the MayaCarBench directory that came out of the zip
    * file > open scene > MayaCarBench.mb
    * you should see ten cars and an aircraft, plus some text disclaimer on the origin of the models, all shaded and with textures
    * press play; the camera will rotate around the models
    * look at the fps indicator, and take note of the HIGHEST fps number achieved (ok, if it's a clear outlier, please discard that one and get the highest non-outlier)

    let's see if this can help us find an answer to questions like these:
    * do the super-expensive FirePro / FireGL / Quadro cards make sense anymore?
    * are the mods we have really working?
    * which is best, nvidia or ati?


    results so far:
    Code:
    n4d444 ------ 9.6 fps -- Q9550@3.6GHz - 8GB - xp.64 - GF8800 GT @ Quadro FX3700
    kocha ------- 9.2 fps -- Q8200@3.2GHz - 4GB - xp.64 - real Quadro FX4600
    SCHOKO ------ 8.6 fps -- T7700@2.4GHz - 4GB - w7.64 - HD2600M@FireGL - BUG SOLVED!!
    kocha ------- 8.0 fps -- Q8200@3.2GHz - 4GB - linux - real Quadro FX4600
    kocha ------- 8.0 fps -- Q8200@3.2GHz - 4GB - linux - real Quadro FX1700 512MB
    kocha ------- 8.0 fps -- Q8200@3.2GHz - 4GB - linux - geforce 9500GT unmodded
    kocha ------- 7.4 fps -- Q8200@3.2GHz - 4GB - xp.64 - real Quadro FX1700 512MB
    Maryus3D ---- 7.0 fps -- q6600@2.4GHz - 8GB - w7.64 - GF8800 GT @ Quadro FX3700
    wpgscorpion - 6.8 fps - i7-860@2.8GHz - 8GB - w7.64 - GF9800GT 512MB
    n4d444 ------ 6.4 fps -- Q9550@2.8GHz - 4GB - xp.64 - real Quadro FX1700 512MB
    CADCAT ------ 6.0 fps ---- C2D@3.8Ghz - 4GB - xp.64 - HD4870 unmodded
    CADCAT ------ 6.0 fps ---- C2D@3.8Ghz - 4GB - xp.64 - HD4870 @ FirePro
    CADCAT ------ 6.0 fps ---- C2D@3.8Ghz - 4GB - xp.64 - FirePro v7700
    Maryus3D ---- 5.7 fps -- q6600@2.4GHz - 8GB - xp.64 - GF8800 GT @ Quadro FX3700
    kocha ------- 5.2 fps -- T7500@2.2GHz - 4GB - xp.32 - real Quadro NVS 320M (NVS!)
    Maryus3D ---- 4.7 fps -- q6600@2.4GHz - 8GB - w7.64 - GF8800 GT @ GeForce driver?
    Glidefan ---- 4.5 fps -- ???????????? - 4GB - ????? - geforce GTX260 unmodded?
    NormanBates - 4.4 fps -- q6600@3.0GHz - 2GB - xp.32 - HD4670 1GB unmodded 8.612
    NormanBates - 4.4 fps -- q6600@3.0GHz - 2GB - xp.32 - HD4670 1GB unmodded 8.612, clocks down by 50% (GPU/GPURAM)
    NormanBates - 4.4 fps -- q6600@3.0GHz - 2GB - xp.32 - HD4670 1GB modded 8.603
    NormanBates - 4.3 fps -- q6600@3.0GHz - 2GB - xp.32 - HD4670 1GB unmodded 8.612 - system RAM 667MHz (instead of 800MHz)
    NormanBates - 3.7 fps -- q6600@2.4GHz - 2GB - xp.32 - HD4670 1GB unmodded 8.612
    



    WHAT WE'VE LEARNED: in this scene...
    * NVIDIA is faster than ATI
    * apart from that, the CPU is the only bottleneck: faster GPU, pro drivers, real pro card... they don't make a difference
    * vista and w7 suck (with respect to xp)


    sister benchmarks:

    MayaBikeBench:
    http://forums.guru3d.com/showthread.php?t=307873

    MaxCarBench:
    http://forums.guru3d.com/showthread.php?p=3338059
     
    Last edited: Nov 16, 2010
  2. Maryus3D

    Maryus3D Active Member

    Messages:
    97
    Likes Received:
    0
    GPU:
    Zotac GTX 285
    I'll try in windows xp with the modded driver for Quadro and win7 with a clean Quadro driver which has the same performance as a geforce :(. My rez is 1680x1050 22 inch monitor :).
     
  3. NormanBates

    NormanBates Master Guru

    Messages:
    376
    Likes Received:
    0
    GPU:
    ATI HD3870 / ATI HD4670
    thanks, I'm really interested in the performance of those cards
    I guess the resolution shouldn't be a problem as long as it's not too different, say 800x600 or 1920x1200
     
  4. Maryus3D

    Maryus3D Active Member

    Messages:
    97
    Likes Received:
    0
    GPU:
    Zotac GTX 285
    I'v got 4,7 under win7 and 5,7 under win xp with the mod. Well the scene won't let me to zoom only rotate donno why. My Q6600 is at 2,4 GHz and rez 1680x1050.

    EDIT: I'v solved the problem :D. It is slow :(. Reminds me of my island scene mayascen2 :bawl:
     
    Last edited: Oct 25, 2009

  5. NormanBates

    NormanBates Master Guru

    Messages:
    376
    Likes Received:
    0
    GPU:
    ATI HD3870 / ATI HD4670
    thanks, I posted your results in the table above

    your gpu smokes mine in raw power and games, and it does so also in MayaCarBench :)
    http://www.tomshardware.com/charts/...arts-2009-high-quality/compare,1538.html?prod[2823]=on&prod[2872]=on

    but we did learn one thing: in line with the 3dsmax review above that found wxp much faster than vista, we now know that in maya it's also best to stick to xp, even compared with w7


    don't cry, it's just a hell of a scene; just more realistically so than the ones we used in previous benchmarks
     
    Last edited: Oct 25, 2009
  6. kocha

    kocha Member

    Messages:
    12
    Likes Received:
    0
    GPU:
    NVidia Quadro FX4600
    MayaCarBench on real Quadro FX4600

    Hello to all!

    I've tested my pc with MayaCarBench and got 9.2

    Maya version 2010 64bit, XP64, 1280x1024.

    My nvidia drivers are set to "Maya 2009 Stereo" profile, with overlay off.

    I hope this throws some light on question "real quadro vs fake quadro"...
     
    Last edited: Oct 26, 2009
  7. NormanBates

    NormanBates Master Guru

    Messages:
    376
    Likes Received:
    0
    GPU:
    ATI HD3870 / ATI HD4670
    wow, that's an impressive result for a card that's basically the same as Maryus3D's

    it can be the drivers, or maybe the memory bandwidth (it can't be the amount, I think yours has 768MB, does it?)

    but it looks like it's most probably the drivers; impressive
     
  8. Maryus3D

    Maryus3D Active Member

    Messages:
    97
    Likes Received:
    0
    GPU:
    Zotac GTX 285
    Well :D the Q8200@ 3,2 is faster then my Q6600 stock and FX 4600 is faster the my FX3700 like 8800GTX > 8800GT :D. Overall nice results :).
     
  9. kocha

    kocha Member

    Messages:
    12
    Likes Received:
    0
    GPU:
    NVidia Quadro FX4600
    Another test with even older Quadro FX 1700 512 MB on the same computer gives me 7.4 FPS.... Not bad at all, better performance/price ratio than FX 4600 :)

    When I choose "Base profile" settings on Quadro FX 1700 result is 7.1, with "Maya" profile = 7.4

    Some nvidia driver settings are not "published" through Control Panel, but do exist in registry (for example OGL_App_SupportBits). It is very difficult to trace the meaning of each bit in that value, I am not an expert, but I've noticed that every application profile has its own value...

    One more thing... I can't find application profile settings in nvidia linux drivers, so I am "convicted" to base profile settings. Benchmark results of FX 4600 in linux are worse than in windows (just 8)... but guess what FX 1700 in Linux scores better (8 = the same as FX 4600).

    Conclusion: For Maya on Linux : cheap original Quadro = expensive original Quadro !!??!!??
     
    Last edited: Oct 26, 2009
  10. Maryus3D

    Maryus3D Active Member

    Messages:
    97
    Likes Received:
    0
    GPU:
    Zotac GTX 285
    Because Maya uses only one core from CPU and the GPU is limited :(.
     

  11. NormanBates

    NormanBates Master Guru

    Messages:
    376
    Likes Received:
    0
    GPU:
    ATI HD3870 / ATI HD4670
    wow, still amazing results: that FX1700 should be definitely slower than a 8800GT: it's the pro version of a really slow 8600GT, it has less raw power, less memory, less bandwith, less everything --> it can only be the drivers!

    and the linux results are also interesting, but my main concern there would be: some people said, long time ago, that in linux and mac drivers for Quadro/FireGL cards were exactly the same as for their gaming counterparts --> can you test a geforce card in linux?



    what's in a quadro, and in a geforce:
    http://en.wikipedia.org/wiki/Nvidia_Quadro
    http://en.wikipedia.org/wiki/Comparison_of_Nvidia_graphics_processing_units
     
    Last edited: Oct 26, 2009
  12. kocha

    kocha Member

    Messages:
    12
    Likes Received:
    0
    GPU:
    NVidia Quadro FX4600
    Yes, both Nvidia and Amd (Ati) pack both non-workstation and workstation Linux drivers in one large archive...

    I can test Gainward GeForce 9500 GT 512 MB in Linux, I will be back with results in 30 minutes :)

    tick tack tick tack ... 30 minutes passed :)


    Thank You Norman for your suggestion to bench "good old" 9500GT, because I've found out something very interesting.

    9500GT 512 MB + Centos 64bit Linux 5.3 + Maya 2009 + MayaCarBench ...
    8 fps (like both Quadros I've tested)

    With Backface culling enabled score is even better = 8.7 :)

    What to say, I am confused. Maya only users, dont throw your GeForce, try Linux :tux:
     
    Last edited: Oct 26, 2009
  13. NormanBates

    NormanBates Master Guru

    Messages:
    376
    Likes Received:
    0
    GPU:
    ATI HD3870 / ATI HD4670
    ok, so linux is the promised not-even-a-mod-is-needed land... as long as you can live with it

    amazing

    edit: yes, truly amazing
     
    Last edited: Oct 27, 2009
  14. Glidefan

    Glidefan Don Booze Staff Member

    Messages:
    12,481
    Likes Received:
    51
    GPU:
    GTX 1070 | 8600M GS
    You sure? i can see my cores going back and forth all the time with maya.
    4.5fps but at 1680*1050 with a 260
     
  15. Maryus3D

    Maryus3D Active Member

    Messages:
    97
    Likes Received:
    0
    GPU:
    Zotac GTX 285
    Yes I'm sure :p.
     

  16. NormanBates

    NormanBates Master Guru

    Messages:
    376
    Likes Received:
    0
    GPU:
    ATI HD3870 / ATI HD4670
    are you using v.2010? it would be nice if they had changed this (though I also doubt it)

    I add your results to the table, with ??? so far for OS and processor, and assuming you didn't use modded drivers
     
    Last edited: Oct 27, 2009
  17. kocha

    kocha Member

    Messages:
    12
    Likes Received:
    0
    GPU:
    NVidia Quadro FX4600
    True, Maya use only one core for viewport rendering, but use all cores for final rendering. I've tried to set affinity of maya.exe process to CPU 0 only and got the same bench results as earlier.

    During the MayaCarBench, that one core is nearly 100% used. That could be the problem if we talk about MayaCarBench "reliability" as graphics card bench. Because of high CPU usage whole test is very CPU dependable. I suppose the main goal is to evaluate GPU effect on viewport performance.

    Maybe this is main reason why I get the same result with 3 different cards in Linux :confused:
     
  18. kocha

    kocha Member

    Messages:
    12
    Likes Received:
    0
    GPU:
    NVidia Quadro FX4600
    MayaCarBench on my laptop

    HP 8710p notebook, C2D T7500@2.2 GHz, 4GB, Quadro NVS 320M

    XP32, Maya 2010 32bit, 1680x1050 resolution

    5.2 FPS (5.6 with Backface culling enabled)


    This notebook Quadro is not like real workstation Quadro, even my control panel is "GeForce like" without workstation options. Technology is the same as GeForce 8400GS.
     
  19. NormanBates

    NormanBates Master Guru

    Messages:
    376
    Likes Received:
    0
    GPU:
    ATI HD3870 / ATI HD4670
    ok, there must be something wrong here if a quadro NVS with 320MB of memory and running on the slowest CPU in the table can score higher than any non-quadro thing we tested

    possible explanations:
    * drivers are still very important, even quadro NVS ones have the necessary bits turned on
    * I'm dumb and I created a scene that needs so much memory in the card that even 1GB is not enough and we're just meassuring the GPU-to-system-RAM performance; I'll try to investigate that

    in any case, I still think it's a useful benchmark: it's a scene with 1.3M triangles and real textures, so it's close to what many of us would find in real world

    if the conclussion is that under these circumstances Maya is so CPU dependant that it doesn't matter which GPU you are using, then it means I don't want to spend big $$$ on a quadro card, maybe I'm just fine working in linux with a 8800GT, or, if I find linux too cumbersome (as I do) I can just decide between that same 8800GT with modded drivers or a real Quadro FX1800, but I won't go above that, and I will make sure I get a really fast CPU

    it would be a nice conclusion, but I think the jury is still out on this one...

    now, if the problem is that I used a scene with too many MB of textures, then I'm dumb because you could always work with slower res textures and then swap them before the final render

    EDIT: one way to see if the main performance driver is the CPU would be to downclock it and test again; if access to the system RAM is the bottleneck, playing with the clocks there should tell us something; my 2 systems are busy until thursday, I'll test all this then
     
    Last edited: Oct 27, 2009
  20. Maryus3D

    Maryus3D Active Member

    Messages:
    97
    Likes Received:
    0
    GPU:
    Zotac GTX 285
    I don't understand why Autodesk live the same poor optimization for maya since 1996 or so :(. There is a lot of powerful hardware out there and maya can't use it at all.
     
    Last edited: Oct 27, 2009

Share This Page