Loading...
> >

OptiX, use it or not?

simonbaush_6f85df6f17simonbaush_6f85df6f17 Posts: 5
in The Commons

Hi folks,

I noticed that when OptiX Prime Acceleration is checked, the GPU load goes down and the 'live' render in Daz does not seem to speed up. Am I missing something? The scene is a single clothed model with hair, loaded on the video card memory. Should I keep it checked or not?

My modest setup: i3 8100 CoffeeLake, 8 GB RAM, MSI GeForce 1050 TI 4GB

Thanks in advance

 

 

Comments

  • KitsumoKitsumo Posts: 1,216

    It works for some and not for others. A lot of people report that memory usage goes up, but render times are shorter. I'd say your best bet is to try a benchmark with it on and off and go by that. You can check the log file to see what your render time is. I usually keep mine off.

  • simonbaush_6f85df6f17simonbaush_6f85df6f17 Posts: 5

    Thank you for the reply Kitsumo, I'll do as you recommend and find out.

  • algovincianalgovincian Posts: 2,618

    Optix Prime replaces a portion of the Optix API, and when used it creates some additional data structures at the beginning of rendering which increase processing speed. The performance increases come at a cost of some memory (how much is scene dependent) and some flexibility as they are low level calls with some of the bells/whistles of the general Optix library no longer available. I believe it's very hardware and driver sensitive which, along with the scene dependence, accounts for the wide range of reported results by end users.

    - Greg

  • linvanchenelinvanchene Posts: 1,382
    edited October 2018

    The answer for Pascal cards (1050, 1060, 1070, 1080, 1080 Ti) is as other posters allready explained.

    For Turing cards like 2070, 2080, 2080 Ti  davemc0 shared the information that Optix Prime Acceleration is not supported for RTX based GPU.


    Another thing to note is that OptiX Prime will not be accelerated on RTX; only OptiX will. This is significant because Iray uses OptiX Prime, not OptiX. Thus, Iray does not get RTX ray tracing acceleration on my RTX 2080, and something big will have to change before it does get accelerated. I don't know whether Nvidia will port Iray to OptiX, which would be a big effort, or whether that will be done by MI Genius or Lightwork Design, or it will just not happen. Another possibility is for some party to implement OptiX Prime on top of OptiX to get access to RTX hardware acceleration.

     

    compare:

    http://www.daz3d.com/forums/discussion/comment/3993836/#Comment_3993836

    - - -

    I can confirm that with OptiX Prime Acceleration On the render time is actually a few seconds slower and not faster.

    1x RTX 2080 Ti
    OptiX Prime Acceleration Off
    4 minutes 37.84 seconds
    OptiX Prime Acceleration On
    4 minutes 44.83 seconds

    - - -

    2x RTX 2080 Ti
    OptiX Prime Acceleration Off
    2 minutes 36.17 seconds
    OptiX Prime Acceleration On
    2 minutes 36.98 seconds

    I have not yet tested how the time difference is with an hour long render.

    Still, I have not found any possible advantage of having Optix Prime Acceleration On for Turing cards.

    For Turing cards it may be the right choice to turn Optix Prime Acceleration off

    - - -

    If someone else got different results with Turing please share...

    - - -

    Included this information in Request #283430

    - - -

    Post edited by linvanchene on
  • AalaAala Posts: 140
    linvanchene said:

    The answer for Pascal cards (1050, 1060, 1070, 1080, 1080 Ti) is as other posters allready explained.

    For Turing cards like 2070, 2080, 2080 Ti  davemc0 shared the information that Optix Prime Acceleration is not supported for RTX based GPU.


    Another thing to note is that OptiX Prime will not be accelerated on RTX; only OptiX will. This is significant because Iray uses OptiX Prime, not OptiX. Thus, Iray does not get RTX ray tracing acceleration on my RTX 2080, and something big will have to change before it does get accelerated. I don't know whether Nvidia will port Iray to OptiX, which would be a big effort, or whether that will be done by MI Genius or Lightwork Design, or it will just not happen. Another possibility is for some party to implement OptiX Prime on top of OptiX to get access to RTX hardware acceleration.

     

    compare:

    http://www.daz3d.com/forums/discussion/comment/3993836/#Comment_3993836

    - - -

    I can confirm that with OptiX Prime Acceleration On the render time is actually a few seconds slower and not faster.

    1x RTX 2080 Ti
    OptiX Prime Acceleration Off
    4 minutes 37.84 seconds
    OptiX Prime Acceleration On
    4 minutes 44.83 seconds

    - - -

    2x RTX 2080 Ti
    OptiX Prime Acceleration Off
    2 minutes 36.17 seconds
    OptiX Prime Acceleration On
    2 minutes 36.98 seconds

    I have not yet tested how the time difference is with an hour long render.

    Still, I have not found any possible advantage of having Optix Prime Acceleration On for Turing cards.

    For Turing cards it may be the right choice to turn Optix Prime Acceleration off

    - - -

    If someone else got different results with Turing please share...

    - - -

    Included this information in Request #283430

    - - -

    What scene have you tested your cards with? I'm curious to know how that stacks up with the 1080 Ti.

  • linvanchenelinvanchene Posts: 1,382
    edited October 2018
    Aala said:
    linvanchene said:

    The answer for Pascal cards (1050, 1060, 1070, 1080, 1080 Ti) is as other posters allready explained.

    For Turing cards like 2070, 2080, 2080 Ti  davemc0 shared the information that Optix Prime Acceleration is not supported for RTX based GPU.


    Another thing to note is that OptiX Prime will not be accelerated on RTX; only OptiX will. This is significant because Iray uses OptiX Prime, not OptiX. Thus, Iray does not get RTX ray tracing acceleration on my RTX 2080, and something big will have to change before it does get accelerated. I don't know whether Nvidia will port Iray to OptiX, which would be a big effort, or whether that will be done by MI Genius or Lightwork Design, or it will just not happen. Another possibility is for some party to implement OptiX Prime on top of OptiX to get access to RTX hardware acceleration.

     

    compare:

    http://www.daz3d.com/forums/discussion/comment/3993836/#Comment_3993836

    - - -

    I can confirm that with OptiX Prime Acceleration On the render time is actually a few seconds slower and not faster.

    1x RTX 2080 Ti
    OptiX Prime Acceleration Off
    4 minutes 37.84 seconds
    OptiX Prime Acceleration On
    4 minutes 44.83 seconds

    - - -

    2x RTX 2080 Ti
    OptiX Prime Acceleration Off
    2 minutes 36.17 seconds
    OptiX Prime Acceleration On
    2 minutes 36.98 seconds

    I have not yet tested how the time difference is with an hour long render.

    Still, I have not found any possible advantage of having Optix Prime Acceleration On for Turing cards.

    For Turing cards it may be the right choice to turn Optix Prime Acceleration off

    - - -

    If someone else got different results with Turing please share...

    - - -

    Included this information in Request #283430

    - - -

    What scene have you tested your cards with? I'm curious to know how that stacks up with the 1080 Ti.

     

    DAZ_Rawb shared the Iray Render Test 2 scene on September 21:

    https://direct.daz3d.com/forums/discussion/comment/3969916/#Comment_3969916

    - - -

    A comparison with the same scene rendered with a 1080, 1080 Ti, 2x 1080 Ti, 2080 Ti, 2x 2080 Ti was shared here:

    https://www.daz3d.com/forums/discussion/comment/4009596/#Comment_4009596

    - - -

    Anonbach shared similar results for the same scene with a 2080 Ti here:


    I have an Ryzen 5 2600 | RTX 2080Ti system checking in - RTX 2080Ti does this test in 4 minutes and 35 seconds. 

    https://www.daz3d.com/forums/discussion/comment/4033951/#Comment_4033951

     

    - - -

    Results from Anonbach

    Ryzen 5 2600

    4 minutes and 35 seconds

    But I do not know if that was OptiX Prime Acceleration On or Off

    - - -

    compared with

    Intel Core i7 5820K
    OptiX Prime Acceleration Off
    4 minutes 37.84 seconds
    OptiX Prime Acceleration On
    4 minutes 44.83 seconds

    - - -

    Just a few seconds difference on two different systems...

    - - -

    Update / Edit: I am now running the scene at max resolution to see if in an hour long render any difference with Optix Prime Acceleration can be detected...

    Post edited by linvanchene on
  • hphoenixhphoenix Posts: 1,335

    Please remember that Optix is a GPU ray-tracing acceleration library.  If the scene being rendered does not require much in the way of raytracing, you may not see much benefit, if any.  For scenes which do involve a LOT of raytracing (reflections, refractions) you should generally see a benefit from using it.

     

  • linvanchenelinvanchene Posts: 1,382
    edited October 2018

    I wanted to know for sure if Optix Prime Acceleration might make a difference with 2x 2080 Ti on longer render times then a few minutes.

    Tldr version: Using Optix Prime Acceleration with Turing cards does not save render time but may cause it to be longer.

     - - -

    Iray Render Test 2 scene can be found here:

    https://direct.daz3d.com/forums/discussion/comment/3969916/#Comment_3969916

    Original Pixel Size
    1000x1300

    Adjusted Pixel Size for testing a longer render time
    7692x10000

    - - -

    Test system

    Win 10 Pro 64bit
    Intel Core i7 5820K
    ASUS X99-E WS
    64 GB RAM


    Asus GTX 1080 Ti FE for display
    2x ASUS GeForce RTX 2080 Ti TURBO for rendering

    Nvidia Driver Version: 416.34
    DAZ Studio 4.11.0.236
    Preview Viewport Wire Shaded
    DAZ Studio was closed and restarted between renders.

    Results:
    OptiX Prime Acceleration Off
    1 hours 38 minutes 24.32 seconds
    OptiX Prime Acceleration On
    1 hours 42 minutes 45.86 seconds

     

    - - -

    Previous Results for Pascal cards:

    1x GTX 1080 Ti
    Optix Prime Acceleration Off
    8 minutes 28.28 seconds
    OptiX Prime Acceleration On
    7 minutes 34.25 seconds

    - - -

    2x GTX 1080 Ti
    OptiX Prime Acceleration Off
    4 minutes 27.10 seconds
    OptiX Prime Acceleration On
    4 minutes 1.66 seconds

    - - -

    My interpretation of the results based on the available information:

    Pascal:

    One GTX 1080 Ti: Optix Prime Acceleration ON saves you ~66 seconds of render time at the Original Pixel Size of 1000x1300.

    Two GTX 1080 Ti: Optix Prime Acceleration ON saves you ~26 seconds of render time at the Original Pixel Size of 1000x1300.

    Turing:

    One RTX 2080 Ti: Optix Prime Acceleration ON increases render time for ~7 seconds at the Original Pixel Size of 1000x1300.

    Two RTX 2080 Ti: Optix Prime Acceleration ON increases render time for ~1 second at the Original Pixel Size of 1000x1300.

    Two RTX 2080 Ti: Optix Prime Acceleration ON increases render time for ~4 minutes at the adjusted Pixel Size of 7692x10000.

    - - -

    Switching Optix Prime Acceleration on does not seem to have any positive effect for Turing cards.

    Based on that I get the impression the information shared by davemc0 may be accurate.

    If anyone can find a scene in which Optix Prime Acceleration improves render time with Turing cards please share your results.

    - - -

    Attached the DAZ Studio log files to Request #283430

    - - -

     

     

    7692x10000 - Optix Prime Acceleration ON.png
    1920 x 1080 - 1M
    Post edited by linvanchene on
  • AalaAala Posts: 140
    linvanchene said:
    Aala said:
    linvanchene said:

    The answer for Pascal cards (1050, 1060, 1070, 1080, 1080 Ti) is as other posters allready explained.

    For Turing cards like 2070, 2080, 2080 Ti  davemc0 shared the information that Optix Prime Acceleration is not supported for RTX based GPU.


    Another thing to note is that OptiX Prime will not be accelerated on RTX; only OptiX will. This is significant because Iray uses OptiX Prime, not OptiX. Thus, Iray does not get RTX ray tracing acceleration on my RTX 2080, and something big will have to change before it does get accelerated. I don't know whether Nvidia will port Iray to OptiX, which would be a big effort, or whether that will be done by MI Genius or Lightwork Design, or it will just not happen. Another possibility is for some party to implement OptiX Prime on top of OptiX to get access to RTX hardware acceleration.

     

    compare:

    http://www.daz3d.com/forums/discussion/comment/3993836/#Comment_3993836

    - - -

    I can confirm that with OptiX Prime Acceleration On the render time is actually a few seconds slower and not faster.

    1x RTX 2080 Ti
    OptiX Prime Acceleration Off
    4 minutes 37.84 seconds
    OptiX Prime Acceleration On
    4 minutes 44.83 seconds

    - - -

    2x RTX 2080 Ti
    OptiX Prime Acceleration Off
    2 minutes 36.17 seconds
    OptiX Prime Acceleration On
    2 minutes 36.98 seconds

    I have not yet tested how the time difference is with an hour long render.

    Still, I have not found any possible advantage of having Optix Prime Acceleration On for Turing cards.

    For Turing cards it may be the right choice to turn Optix Prime Acceleration off

    - - -

    If someone else got different results with Turing please share...

    - - -

    Included this information in Request #283430

    - - -

    What scene have you tested your cards with? I'm curious to know how that stacks up with the 1080 Ti.

     

    DAZ_Rawb shared the Iray Render Test 2 scene on September 21:

    https://direct.daz3d.com/forums/discussion/comment/3969916/#Comment_3969916

    - - -

    A comparison with the same scene rendered with a 1080, 1080 Ti, 2x 1080 Ti, 2080 Ti, 2x 2080 Ti was shared here:

    https://www.daz3d.com/forums/discussion/comment/4009596/#Comment_4009596

    - - -

    Anonbach shared similar results for the same scene with a 2080 Ti here:


    I have an Ryzen 5 2600 | RTX 2080Ti system checking in - RTX 2080Ti does this test in 4 minutes and 35 seconds. 

    https://www.daz3d.com/forums/discussion/comment/4033951/#Comment_4033951

     

    - - -

    Results from Anonbach

    Ryzen 5 2600

    4 minutes and 35 seconds

    But I do not know if that was OptiX Prime Acceleration On or Off

    - - -

    compared with

    Intel Core i7 5820K
    OptiX Prime Acceleration Off
    4 minutes 37.84 seconds
    OptiX Prime Acceleration On
    4 minutes 44.83 seconds

    - - -

    Just a few seconds difference on two different systems...

    - - -

    Update / Edit: I am now running the scene at max resolution to see if in an hour long render any difference with Optix Prime Acceleration can be detected...

    I just tested that scene on my 1080 Ti again, as I did in the thread you linked (one page after). I'm still getting 5 mins 40 secs with Optix on. Not sure why you're getting 7 minutes 34.25 seconds as stated in the benchmakr thread.

  • AalaAala Posts: 140
    edited October 2018

    Okay, I found the culprit:

    DAZ Studio 4.10 - 5 mins 38 seconds

    DAZ Studio 4.11 BETA - 7 mins 42 seconds

    Seems like Iray has changed a lot between the two versions. We need to do more testing. I'm assuming the 2080 Ti doesn't run on 4.10?

    EDIT: I'm also attaching the two rendered images from both DAZ versions. Seems to me that there's quite a bit of a difference between both Iray versions

    4.10-T1.png
    1000 x 1300 - 2M
    4.11-T1.png
    1000 x 1300 - 2M
    Post edited by Aala on
  • linvanchenelinvanchene Posts: 1,382
    edited October 2018
    Aala said:

    Okay, I found the culprit:

    DAZ Studio 4.10 - 5 mins 38 seconds

    DAZ Studio 4.11 BETA - 7 mins 42 seconds

    Seems like Iray has changed a lot between the two versions. 

    Thank you alot for sharing this. 

    I did not expect the difference to be that extreme.

    Does anyone know if this has been reported as well and is known by Iray developers?

     

    I'm assuming the 2080 Ti doesn't run on 4.10?

    Afaik no. You need to use the 4.11 Beta for Turing cards like 2080 Ti.

     

    EDIT: I'm also attaching the two rendered images from both DAZ versions. Seems to me that there's quite a bit of a difference between both Iray versions

     Another look trough the list of Iray updates may indicate which material properties have changed.

     

    We need to do more testing.

    I only install the latest Beta version and then stick with that.

    The latest general release is 

    • Version: 4.10.0.123
    • Date: December 8, 2017

    There certainly has changed a lot since then...

    It probably would make it easier to find issues if more people would be using the latest beta version.

    - - -

    Post edited by linvanchene on
  • outrider42outrider42 Posts: 3,679

    There have been some comments in the forums about the beta being slower. There is also a beta thread in the forum, I'd look there. There could be other factors at play, though. I have read that Iray can change how it calculates convergence from one version to the next. That could be enough to make a big difference if the benchmark being used is based only off convergence. That is why when I made my benchmark I capped it at 5000 iterations. That prevents any changes to convergence from skewing the results. Back in the benchmark thread in my sig, I noticed that one person posted the iteration count from the SY bench. They had a higher iteration count for the SY bench in 4.11 compared to 4.10, and that would certainly explain a longer render time.

    You can take a look at the renders from 4.10 and 4.11 if you happened to save them and see if one looks better. If 4.11 is doing more to converge then the image should at least look a bit better.

    I think the help file also lists the average time per iteration, look at this stat and see if it is different. If the time per iteration is higher in 4.11, then we have proof that it is truly slower than 4.10.

    I certainly hope Daz is getting into Nvidia's ear about improvements for whatever is next. Octane sure is looking like the place to be in 2019 to me.

  • david_23523141david_23523141 Posts: 40
    edited November 2019

    A great article on Optix and the non-RT cards VERSUS the newer RTX 2080 and RTX 2080 Ti Cards  http://boostclock.com/show/000219/gpu-rendering-nv-fermat-gtx980ti-gtx1080-gtx1080ti-titanv.html

    Here is the WIKI page on OptiX in general https://en.wikipedia.org/wiki/OptiX

    Post edited by david_23523141 on
  • outrider42outrider42 Posts: 3,679
    That link is kind of old now, as Iray has been completely overhauled since this thread and those benchmarks have been made.

    So the advice from 2018 doesn't quite apply now.

    Iray 2019 uses OptiX 6.0. The previous versions of Iray used OptiX Prime. They have some key differences, the most obvious being that 6.0 can utilize the RT cores found in RTX cards.

    Anyway, for Iray 2019, OptiX Prime acceleration is not available, because they are using 6.0. The check box may be there in some versions of Daz Studio, but it should be removed soon if a beta hasn't already done. So if you have Turing, it doesn't matter if you turn that box on or not, it doesn't work and there is no impact on rendering either way.

  • nothinglifeart_366ec5bc62nothinglifeart_366ec5bc62 Posts: 2
    linvanchene said:

    The answer for Pascal cards (1050, 1060, 1070, 1080, 1080 Ti) is as other posters allready explained.

    For Turing cards like 2070, 2080, 2080 Ti  davemc0 shared the information that Optix Prime Acceleration is not supported for RTX based GPU.


    Another thing to note is that OptiX Prime will not be accelerated on RTX; only OptiX will. This is significant because Iray uses OptiX Prime, not OptiX. Thus, Iray does not get RTX ray tracing acceleration on my RTX 2080, and something big will have to change before it does get accelerated. I don't know whether Nvidia will port Iray to OptiX, which would be a big effort, or whether that will be done by MI Genius or Lightwork Design, or it will just not happen. Another possibility is for some party to implement OptiX Prime on top of OptiX to get access to RTX hardware acceleration.

     

    compare:

    http://www.daz3d.com/forums/discussion/comment/3993836/#Comment_3993836

    - - -

    I can confirm that with OptiX Prime Acceleration On the render time is actually a few seconds slower and not faster.

    1x RTX 2080 Ti
    OptiX Prime Acceleration Off
    4 minutes 37.84 seconds
    OptiX Prime Acceleration On
    4 minutes 44.83 seconds

    - - -

    2x RTX 2080 Ti
    OptiX Prime Acceleration Off
    2 minutes 36.17 seconds
    OptiX Prime Acceleration On
    2 minutes 36.98 seconds

    I have not yet tested how the time difference is with an hour long render.

    Still, I have not found any possible advantage of having Optix Prime Acceleration On for Turing cards.

    For Turing cards it may be the right choice to turn Optix Prime Acceleration off

    - - -

    If someone else got different results with Turing please share...

    - - -

    Included this information in Request #283430

    - - -

     

  • nothinglifeart_366ec5bc62nothinglifeart_366ec5bc62 Posts: 2

    Hi , when you said "Turing" is for RTX GIbabyte made ?

  • fastbike1fastbike1 Posts: 4,078

     All RTX cards are Turing. Turing is the GPU chip.

    nothinglifeart_366ec5bc62 said:

    Hi , when you said "Turing" is for RTX GIbabyte made ?

     

Sign In or Register to comment.