Iray Starter Scene: Post Your Benchmarks!

1404143454649

Comments

  • timon630timon630 Posts: 37
    LenioTG said:
    Heya, where u got results for a 1070 ti ?

    All results are taken from this topic. For the 1070 ti, there are few results, so I pointed out "*Few results. Possibly incorrect data."

    I doubt your 1070 TI (2:27) works better than 1080 (2:55), for which the data is sufficient.

  • kattyg911kattyg911 Posts: 7
    timon630 said:
    LenioTG said:
    Heya, where u got results for a 1070 ti ?

    All results are taken from this topic. For the 1070 ti, there are few results, so I pointed out "*Few results. Possibly incorrect data."

    I doubt your 1070 TI (2:27) works better than 1080 (2:55), for which the data is sufficient.

    https://www.youtube.com/watch?v=3GzPg4BtaLc&feature=youtu.be

  • RayDAntRayDAnt Posts: 1,120
    edited May 2019
    kattyg911 said:
    timon630 said:
    LenioTG said:
    Heya, where u got results for a 1070 ti ?

    All results are taken from this topic. For the 1070 ti, there are few results, so I pointed out "*Few results. Possibly incorrect data."

    I doubt your 1070 TI (2:27) works better than 1080 (2:55), for which the data is sufficient.

    https://www.youtube.com/watch?v=3GzPg4BtaLc&feature=youtu.be

    See this post. In short, Total Rendering Time introduces a margin-of-error of +-30 seconds (or more on older systems) between different computers running the same benchmark into all rendering performance measurements derived from it.

    Post edited by RayDAnt on
  • outrider42outrider42 Posts: 3,679
    edited May 2019
    timon630 said:
    LenioTG said:
    Heya, where u got results for a 1070 ti ?

    All results are taken from this topic. For the 1070 ti, there are few results, so I pointed out "*Few results. Possibly incorrect data."

    I doubt your 1070 TI (2:27) works better than 1080 (2:55), for which the data is sufficient.

    When migenius benchmarked the 1070ti they actually graded it faster than the 1080 for Iray. And they used the same version of Iray for all cards tested. The 1070ti is just a few cores short of a 1080, so I think its pretty reasonable for a 1070ti to perform like a 1080, especially if you have an overclocked model. Don't forget that clockspeeds can be a factor in these times. Also the 1070ti released a lot later, I believe it has a slightly updated CUDA compute, though I could be wrong. I thought I saw it was 6.2 vs the 6.1 of other Pascal models. Even if it the same 6.1, overclocking can net 1080 performance.

    https://www.migenius.com/products/nvidia-iray/iray-2017-1-benchmarks

    Interestingly, migenius has added the 2080ti to the list.
    Post edited by outrider42 on
  • LenioTGLenioTG Posts: 2,116
    edited May 2019

    Wow :O

    PS: do you know that if you're not pressing Shift, the "Enter" button already gives some spacing? It looks like you're hitting enter two times pressing shift at the same time from the quotes.

    Thanks for sharing this benchmark!

    It's a bummer that it doesn't have all the GPUs, like with Octanebench :(

    Post edited by LenioTG on
  • outrider42outrider42 Posts: 3,679
    I'm currently on mobile, and using the Daz forum on mobile fiercely sucks. The formatting is all screwed up. Users have had to manually enter HTML code to make a space between paragraphs, unless Daz finally upgraded the forum to something from the last decade. Lets see if this shows up correctly.
  • HP Z820 V2

    DDR3 1866 ECC RAM 
    3X 1080ti no OC
    Dual Xeon 2667 V2 water cooled, 16 cores

    render time: 1 min 59 seconds 

  • outrider42outrider42 Posts: 3,679

    HP Z820 V2

    DDR3 1866 ECC RAM 
    3X 1080ti no OC
    Dual Xeon 2667 V2 water cooled, 16 cores

    render time: 1 min 59 seconds 

    Is this the Sickleyield scene?
  • jonascsjonascs Posts: 35

    Just thought I'd mention my latest test with the outrider scene.

    I let my 2x2018ti render the scene to 100 iterations, once with denoiser off, and the other with it on.

    I took about 5 seconds for both renders but the result is telling.

     

    benchtest.jpg
    720 x 520 - 269K
    benchtest denoise.jpg
    720 x 520 - 151K
  • bailaowaibailaowai Posts: 44
    timon630 said:

    Hi, everybody. Sorry for my English.
    I put all the data in one table.

     

    OMG finally.  Thank you so much.

  • RayDAntRayDAnt Posts: 1,120
    bailaowai said:
    timon630 said:

    Hi, everybody. Sorry for my English.
    I put all the data in one table.

     

    OMG finally.  Thank you so much.

    Just keep in mind that there is a relatively HUGE margin of error in the data it's derived from.

  • LenioTGLenioTG Posts: 2,116
    RayDAnt said:
    bailaowai said:
    timon630 said:

    Hi, everybody. Sorry for my English.
    I put all the data in one table.

     

    OMG finally.  Thank you so much.

    Just keep in mind that there is a relatively HUGE margin of error in the data it's derived from.

    I think that for now the Octanebench results are the best ones for reference!

    We should suggest Daz to realize a short benchmark themselves, imagine all the data they could collect!

  • asielasusasielasus Posts: 0

    Benchmark scene (IrayTestSceneStarterEssentialsonly.duf)

    Ran on DAZ 4.11.0.380 Beta 64bit

    Total render time: 54.48 seconds (GPUs only)

     

    System specs

    Note: No monitors connected on the 4 cards. Driver installed is 419.67.

    2x2070 RTX 8GB (GeForce RTX 2070): compute capability 7.5, 8 GiB total) - 2xGTX Titan 6GB  (GeForce GTX TITAN): compute capability 3.5, 6 GiB total) 

    CPU: i7 5960X @3 GHz

    Ram:32 GB DDR4

    OS Windows 10 (10.0.17134)

  • outrider42outrider42 Posts: 3,679

    A benchmark would be fantastic, but I don't think Daz is all that interested in it.

    However, I think a built in benchmark would be actually be a great marketing tool. Just think if some media outlets used Daz Iray to test out new GPUs. That can get your software noticed. In the gaming world, there is a game called "Ashes of the Singularity". To be blunt the game itself is a very average RTS game, and it probably would be totally forgotten by now except for one thing...it has a fantastic benchmarking suite. The game is constantly optimized for all sorts of new tech. It was one of the first games to have DirectX 12, for example. The game also makes use of async compute, which was especially rare in gaming. Plus Ashes did something almost nobody else even tried to do, you can actually use AMD and Nvidia GPU's TOGETHER at the same time to boost performance! How many of you knew this was even possible? LOL. It is a feature of DirectX 12, BTW. So of course media outlets had to test this, giving Ashes more exposure again.

    As a result, Ashes has become a mainstay in many benchmark lists since its release. Some of the people who run benchmarks will jokingly call the game "Ashes of the Benchmark" because it is mostly known for benchmarks more than being a game. Still, it cannot be argued that Ashes has benefitted greatly because of this exposure, as some people have bought Ashes purely to use it as a benchmark.

    So I believe that the argument can be made for having a built in benchmark from a marketing perspective. It could show off just what Daz Iray can do and expose Daz Studio to a lot more people who probably never heard of it. A benchmark can have real value.

  • fred9803fred9803 Posts: 1,558

    Thank you timon630 for all that work. The biggie will be when Iray gets support for those RTX cors, and the results may change quite a bit.

    DAZ is probably waiting for this to happen before their next General Release build.

  • LightZergoLightZergo Posts: 2
    edited June 2019
    timon630 said:
    LenioTG said:
    Heya, where u got results for a 1070 ti ?

    All results are taken from this topic. For the 1070 ti, there are few results, so I pointed out "*Few results. Possibly incorrect data."

    I doubt your 1070 TI (2:27) works better than 1080 (2:55), for which the data is sufficient.

    When migenius benchmarked the 1070ti they actually graded it faster than the 1080 for Iray. And they used the same version of Iray for all cards tested. The 1070ti is just a few cores short of a 1080, so I think its pretty reasonable for a 1070ti to perform like a 1080, especially if you have an overclocked model. Don't forget that clockspeeds can be a factor in these times. Also the 1070ti released a lot later, I believe it has a slightly updated CUDA compute, though I could be wrong. I thought I saw it was 6.2 vs the 6.1 of other Pascal models. Even if it the same 6.1, overclocking can net 1080 performance.

    https://www.migenius.com/products/nvidia-iray/iray-2017-1-benchmarks

    Interestingly, migenius has added the 2080ti to the list.

    In both cases, CUDA version 6.1. The built-in render test is a great idea!
    Post edited by LightZergo on
  • RobinsonRobinson Posts: 751
    edited June 2019

    Benchmark scene (IrayTestSceneStarterEssentialsonly.duf)

    Ran on DAZ 4.11.0.382 Beta 64bit

    Total render time: 1 minutes 55.73 seconds (GPU only, Optix enabled) - RTX 2070, 4790K, 16Gb DDR 4.

     

    Interestingly I pulled up the results I posted on the OPs Deviant Art page, which was: 1 minutes 49 seconds.  So it's about 6 seconds slower with the current beta than 4.11 previous beta.  If I enable post denoiser, I get 1 minute 56 seconds, so pretty much the same speed.  I also don't think 1m 49 and 1m 55 are significantly different in the grand scheme of things.  I mean if you could put error bars on the results...

     

    Gosh, I can't wait for the iRay update with RTX enabled.  Maybe next year...

    Post edited by Robinson on
  • outrider42outrider42 Posts: 3,679
    Which GPU do you have?
  • RobinsonRobinson Posts: 751
    Which GPU do you have?

    Doh! I edited my post.

  • outrider42outrider42 Posts: 3,679
    Oh nice, another 2070 bench, so a 2070 is indeed a bit faster than a 1080ti right out of the box, and that is without RTX. However, what do you mean by it being 6 seconds slower than 4.10? The 2070 does't work in 4.10 at all. Do you mean a previous version of the beta?
  • RobinsonRobinson Posts: 751
    Oh nice, another 2070 bench, so a 2070 is indeed a bit faster than a 1080ti right out of the box, and that is without RTX. However, what do you mean by it being 6 seconds slower than 4.10? The 2070 does't work in 4.10 at all. Do you mean a previous version of the beta?

    I looked at the previous one and it was the previous 4.11 beta, not 4.10.  The current beta is slower by that much.  However... I had two cards plugged in previously (a 970 as well) which I was using to drive the monitor.  I don't think Windows Desktop Manager would make a 6s difference though, so I'm putting it down to the iRay changes in the current beta.

  • outrider42outrider42 Posts: 3,679
    Okay. You can also see slightly different times just from run to run. All sorts of things can effect it. Even a difference in ambient room temperature can effect benchmarking by effecting the case cooling. Your GPU driver can impact it, too. I don't think the beta has updated its Iray plugin for quite a while, so it may not be Daz. It can also just be within margin of error.
  • RobinsonRobinson Posts: 751
    Okay. You can also see slightly different times just from run to run. All sorts of things can effect it. Even a difference in ambient room temperature can effect benchmarking by effecting the case cooling. Your GPU driver can impact it, too. I don't think the beta has updated its Iray plugin for quite a while, so it may not be Daz. It can also just be within margin of error.

    I think this most recent beta did have significant iRay changes (CUDA version for example).  But yes, otherwise the variance is not significant imho.  I was surprised the filtering didn't speed it up though.  Perhaps the benefit from that increases with the image resolution.

  • outrider42outrider42 Posts: 3,679
    The filter doesn't speed up convergence, convergence is fully independent. All it does is clean up the noise. It can produce a render that looks "done" faster, so that you can manually stop the render sooner than you might otherwise, that is the benefit. The denoiser basically "guesses" what the missing pixels are, but as the image converges, these pixels will fill in with the calculated pixels. Your mileage will vary. Some people hate the filter because it can wash some detail out. But it depends on your scene and what your goal is. I think the filter is excellent for getting fast results when rendering full 3D environments. My results with humans are more mixed, but tilt positive.
  • LenioTGLenioTG Posts: 2,116
    edited June 2019

    So, this is my benchmark with the new GPU

    RTX 2060, GPU ONLY, OPTIX ON, SICKLEYIELD SCENE
    2 MIN AND 13 SEC

    I have the latest Beta and the latest Creator drivers.

    I just can't find my previous benchmark. But I see in the graph that a GTX 1060 basically took double the time.

    I see the other benchmark with an RTX 2060 had the exact same time.

    What to say? So far, I'm very happy of my purchase :)

    Post edited by LenioTG on
  • outrider42outrider42 Posts: 3,679
    LenioTG said:

    So, this is my benchmark with the new GPU

    RTX 2060, GPU ONLY, OPTIX ON, SICKLEYIELD SCENE
    2 MIN AND 13 SEC

    I have the latest Beta and the latest Creator drivers.

    I just can't find my previous benchmark. But I see in the graph that a GTX 1060 basically took double the time.

    I see the other benchmark with an RTX 2060 had the exact same time.

    What to say? So far, I'm very happy of my purchase :)

    Just 2 seconds slower than the Titan X Pascal, how cool is that? The 2060 is a gem. I'm glad you chose that over the 1660ti.

  • fred9803fred9803 Posts: 1,558
    LenioTG said:

    So, this is my benchmark with the new GPU

    RTX 2060, GPU ONLY, OPTIX ON, SICKLEYIELD SCENE
    2 MIN AND 13 SEC

    I have the latest Beta and the latest Creator drivers.

    I just can't find my previous benchmark. But I see in the graph that a GTX 1060 basically took double the time.

    I see the other benchmark with an RTX 2060 had the exact same time.

    What to say? So far, I'm very happy of my purchase :)

    I should update my drivers and use the latest version of DS because your 2060 is a whole lot faster than my 2080 on that scene. In theory my 2080 should be 37% faster than your GPU.

    Then again my benchmark included the time for the render to start and was timed to 100% convergence. Is that the way we should be benchmarking our GPUs? I'm not sure because some people are posting results at 90% or 95% convernence and only time it from when the render kicks in.

  • LenioTGLenioTG Posts: 2,116
    LenioTG said:

    So, this is my benchmark with the new GPU

    RTX 2060, GPU ONLY, OPTIX ON, SICKLEYIELD SCENE
    2 MIN AND 13 SEC

    I have the latest Beta and the latest Creator drivers.

    I just can't find my previous benchmark. But I see in the graph that a GTX 1060 basically took double the time.

    I see the other benchmark with an RTX 2060 had the exact same time.

    What to say? So far, I'm very happy of my purchase :)

    Just 2 seconds slower than the Titan X Pascal, how cool is that? The 2060 is a gem. I'm glad you chose that over the 1660ti.

    And thank you for your suggestions and because you care! :D

    Where can I find your benchmark scene? I don't know why but the download button from your gallery doesn't seem to work for me

     

    fred9803 said:
    LenioTG said:

    So, this is my benchmark with the new GPU

    RTX 2060, GPU ONLY, OPTIX ON, SICKLEYIELD SCENE
    2 MIN AND 13 SEC

    I have the latest Beta and the latest Creator drivers.

    I just can't find my previous benchmark. But I see in the graph that a GTX 1060 basically took double the time.

    I see the other benchmark with an RTX 2060 had the exact same time.

    What to say? So far, I'm very happy of my purchase :)

    I should update my drivers and use the latest version of DS because your 2060 is a whole lot faster than my 2080 on that scene. In theory my 2080 should be 37% faster than your GPU.

    Then again my benchmark included the time for the render to start and was timed to 100% convergence. Is that the way we should be benchmarking our GPUs? I'm not sure because some people are posting results at 90% or 95% convernence and only time it from when the render kicks in.

    My result included the start time and it was at 100% convergence! I've read the time on the log file.
    90% was reached in a matter of seconds, I was very surprised.

    I don't know why your 2080 is having that speed! I have just the GPU checked, OptiX ON and the latest drivers and Daz Public Build. Since there's another entry in this thread with the exact same time, I think it's a 2060 speed in Daz!

  • outrider42outrider42 Posts: 3,679
    fred9803 said:
    LenioTG said:

    So, this is my benchmark with the new GPU

    RTX 2060, GPU ONLY, OPTIX ON, SICKLEYIELD SCENE
    2 MIN AND 13 SEC

    I have the latest Beta and the latest Creator drivers.

    I just can't find my previous benchmark. But I see in the graph that a GTX 1060 basically took double the time.

    I see the other benchmark with an RTX 2060 had the exact same time.

    What to say? So far, I'm very happy of my purchase :)

    I should update my drivers and use the latest version of DS because your 2060 is a whole lot faster than my 2080 on that scene. In theory my 2080 should be 37% faster than your GPU.

    Then again my benchmark included the time for the render to start and was timed to 100% convergence. Is that the way we should be benchmarking our GPUs? I'm not sure because some people are posting results at 90% or 95% convernence and only time it from when the render kicks in.

    Load the scene and don't touch anything. Just hit render. The scene renders to 95% convergence by default. A lot of people say 100%, but that it is actually not referring to convergence, most people get this confused.

    The only settings that you may change are Optix on/off and the speed or memory optimization toggle, and report if you do alter these. Any other changes will invalidate the bench.
Sign In or Register to comment.