Daz Studio Iray - Rendering Hardware Benchmarking

145791045

Comments

  • fred9803fred9803 Posts: 1,558

    I think I'll come back to this thread when NVIDIA releases their GeForce RTX 3080 Ti with next-gen Ampere this summer. 12-20Gb VRAM. Predicted to be much faster and even cheaper than the 2080Ti. Anyone want to buy a second hand and heavily used RTX 2080? .... or in the recycle bin.

  • DripDrip Posts: 1,125
    fred9803 said:

    I think I'll come back to this thread when NVIDIA releases their GeForce RTX 3080 Ti with next-gen Ampere this summer. 12-20Gb VRAM. Predicted to be much faster and even cheaper than the 2080Ti. Anyone want to buy a second hand and heavily used RTX 2080? .... or in the recycle bin.

    Latest rumours about NVidia's new cards suggest that it might not be a new 30xx series, but some new Quadro or maybe Titan cards, which would suggest that 30xx cards wouldn't be announced until the end of this year, or maybe even next year.

  • LeanaLeana Posts: 10,932
    Drip said:
    fred9803 said:

    I think I'll come back to this thread when NVIDIA releases their GeForce RTX 3080 Ti with next-gen Ampere this summer. 12-20Gb VRAM. Predicted to be much faster and even cheaper than the 2080Ti. Anyone want to buy a second hand and heavily used RTX 2080? .... or in the recycle bin.

    Latest rumours about NVidia's new cards suggest that it might not be a new 30xx series, but some new Quadro or maybe Titan cards, which would suggest that 30xx cards wouldn't be announced until the end of this year, or maybe even next year.

    And even if the 30xx series is actually released this summer, if it's anything like with the 20xx series then Iray support for the new cards won't be available from Nvidia (let alone included in DS) for quite some time after release.

  • droidy001droidy001 Posts: 277
    fred9803 said:

    I think I'll come back to this thread when NVIDIA releases their GeForce RTX 3080 Ti with next-gen Ampere this summer. 12-20Gb VRAM. Predicted to be much faster and even cheaper than the 2080Ti. Anyone want to buy a second hand and heavily used RTX 2080? .... or in the recycle bin.

    If you're correct that it will be cheaper then I'd watch out for a sudden drop in price of the 2080 that might could indicate it's close to release. Suppliers wouldn't want to get stuck with a load of older gen cards
  • Leana said:
    Drip said:
    fred9803 said:

    I think I'll come back to this thread when NVIDIA releases their GeForce RTX 3080 Ti with next-gen Ampere this summer. 12-20Gb VRAM. Predicted to be much faster and even cheaper than the 2080Ti. Anyone want to buy a second hand and heavily used RTX 2080? .... or in the recycle bin.

    Latest rumours about NVidia's new cards suggest that it might not be a new 30xx series, but some new Quadro or maybe Titan cards, which would suggest that 30xx cards wouldn't be announced until the end of this year, or maybe even next year.

    And even if the 30xx series is actually released this summer, if it's anything like with the 20xx series then Iray support for the new cards won't be available from Nvidia (let alone included in DS) for quite some time after release.

    and the delay with the 10x0 cards was even longer.

  • RayDAntRayDAnt Posts: 1,120
    droidy001 said:
    Are there any scores for the Titan X (not xp)? A handful on ebay for £220 If it's anywhere near my 2060 super I'd be tempted to buy one, for that extra vram.

    Jsut a single composite bench for 2 Ttan Xp's plus a single Titan X (Pascal edition) I'm afraid. Keep mind that there are two versions of the Titan X out there: the orginal Maxwell based Titan X and an upgraded Pascal based Titan X (not to be confused with the Titan Xp, which is also Pascal based but has more active cores.)

  • droidy001droidy001 Posts: 277
    RayDAnt said:
    droidy001 said:
    Are there any scores for the Titan X (not xp)? A handful on ebay for £220 If it's anywhere near my 2060 super I'd be tempted to buy one, for that extra vram.

    Jsut a single composite bench for 2 Ttan Xp's plus a single Titan X (Pascal edition) I'm afraid. Keep mind that there are two versions of the Titan X out there: the orginal Maxwell based Titan X and an upgraded Pascal based Titan X (not to be confused with the Titan Xp, which is also Pascal based but has more active cores.)

    Cheers, they're the Maxwell's. Even with the pascal I would probably take a hit in performance, so I'm guessing the maxwell would be a bigger hit.
  • xionisxionis Posts: 18

    Well I decided to upgrade my 2080 Super cards to 2 Titan RTX's. So then I ran the benchmarks...here are the results...

    System Configuration (Single GPU ONLY)
    System/Motherboard: Asus ROG STRIX TRX40-E Gaming
    CPU: AMD Threadripper 3970x @ Stock
    GPU: GPU1 Asus RTX 2080 Super @ stock , GPU2 Asus RTX 2080 Super @ stock 
    System Memory: GSKILL Trident Z Neo 64GB/Quad Channel @ 3600MHz
    OS Drive: Corsair MP600 NVMe Gen4 1TB
    Asset Drive: Samsung 970 EVO Plus  NVMe 1TB
    Operating System: Windows 10 Pro 64bit 1909
    Nvidia Drivers Version: Gameready driver version 442.59
    Daz Studio Version: 4.12.0.86 64bit
    Optix Prime Acceleration: OFF

    Benchmark Results
    DAZ_STATS

    2020-03-19 00:04:21.325 Finished Rendering
    2020-03-19 00:04:21.370 Total Rendering Time: 4 minutes 0.18 seconds
    IRAY_STATS

    2020-03-19 00:05:09.370 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:
    2020-03-19 00:05:09.370 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (TITAN RTX): 1800 iterations, 1.746s init, 235.428s render


    Rendering Performance: 1800 / 235.428s] 7.645 iterations per second
    Loading Time: [(240.18) - 235.428] 4.752 seconds

     

    System Configuration (BOTH TITAN RTX ONLY)
    System/Motherboard: Asus ROG STRIX TRX40-E Gaming
    CPU: AMD Threadripper 3970x @ Stock
    GPU: GPU1 Asus RTX 2080 Super @ stock , GPU2 Asus RTX 2080 Super @ stock 
    System Memory: GSKILL Trident Z Neo 64GB/Quad Channel @ 3600MHz
    OS Drive: Corsair MP600 NVMe Gen4 1TB
    Asset Drive: Samsung 970 EVO Plus  NVMe 1TB
    Operating System: Windows 10 Pro 64bit 1909
    Nvidia Drivers Version: Gameready driver version 442.59
    Daz Studio Version: 4.12.0.86 64bit
    Optix Prime Acceleration: OFF

    Benchmark Results
    DAZ_STATS

    2020-03-19 00:07:28.363 Finished Rendering
    2020-03-19 00:07:28.398 Total Rendering Time: 2 minutes 4.57 seconds
    IRAY_STATS

    2020-03-19 00:08:00.462 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:
    2020-03-19 00:08:00.462 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (TITAN RTX): 895 iterations, 1.764s init, 119.963s render
    2020-03-19 00:08:00.462 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (TITAN RTX): 905 iterations, 1.960s init, 120.305s render


    Rendering Performance: 1800 / 120.305] 14.961 iterations per second
    Loading Time: [(124.57) - 235.428] 4.265 seconds

     

     

    System Configuration (THREADRIPPER 3970x and BOTH TITAN RTX )
    System/Motherboard: Asus ROG STRIX TRX40-E Gaming
    CPU: AMD Threadripper 3970x @ Stock
    GPU: GPU1 Asus RTX 2080 Super @ stock , GPU2 Asus RTX 2080 Super @ stock 
    System Memory: GSKILL Trident Z Neo 64GB/Quad Channel @ 3600MHz
    OS Drive: Corsair MP600 NVMe Gen4 1TB
    Asset Drive: Samsung 970 EVO Plus  NVMe 1TB
    Operating System: Windows 10 Pro 64bit 1909
    Nvidia Drivers Version: Gameready driver version 442.59
    Daz Studio Version: 4.12.0.86 64bit
    Optix Prime Acceleration: OFF

    Benchmark Results
    DAZ_STATS
    2020-03-19 00:27:37.375 Finished Rendering
    2020-03-19 00:27:37.409 Total Rendering Time: 1 minutes 51.73 seconds

    IRAY_STATS
    2020-03-19 00:28:06.823 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:
    2020-03-19 00:28:06.823 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (TITAN RTX): 757 iterations, 1.950s init, 106.770s render
    2020-03-19 00:28:06.823 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (TITAN RTX): 772 iterations, 2.310s init, 106.338s render
    2020-03-19 00:28:06.823 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CPU: 271 iterations, 1.670s init, 106.482s render

    Rendering Performance: 1800 /106.770] 16.858 iterations per second

    Loading Time: [(111.73) - 106.77] 4.96 seconds

  • RayDAntRayDAnt Posts: 1,120

    Important Announcement

    Just an FYI that the benchmarking reference scene ("RayDAnt - DS Iray Benchmark 2019A_r3.duf") has been refreshed to maintain statistical parity with past benchmarknig runs despite newly added Iray features in DS 4.12.1.109 (like SSIM.) So everyone should re-download the scene before performnig any future benchmarks.

  • outrider42outrider42 Posts: 3,679

    What do you mean by 'refreshed'? Is this just to disable SSIM to make sure people don't have it on?

    SSIM seems interesting, BTW.

  • RayDAntRayDAnt Posts: 1,120
    edited April 2020

    What do you mean by 'refreshed'? Is this just to disable SSIM to make sure people don't have it on?

    Pretty much. DS employs a file saving scheme whereby only specific scene parameters are saved in a file. Meaning that yet-to-be-released features are left to their own devices as to parameter states when loading a file from a previous applciation version. Simply opening a past scene in the new DS version and re-saving it binds the values of those new parameters going forward. Luckily for us in a non-destructive fashion (unlike eg. many Adobe applications, that require a sometimes lengthy "conversion" process to open old file saves in more recent application versions.)

     

    ETA: And yeah, also very intrigued to see how SSIM performs (been tracking mentions of its existence in Iray documentation for at least 4 months now.)

    Am also interested to see how Daz's implementation of Filament as a PBR alternative to Iray turns out once it actually makes it into the public release.

    Post edited by RayDAnt on
  • outrider42outrider42 Posts: 3,679
    edited April 2020

    I did some new tests with the 4.12.1.109 beta since Iray has changed a bit. I also wanted to see how much VRAM was being used, not just by the scene, but by the new SSID feature, and also denoising. And to see what impact on render speed those have, as they are supposed to hurt rendering speed. I found out something else, my times are slower overall, however it is not the new Iray. I still have 4.11 tucked away on my system, and my times are slower on it as well. This is also with 2 different drivers. We'll see below.

    The first test with driver 441.66. Everything about my system is the same as before.

    Windows 10 1903

    CPU: i5 4690K

    GPU #1:  EVGA 1080ti SC2

    GPU #2: MSI 1080ti Gaming  <--this is my display, yes, I use GPU 2 for display.

    RAM 32GB HyperX

    OS Drive Samsung 860 EVO 1TB

    Asset Drive: Samsung 860 EVO 1TB and WB 4TB Black HDD

    2020-04-22 16:41:55.940 Total Rendering Time: 4 minutes 38.7 seconds

    2020-04-22 16:43:20.426 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2020-04-22 16:43:20.426 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce GTX 1080 Ti): 903 iterations, 7.075s init, 266.593s render

    2020-04-22 16:43:20.426 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (GeForce GTX 1080 Ti): 897 iterations, 7.362s init, 266.475s render

    VRAM: GPU1 2885  GPU2 3415 MB

    That establishes a baseline for VRAM use. You can see that my display GPU is using 530 MB more than the non display card. The non display card had 0 MB before starting Daz.

    This next test is with SSID on.

    2020-04-22 16:50:32.580 Total Rendering Time: 4 minutes 38.52 seconds

    2020-04-22 16:51:01.109 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2020-04-22 16:51:01.110 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce GTX 1080 Ti): 899 iterations, 6.446s init, 268.581s render

    2020-04-22 16:51:01.110 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (GeForce GTX 1080 Ti): 901 iterations, 6.730s init, 268.494s render

    VRAM: GPU1 3465  GPU2 3436

    So here you can see that the time is 2 seconds slower, a very minor drop. The VRAM used is very odd. Somehow the first GPU, which is not the display, now uses slightly more VRAM than the display GPU! This is a most odd finding. However, the display GPU is using just a tiny amount of VRAM for SSID, just 21 MB, while the non display jumped 580 MB. Overall though, notice how balanced they are. I am thinking that the SSID is smart enough to balance the load between the GPUs so that they use similar amounts. And notice that in the first test, the non display GPU rendered 6 more frames, but in this test, the display GPU actually rendered 2 more frames. I believe this is due to the SSID running more on the display GPU since that is where most of the VRAM went for SSID. Interesting stuff. I saw no difference in the actual renders between these two tests, which makes sense.

    This next test is with the good old denoiser. You either love it or hate it. Lets see how it effects performance.

    2020-04-22 16:58:58.783 Total Rendering Time: 4 minutes 37.50 seconds

    2020-04-22 16:59:27.262 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2020-04-22 16:59:27.262 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce GTX 1080 Ti): 902 iterations, 5.943s init, 268.297s render

    2020-04-22 16:59:27.262 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (GeForce GTX 1080 Ti): 898 iterations, 6.328s init, 268.191s render

    VRAM: GPU1 3225 GPU2 3436

    Well, it actually clocked in a tiny fraction faster than SSID, though a still second or so slower than no post. Still, this difference is so minor as to be within a margin of error. It could depend on the scene. The VRAM used is less than the SSID, at least for the non display card. The display card is actually identical. And this is pretty interesting, because again, most of the VRAM that the denoiser asked for came from just one GPU. And this time the non display GPU rendered 4 more frames than the other.

    Another note, the non display GPU settled at 1911 Mhz, while the display GPU settled at 1936. So by clockspeed, the display GPU should be rendering more frames most of the time, they are both 1080tis. So I imagine being the display is costing it a few frames allowing the non display to 'win'. In my experience, the non display almost always wins, by just a few frames.

    At this point I updated my drivers to 442.92. Not much changed, in fact they are so similar I don't think I need to post them. But I had noticed my times were slower than before. I was hitting right at 4 minutes before on this scene. I still have 4.11 on my PC, so I fired it up.

    4.11.0.0236 optix on This predates the RTX update.

    2020-04-22 22:51:51.224 Total Rendering Time: 4 minutes 28.61 seconds

    2020-04-22 22:52:12.199 Iray INFO - module:category(IRAY:RENDER):   1.0   IRAY   rend info : Device statistics:

    2020-04-22 22:52:12.199 Iray INFO - module:category(IRAY:RENDER):   1.0   IRAY   rend info : CUDA device 0 (GeForce GTX 1080 Ti): 898 iterations, 6.276s init, 259.093s render

    2020-04-22 22:52:12.200 Iray INFO - module:category(IRAY:RENDER):   1.0   IRAY   rend info : CUDA device 1 (GeForce GTX 1080 Ti): 902 iterations, 6.277s init, 259.691s render

    So I did render faster here, but by about 6 seconds or so. 4.12.1.109 seems to take much longer to load the scene into the GPU. But that's not all, you can look at my old times I posted before, and the render times are about 30 seconds faster, too. So something has happened, probably with drivers, to effect speed. But we need newer drivers to access certain features of Iray and dforce. It also seems like my GPUs are topping out at slightly lower clocks than I remember, but I tried overclocking and that didn't make much difference, about 4-6 seconds, not 30. I was getting 229 second renders before.

    Post edited by outrider42 on
  • RayDAntRayDAnt Posts: 1,120

    I did some new tests with the 4.12.1.109 beta since Iray has changed a bit. I also wanted to see how much VRAM was being used, not just by the scene, but by the new SSID feature, and also denoising. And to see what impact on render speed those have, as they are supposed to hurt rendering speed. I found out something else, my times are slower overall, however it is not the new Iray. I still have 4.11 tucked away on my system, and my times are slower on it as well. This is also with 2 different drivers. We'll see below.

    First thought that comes to mind is that it could be loss of performance stemming from all those OS updates for Spectre-like security vulnerabilities that seem to pop up every two weeks or so. While it is true that Iray rendering performance iwith GPUs s almost entirely GPU limited, CPU still plays a fundamental role in scheduling which GPU processes which iterations in a multi-GPU system like yours. Meaning that a high performance set of cards could expose sublte changes in CPU performance on lower-tier previous-gen CPUs .

     

    The first test with driver 441.66. Everything about my system is the same as before.

    Windows 10 1903

    CPU: i5 4690K

    GPU #1:  EVGA 1080ti SC2

    GPU #2: MSI 1080ti Gaming  <--this is my display, yes, I use GPU 2 for display.

    RAM 32GB HyperX

    OS Drive Samsung 860 EVO 1TB

    Asset Drive: Samsung 860 EVO 1TB and WB 4TB Black HDD

    2020-04-22 16:41:55.940 Total Rendering Time: 4 minutes 38.7 seconds

    2020-04-22 16:43:20.426 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2020-04-22 16:43:20.426 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce GTX 1080 Ti): 903 iterations, 7.075s init, 266.593s render

    2020-04-22 16:43:20.426 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (GeForce GTX 1080 Ti): 897 iterations, 7.362s init, 266.475s render

    VRAM: GPU1 2885  GPU2 3415 MB

    That establishes a baseline for VRAM use. You can see that my display GPU is using 530 MB more than the non display card. The non display card had 0 MB before starting Daz.

    This next test is with SSID on.

    2020-04-22 16:50:32.580 Total Rendering Time: 4 minutes 38.52 seconds

    2020-04-22 16:51:01.109 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2020-04-22 16:51:01.110 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce GTX 1080 Ti): 899 iterations, 6.446s init, 268.581s render

    2020-04-22 16:51:01.110 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (GeForce GTX 1080 Ti): 901 iterations, 6.730s init, 268.494s render

    VRAM: GPU1 3465  GPU2 3436

    So here you can see that the time is 2 seconds slower, a very minor drop. The VRAM used is very odd. Somehow the first GPU, which is not the display, now uses slightly more VRAM than the display GPU! This is a most odd finding. However, the display GPU is using just a tiny amount of VRAM for SSID, just 21 MB, while the non display jumped 580 MB.

    Overall though, notice how balanced they are. I am thinking that the SSID is smart enough to balance the load between the GPUs so that they use similar amounts. And notice that in the first test, the non display GPU rendered 6 more frames, but in this test, the display GPU actually rendered 2 more frames. I believe this is due to the SSID running more on the display GPU since that is where most of the VRAM went for SSID. Interesting stuff. I saw no difference in the actual renders between these two tests, which makes sense.

    Check the official Iray documentation for SSID. SSID runs exclusively on the first GPU found by your system in multi-GPU setups (presumably whichever one is in the PCI-E slot closest to the CPU.) Hence the memory usage and individual GPU productivity patterns you are observing.

    The fact that memory usage ended up almost identical between your GPUs is honestly probably just a coincidence (granted, a kind of cool one.) Mostl ikely if you were to go into Windows display settings and change your primary monitor(s)' resolution and re-ran these same tests, you would find that that memory usage parity would no longer exist.

     

    This next test is with the good old denoiser. You either love it or hate it.

    Fwiw the denoiser in this release has been revamped versus previous releases:

    Deep Learning based Denoiser
    • Improved quality with new training methodology, better preservation of brightness and details for low sample images.

    Meaning that it could give radically better (or worse) results this time around. Some testing for that is likely in order...

     

    Lets see how it effects performance.

    2020-04-22 16:58:58.783 Total Rendering Time: 4 minutes 37.50 seconds

    2020-04-22 16:59:27.262 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2020-04-22 16:59:27.262 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce GTX 1080 Ti): 902 iterations, 5.943s init, 268.297s render

    2020-04-22 16:59:27.262 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (GeForce GTX 1080 Ti): 898 iterations, 6.328s init, 268.191s render

    VRAM: GPU1 3225 GPU2 3436

    Well, it actually clocked in a tiny fraction faster than SSID, though a still second or so slower than no post. Still, this difference is so minor as to be within a margin of error. It could depend on the scene. The VRAM used is less than the SSID, at least for the non display card. The display card is actually identical. And this is pretty interesting, because again, most of the VRAM that the denoiser asked for came from just one GPU. And this time the non display GPU rendered 4 more frames than the other.

    Another note, the non display GPU settled at 1911 Mhz, while the display GPU settled at 1936. So by clockspeed, the display GPU should be rendering more frames most of the time, they are both 1080tis. So I imagine being the display is costing it a few frames allowing the non display to 'win'. In my experience, the non display almost always wins, by just a few frames.

    At this point I updated my drivers to 442.92. Not much changed, in fact they are so similar I don't think I need to post them. But I had noticed my times were slower than before. I was hitting right at 4 minutes before on this scene. I still have 4.11 on my PC, so I fired it up.

    4.11.0.0236 optix on This predates the RTX update.

    2020-04-22 22:51:51.224 Total Rendering Time: 4 minutes 28.61 seconds

    2020-04-22 22:52:12.199 Iray INFO - module:category(IRAY:RENDER):   1.0   IRAY   rend info : Device statistics:

    2020-04-22 22:52:12.199 Iray INFO - module:category(IRAY:RENDER):   1.0   IRAY   rend info : CUDA device 0 (GeForce GTX 1080 Ti): 898 iterations, 6.276s init, 259.093s render

    2020-04-22 22:52:12.200 Iray INFO - module:category(IRAY:RENDER):   1.0   IRAY   rend info : CUDA device 1 (GeForce GTX 1080 Ti): 902 iterations, 6.277s init, 259.691s render

    So I did render faster here, but by about 6 seconds or so. 4.12.1.109 seems to take much longer to load the scene into the GPU. But that's not all, you can look at my old times I posted before, and the render times are about 30 seconds faster, too. So something has happened, probably with drivers, to effect speed. But we need newer drivers to access certain features of Iray and dforce. It also seems like my GPUs are topping out at slightly lower clocks than I remember, but I tried overclocking and that didn't make much difference, about 4-6 seconds, not 30. I was getting 229 second renders before.

     

  • outrider42outrider42 Posts: 3,679
    edited April 2020

    Well, there is an easy way to test the idea that the security updates harmed the CPU scheduling for multi GPU...just run a test for single GPU. If the result is similar to the previous tests, then that would prove it. I have a test from November 2019 in my notepad testing the general release of 4.12. So here we go.

    For this test I am using the EVGA card by itself. It is the non display GPU plugged into the first port.

    2020-04-23 17:07:05.959 Total Rendering Time: 8 minutes 50.99 seconds

    2020-04-23 17:07:39.960 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2020-04-23 17:07:39.960 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce GTX 1080 Ti): 1800 iterations, 6.107s init, 520.741s render

    VRAM: 2885 MB

    Here is the result from November with driver 436.02

    2019-11-09 00:22:21.167 Total Rendering Time: 7 minutes 38.53 seconds

    2019-11-09 00:22:28.149 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2019-11-09 00:22:28.149 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce GTX 1080 Ti): 1800 iterations, 5.600s init, 449.931s render

    There is large 30 second difference between the two. Plus there is a much longer loading time on top of that, as the test was overall 1 minute 12 seconds faster.

    So I do not believe security updates are at play here, because single GPU use should not be impacted, certainly not by this much. I believe that most GPU rendering is not too effected by these updates. My previous tests here was only back in November, many updates had already came through at that point.

    I also ran the MSI by itself, it got similar results, it too is about 30 seconds slower than the November test.

    2020-04-23 18:32:38.351 Total Rendering Time: 8 minutes 52.98 seconds

    2020-04-23 18:39:29.863 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2020-04-23 18:39:29.863 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (GeForce GTX 1080 Ti): 1800 iterations, 6.452s init, 522.513s render

    Post edited by outrider42 on
  • RayDAntRayDAnt Posts: 1,120

    @outrider42 how odd. I just went back and re-ran the benchmark in 4.12 under an identical configuration to the last time I did on my Titan RTX based system (with the exception of driver version) and got virtually the same (actually a few seconds better) rendering performance. And my loading times are now several seconds faster as well. Perhaps there's been a driver-level regression in performance for GTX and not RTX cards since then? You could try a driver rollback and see if that changes anything. The next thing would be to see if other GTX users are experiencing the same thing.

  • outrider42outrider42 Posts: 3,679
    edited April 2020
    RayDAnt said:

    @outrider42 how odd. I just went back and re-ran the benchmark in 4.12 under an identical configuration to the last time I did on my Titan RTX based system (with the exception of driver version) and got virtually the same (actually a few seconds better) rendering performance. And my loading times are now several seconds faster as well. Perhaps there's been a driver-level regression in performance for GTX and not RTX cards since then? You could try a driver rollback and see if that changes anything. The next thing would be to see if other GTX users are experiencing the same thing.

    It is quite odd. I was downloading a previous driver when I saw your post, LOL. So I rolled back to driver 436.3. I booted up 4.12.0.86, which is the one from November. In fact, my full post with my marks is on page 4 of this thread. So here is my new test using driver 436.3:

    2020-04-23 19:20:47.762 Total Rendering Time: 4 minutes 4.95 seconds

    2020-04-23 19:20:55.951 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2020-04-23 19:20:55.951 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce GTX 1080 Ti): 907 iterations, 7.064s init, 233.812s render

    2020-04-23 19:20:55.951 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (GeForce GTX 1080 Ti): 893 iterations, 7.312s init, 233.041s render

    And there it is. Just a few seconds off my November time. Not only is the render time faster, but so is the load time.

    So is this really a driver thing? It looks that way. Remember I tested this with 2 different drivers, but not the one I used back in November, so when did this happen? I would really like some other GTX users to test this out and see if anybody else has experienced this with the newer Nvidia drivers. 

    BTW the, if there is anybody who is not sure how to install older drivers, Nvidia has a page for that here. https://www.nvidia.com/Download/Find.aspx?lang=en-us I'll also take this as yet another opportunity to remind the people running Daz that, hey, they need to provide a place for past versions of Studio like Nvidia does here. It is good for customers. Does the beta even run on the driver I used? Or will it crash? See, if I am unhappy with the beta, for any reason, I should be able to revert it to the previous version. I have some different versions of Daz, but that is only because I planned ahead. The vast majority of users do not even know they can do this.

    *EDIT: Daz 4.12.1.109 will boot up with driver 436.3, but Iray will not work. So you cannot render anything.

    Post edited by outrider42 on
  • RayDAntRayDAnt Posts: 1,120
    edited April 2020
    RayDAnt said:

    @outrider42 how odd. I just went back and re-ran the benchmark in 4.12 under an identical configuration to the last time I did on my Titan RTX based system (with the exception of driver version) and got virtually the same (actually a few seconds better) rendering performance. And my loading times are now several seconds faster as well. Perhaps there's been a driver-level regression in performance for GTX and not RTX cards since then? You could try a driver rollback and see if that changes anything. The next thing would be to see if other GTX users are experiencing the same thing.

    It is quite odd. I was downloading a previous driver when I saw your post, LOL. So I rolled back to driver 436.3. I booted up 4.12.0.86, which is the one from November. In fact, my full post with my marks is on page 4 of this thread. So here is my new test using driver 436.3:

    2020-04-23 19:20:47.762 Total Rendering Time: 4 minutes 4.95 seconds

    2020-04-23 19:20:55.951 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2020-04-23 19:20:55.951 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce GTX 1080 Ti): 907 iterations, 7.064s init, 233.812s render

    2020-04-23 19:20:55.951 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (GeForce GTX 1080 Ti): 893 iterations, 7.312s init, 233.041s render

    And there it is. Just a few seconds off my November time. Not only is the render time faster, but so is the load time.

    So is this really a driver thing? It looks that way. Remember I tested this with 2 different drivers, but not the one I used back in November, so when did this happen? I would really like some other GTX users to test this out and see if anybody else has experienced this with the newer Nvidia drivers. 

    BTW the, if there is anybody who is not sure how to install older drivers, Nvidia has a page for that here. https://www.nvidia.com/Download/Find.aspx?lang=en-us I'll also take this as yet another opportunity to remind the people running Daz that, hey, they need to provide a place for past versions of Studio like Nvidia does here. It is good for customers. Does the beta even run on the driver I used? Or will it crash? See, if I am unhappy with the beta, for any reason, I should be able to revert it to the previous version. I have some different versions of Daz, but that is only because I planned ahead. The vast majority of users do not even know they can do this.

    *EDIT: Daz 4.12.1.109 will boot up with driver 436.3, but Iray will not work. So you cannot render anything.

    Are you on Studio or Game Ready drivers? I've been running all my systems on Game Ready drivers since my past testing has shown that performance (and even the binary files between them) were virtually identical. And Game Ready get bugfixed/updated more often. However, it has been at least 3 months since I last compared them, and it is possible that one could now offer benefits over the other in Iray specific workloads. Especially since they just updated the minmimum driver requirement for Iray with this latest release.

    Post edited by RayDAnt on
  • outrider42outrider42 Posts: 3,679
    edited April 2020

    The last one was a Studio driver, though I haven't seen any difference between them. There have been a couple of times when a Game Ready driver caused some unexpected issues, though not for me, so that has happened in the past. I'll try a newer Game Ready to see if its any different.

    Game Ready drivers are indeed updated more frequently, but they those updates are only targeted at optimizing performance specifically for the newest big games. Like when the new Doom came out, they released a driver update just for it.

    Sometimes you just get unlucky and a driver doesn't quite mesh. I recall a youtuber who had an issue with a driver. At the time, they thought that Nvidia drivers had cost them performance, but they after a couple reinstalls it finally performed like it was supposed to.

    So here is a test with the latest Game Ready driver, 445.87

    2020-04-23 21:48:46.128 Total Rendering Time: 4 minutes 38.59 seconds

    2020-04-23 21:48:51.208 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2020-04-23 21:48:51.208 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce GTX 1080 Ti): 903 iterations, 8.376s init, 265.788s render

    2020-04-23 21:48:51.208 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (GeForce GTX 1080 Ti): 897 iterations, 8.675s init, 265.548s render

    All I can say is that this is not looking good for GTX at this point. This brings me right back to where I was before. So that is 3 different drivers that all make my GPUs run 30+ seconds slower than I did with drivers 436.3 and previous. I find it very hard to believe that 3 drivers would present this same performance. If this is a trend that effects GTX, that is very bad news for both Daz and its GTX users. Obviously, Daz would like its users to upgrade, but what if that causes a drop in performance? Have you tried your test with your 1050 based machine? If I am losing 30 seconds with 1080tis, I hate to think what others might get with lesser cards.

    I am going to try one last time, using the bare minimum driver for the new beta features. The Beta thread lists 441.22 as being the minimum for GTX cards, though I do not see a driver on the list with that number. So I am going to try 441.28. We shall see.

    Post edited by outrider42 on
  • RayDAntRayDAnt Posts: 1,120
    edited April 2020

    Finally got around to benchmarking the latest DS release on my main rendering system.

    System Configuration
    System/Motherboard: Gigabyte Z370 Aorus Gaming 7
    CPU: Intel i7-8700K @ stock (MCE enabled)
    GPU: Nvidia Titan RTX @ stock (watercooled)
    System Memory: Corsair Vengeance LPX 32GB DDR4 @ 3000Mhz
    OS Drive: Samsung Pro 970 512GB NVME SSD
    Asset Drive: Sandisk Extreme Portable SSD 1TB
    Operating System: Windows 10 Pro 1909
    Nvidia Drivers Version: 445.87
    Daz Studio Version: 4.12.1.109 Beta x64


    Benchmark Results: Titan RTX (TCC)
    Total Rendering Time: 4 minutes 7.88 seconds
    CUDA device 0 (TITAN RTX): 1800 iterations, 2.991s init, 242.131s render
    Iteration Rate: (1800 / 242.131) = 7.434 iterations per second
    Loading Time: ((0 + 4 * 60 + 7.88) - 242.131) = 5.749 seconds

    Benchmark Results: Titan RTX (WDDM, used for display)
    Total Rendering Time: 4 minutes 16.24 seconds
    CUDA device 0 (TITAN RTX): 1800 iterations, 2.240s init, 251.320s render
    Iteration Rate: (1800 / 251.320) = 7.162 iterations per second
    Loading Time: ((0 + 4 * 60 + 16.24) - 251.320) = 4.920 seconds

     

    Benchmark Results: Titan RTX (TCC) + i7-8700K (used for display)
    Total Rendering Time: 4 minutes 20.38 seconds
    CUDA device 0 (TITAN RTX): 1684 iterations, 2.433s init, 255.144s render
    CPU: 116 iterations, 2.201s init, 255.544s render
    Iteration Rate: (1800 / 255.544) = 7.044 iterations per second
    Loading Time: ((0 + 4 * 60 + 20.38) - 255.544) = 4.836 seconds

    Benchmark Results: Titan RTX (WDDM, used for display) + i7-8700K
    Total Rendering Time: 4 minutes 16.74 seconds
    CUDA device 0 (TITAN RTX): 1690 iterations, 2.641s init, 251.068s render
    CPU: 110 iterations, 2.205s init, 251.716s render
    Iteration Rate: (1800 / 251.716) = 7.151 iterations per second
    Loading Time: ((0 + 4 * 60 + 16.74) - 251.716) = 5.024 seconds

     

    Benchmark Results: i7-8700K
    Total Rendering Time: 1 hours 36.29 seconds
    CPU: 1800 iterations, 2.372s init, 3631.272s render
    Iteration Rate: (1800 / 3631.272) = 0.496 iterations per second
    Loading Time: ((1 * 3600 + 36.29) - 3631.272) = 5.018 seconds

    Benchmark Results: i7-8700K (used for display)
    Total Rendering Time: 1 hours 1 minutes 41.67 seconds
    CPU: 1800 iterations, 2.522s init, 3696.342s render
    Iteration Rate: (1800 / 3696.342) = 0.487 iterations per second
    Loading Time: ((1 * 60 + 1 * 60 + 36.29) - 3696.342) = 5.328 seconds

     

    Post edited by RayDAnt on
  • RayDAntRayDAnt Posts: 1,120
    edited April 2020

    @outrider42 are you sure your unexpectedly longer benchmark times weren't actually from 4.12.1.109 (maybe you referred to the Release channel log file rather than the Beta one?) I ask because my results with 4.12.1.109 indicate that it is approximately 8% slower at achieving a set amount of iterations than the previous release was - which doesn't sound far off from the amount of apparentt rendering performance loss you are currently investigating.

    If so, keep in mind that 4.12.1.109 taking a signficantly longer (or shorter for that matter) amount of time to complete this benchmark than other releaes isn't in itself meaningful. Because every new version of Iray has a different distribution of processing time to per-iteration completion and overall image convergance rate stemming from featureset/rendering algorithm tweaks, the raw statistics generated by running the benchmark in this thread are only technically comparable to each other if they come from the same version of Iray. Hence why the main results table has a separate column for each Iray release.

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

    Since yesterday I have rendered in 4.11 and now 3 versions of 4.12. And I have tried at this point 5 different drivers. My peak performance was with driver 436.3, every driver after that has a drop off in all of the versions of Daz available to that driver.

    Your time is not effected as much, and convergence metrics can and do change, but still you posted the worst Iray RTX bench you have had so far. And I think 8% is outside the margin of error. Though that only happened with 109 for you. You could save the images and compare them, assuming you have a saved render from previous benches. I do not see any difference in the images I have. The noise is not more or less. If anything, looking at the green ball at the bottom left corner, maybe my eyes are wrong, but the 109 version might have a few more red dots present on that ball than the images rendered in previous versions do. That includes my images from November. 

    Do you have other versions of Daz installed? The general release is still behind the beta, so that might be an interesting test. If your time is as fast as before in the general release, then that points to 109 for you. But if your time is 8% slower in the general release like it is here, then the driver might be an issue.

  • RayDAntRayDAnt Posts: 1,120
    edited April 2020

    Do you have other versions of Daz installed? The general release is still behind the beta, so that might be an interesting test. If your time is as fast as before in the general release, then that points to 109 for you. But if your time is 8% slower in the general release like it is here, then the driver might be an issue.

    I am in the habit of archiving both the DIM installer and installation directory of each and every outdated DS release before installing a new one. Disk space is cheap ;)

    Re-benched 4.12.0.086 (aka Iray 2019.1.3) and got identical to slightly better (by about 3 seconds) results to my previous bench of that version with the drivers available at that time. So everything from my end points to that 8% variance being DS/Iray version related and not driver/OS/something else related (as your variance seems to be.)

    Also, it's a purely technical point (doubt there are many other statisticians reading this thread, but who knows) but the 8% shift in render times I'm describing is a result of expected variance between different test cases - not a margin of error. If I were to do multiple benchmark passes of the same DS version under identical test conditions and get up to an 8% difference in results between them, then that would indicate an 8% margin of error. I just did a couple of extra passes to verify, and the margin of error I am currently seeing is around 1-2%.

    Post edited by RayDAnt on
  • Jason GalterioJason Galterio Posts: 2,562

    Did an upgrade. Thought I would run this again to see the results:

    System Configuration
    System/Motherboard: Asus TUF Z390-Plus
    CPU: Intel Core i9-9900k 3.6 GHz
    GPU: RTX 2080 Super
    System Memory: 32 GB DDR4 2666 MHz
    OS Drive: Intel M.2 1 TB NVMe SSD
    Asset Drive: Same
    Operating System: Windows 10 Pro
    Nvidia Drivers Version: Game Ready 445.87
    Daz Studio Version: 4.12.1.117
    Optix Prime Acceleration: N/A

    2020-05-15 17:05:08.199 Finished Rendering
    2020-05-15 17:05:08.227 Total Rendering Time: 5 minutes 22.7 seconds

    2020-05-15 17:05:39.757 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:
    2020-05-15 17:05:39.757 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 2080 SUPER):   1800 iterations, 6.378s init, 313.040s render

    Iteration Rate: 5.75 iterations per second
    Loading Time: 9.96 seconds

  • RayDAntRayDAnt Posts: 1,120

    Finally got around to doing a fresh test on my Surface Book 2.

    System Configuration
    System/Motherboard: Microsoft Surface Book 2
    CPU: Intel i7-8650U @ stock
    GPU: Nvidia GTX 1050 2GB @ stock
    System Memory: 16GB DDR3 @ 1867Mhz
    OS Drive: Samsung OEM 512GB NVME SSD
    Asset Drive: Sandisk Extreme 1TB External SSD
    Operating System: W10 1909
    Nvidia Drivers Version: 442.92
    Daz Studio Version: 4.12.2.006 Beta x64

     

    Benchmark Results: GTX 1050 2GB
    Total Rendering Time: 46 minutes 32.33 seconds
    CUDA device 0 (GeForce GTX 1050): 1800 iterations, 5.428s init, 2783.170s render
    Iteration Rate: (1800 / 2783.170) = 0.647 iterations per second
    Loading Time: ((0 + 2760 + 32.33) - 2783.170) = 9.160 seconds

  • outrider42outrider42 Posts: 3,679
    edited May 2020
    RayDAnt said:

    Finally got around to doing a fresh test on my Surface Book 2.

    System Configuration
    System/Motherboard: Microsoft Surface Book 2
    CPU: Intel i7-8650U @ stock
    GPU: Nvidia GTX 1050 2GB @ stock
    System Memory: 16GB DDR3 @ 1867Mhz
    OS Drive: Samsung OEM 512GB NVME SSD
    Asset Drive: Sandisk Extreme 1TB External SSD
    Operating System: W10 1909
    Nvidia Drivers Version: 442.92
    Daz Studio Version: 4.12.2.006 Beta x64

     

    Benchmark Results: GTX 1050 2GB
    Total Rendering Time: 46 minutes 32.33 seconds
    CUDA device 0 (GeForce GTX 1050): 1800 iterations, 5.428s init, 2783.170s render
    Iteration Rate: (1800 / 2783.170) = 0.647 iterations per second
    Loading Time: ((0 + 2760 + 32.33) - 2783.170) = 9.160 seconds

    Just thought I would point out that time is also slower than previous tests you did. Like 8 minutes slower. That's the slowest time you have recorded with Daz 4.12 on your Surface, period. All your other 4.12 tests were right around 38 minutes.

    I don't think this is a coincidence. I am rendering slower, you are rendering slower with 2 different machines. And Jason's 2080 Super also rendered slower than the only existing 2080 Super benched so far, which was a previous version of Iray. Its not a huge loss, but it is a solid 30 seconds slower. And that's the device render time, not the total time. Out of a 5 minute render that is a sizable chunk of time. In fact, he just rendered the same iteration rate as the regular 2080 did!

    I think this is a legit trend. I would really like to see more people run new benchmarks on the updated Iray!

    Please, anybody who hasn't ran the bench in while, I think it would be good to run one now.

    Post edited by outrider42 on
  • cajhincajhin Posts: 154
    edited May 2020

    Updated from Radeon to RTX 2070 Super, just for Iray. It's a game changer how much faster this is compared to CPU. Happy :-)

    The card was a whopping (for my taste) €530, but well I don't spend much these days...

     

    System Configuration
    System/Motherboard: ASRock X570M Pro4 mATX
    CPU: Ryzen 7 3700X 8x 3.60GHz / stock
    GPU: PNY GEFORCE RTX 2070 SUPER V2 (VCG20708SBLPPB) / stock 1605/1170 MHz
    System Memory:32GB Crucial Ballistix Sport LT V2 / stock 3200 CL16
    OS Drive: Sabrent Rocket 1TB Nvme PCIe 4.0 M.2, 5 GBps read seq
    Asset Drive: Same
    Operating System: Win10 Pro 1909 18363.836
    Nvidia Drivers Version: 442.92 Studio
    Daz Studio Version: 4.12.1.117 x64
    Optix Prime Acceleration: NA

    Benchmark Results
    Total Rendering Time: 5 minutes 44.2 seconds
    CUDA device 0 (GeForce RTX 2070 SUPER): 1800 iterations, 2.170s init, 339.057s render
    Iteration Rate: 5.31 it/s (1800 / 339.06)
    Loading Time: 5.1 s (344.2 - 339.1)

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

    Updated from Radeon to RTX 2070 Super, just for Iray. It's a game changer how much faster this is compared to CPU. Happy :-)

    The card was a whopping (for my taste) €530, but well I don't spend much these days...

     

    System Configuration
    System/Motherboard: ASRock X570M Pro4 mATX
    CPU: Ryzen 7 3700X 8x 3.60GHz / stock
    GPU: PNY GEFORCE RTX 2070 SUPER V2 (VCG20708SBLPPB) / stock 1605/1170 MHz
    System Memory:32GB Crucial Ballistix Sport LT V2 / stock 3200 CL16
    OS Drive: Sabrent Rocket 1TB Nvme PCIe 4.0 M.2, 5 GBps read seq
    Asset Drive: Same
    Operating System: Win10 Pro 1909 18363.836
    Nvidia Drivers Version: 442.92 Studio
    Daz Studio Version: 4.12.1.117 x64
    Optix Prime Acceleration: NA

    Benchmark Results
    Total Rendering Time: 5 minutes 44.2 seconds
    CUDA device 0 (GeForce RTX 2070 SUPER): 1800 iterations, 2.170s init, 339.057s render
    Iteration Rate: 5.31 it/s (1800 / 339.06)
    Loading Time: 5.1 s (344.2 - 339.1)

    <span style="color:rgb(129, 162, 190)">Jason</span> <span style="color:rgb(129, 162, 190)">Galterio</span>     <span style="color:rgb(129, 162, 190)">W10</span> 1903     441<span style="color:rgb(222, 147, 95)">.20</span>     4<span style="color:rgb(222, 147, 95)">.12</span><span style="color:rgb(222, 147, 95)">.0</span><span style="color:rgb(222, 147, 95)">.086</span> <span style="color:rgb(129, 162, 190)">x64</span>     <span style="color:rgb(129, 162, 190)">NA</span>        5 <span style="color:rgb(129, 162, 190)">minutes</span> 1<span style="color:rgb(222, 147, 95)">.95</span> <span style="color:rgb(129, 162, 190)">seconds</span>     290<span style="color:rgb(222, 147, 95)">.227</span>           06<span style="color:rgb(222, 147, 95)">.202</span>             11<span style="color:rgb(222, 147, 95)">.723</span>

    So far not a single time posted with the new update is as fast as times posted before. Here Jason's 2070 Super has a device render time of 290 seconds, compared to 339. That's a full 50 seconds slower between updates! His driver was right before this stuff started happening. Granted, Jason's time seems to be abnormally fast as it came really close to a 2080 Super. Now obviously things can change and what an iteration may be can change. But I just don't believe that Iray would change the work done in an iteration that much. If you ask me, Iray performance has been gimped by recent drivers. Like I said before, I was able to get my speed back by rolling back drivers and using a previous version of Daz Studio I still had. When I updated drivers, I lost speed, even in that same previous version of Studio. That points to a driver issue, and everybody so far has seen a drop in render speed.

    Also, when it comes to Iray, I have seen the times it benches to be extremely similar to each other. It is very rare for my PC to suddenly drop or gain any percentage of time over different bench runs, even 8%. Nearly all my bench runs are just a few seconds apart at most, until the recent drivers. It could just be that the Titan for whatever reason is less effected than other cards, it is a different chipset from everybody else that has tested since the new drivers and Daz updates.

    There is another driver update, so I'll try that.

  • RayDAntRayDAnt Posts: 1,120
    edited June 2020
    cajhin said:

    Updated from Radeon to RTX 2070 Super, just for Iray. It's a game changer how much faster this is compared to CPU. Happy :-)

    The card was a whopping (for my taste) €530, but well I don't spend much these days...

     

    System Configuration
    System/Motherboard: ASRock X570M Pro4 mATX
    CPU: Ryzen 7 3700X 8x 3.60GHz / stock
    GPU: PNY GEFORCE RTX 2070 SUPER V2 (VCG20708SBLPPB) / stock 1605/1170 MHz
    System Memory:32GB Crucial Ballistix Sport LT V2 / stock 3200 CL16
    OS Drive: Sabrent Rocket 1TB Nvme PCIe 4.0 M.2, 5 GBps read seq
    Asset Drive: Same
    Operating System: Win10 Pro 1909 18363.836
    Nvidia Drivers Version: 442.92 Studio
    Daz Studio Version: 4.12.1.117 x64
    Optix Prime Acceleration: NA

    Benchmark Results
    Total Rendering Time: 5 minutes 44.2 seconds
    CUDA device 0 (GeForce RTX 2070 SUPER): 1800 iterations, 2.170s init, 339.057s render
    Iteration Rate: 5.31 it/s (1800 / 339.06)
    Loading Time: 5.1 s (344.2 - 339.1)

    <span style="color:rgb(129, 162, 190)">Jason</span> <span style="color:rgb(129, 162, 190)">Galterio</span>     <span style="color:rgb(129, 162, 190)">W10</span> 1903     441<span style="color:rgb(222, 147, 95)">.20</span>     4<span style="color:rgb(222, 147, 95)">.12</span><span style="color:rgb(222, 147, 95)">.0</span><span style="color:rgb(222, 147, 95)">.086</span> <span style="color:rgb(129, 162, 190)">x64</span>     <span style="color:rgb(129, 162, 190)">NA</span>        5 <span style="color:rgb(129, 162, 190)">minutes</span> 1<span style="color:rgb(222, 147, 95)">.95</span> <span style="color:rgb(129, 162, 190)">seconds</span>     290<span style="color:rgb(222, 147, 95)">.227</span>           06<span style="color:rgb(222, 147, 95)">.202</span>             11<span style="color:rgb(222, 147, 95)">.723</span>

    So far not a single time posted with the new update is as fast as times posted before. Here Jason's 2070 Super has a device render time of 290 seconds, compared to 339. That's a full 50 seconds slower between updates! His driver was right before this stuff started happening. Granted, Jason's time seems to be abnormally fast as it came really close to a 2080 Super. Now obviously things can change and what an iteration may be can change. But I just don't believe that Iray would change the work done in an iteration that much. If you ask me, Iray performance has been gimped by recent drivers. Like I said before, I was able to get my speed back by rolling back drivers and using a previous version of Daz Studio I still had. When I updated drivers, I lost speed, even in that same previous version of Studio. That points to a driver issue, and everybody so far has seen a drop in render speed.

    Also, when it comes to Iray, I have seen the times it benches to be extremely similar to each other. It is very rare for my PC to suddenly drop or gain any percentage of time over different bench runs, even 8%. Nearly all my bench runs are just a few seconds apart at most, until the recent drivers. It could just be that the Titan for whatever reason is less effected than other cards, it is a different chipset from everybody else that has tested since the new drivers and Daz updates.

    There is another driver update, so I'll try that.

    Which driver version did you roll back to that seemingly restored performance even on the latest Iray build? (aka the latest DS release.) I'll try rolling back to it with my Titan RTX and see if I get similar results.

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

    Updated from Radeon to RTX 2070 Super, just for Iray. It's a game changer how much faster this is compared to CPU. Happy :-)

    The card was a whopping (for my taste) €530, but well I don't spend much these days...

     

    System Configuration
    System/Motherboard: ASRock X570M Pro4 mATX
    CPU: Ryzen 7 3700X 8x 3.60GHz / stock
    GPU: PNY GEFORCE RTX 2070 SUPER V2 (VCG20708SBLPPB) / stock 1605/1170 MHz
    System Memory:32GB Crucial Ballistix Sport LT V2 / stock 3200 CL16
    OS Drive: Sabrent Rocket 1TB Nvme PCIe 4.0 M.2, 5 GBps read seq
    Asset Drive: Same
    Operating System: Win10 Pro 1909 18363.836
    Nvidia Drivers Version: 442.92 Studio
    Daz Studio Version: 4.12.1.117 x64
    Optix Prime Acceleration: NA

    Benchmark Results
    Total Rendering Time: 5 minutes 44.2 seconds
    CUDA device 0 (GeForce RTX 2070 SUPER): 1800 iterations, 2.170s init, 339.057s render
    Iteration Rate: 5.31 it/s (1800 / 339.06)
    Loading Time: 5.1 s (344.2 - 339.1)

    <span style="color:rgb(129, 162, 190)">Jason</span> <span style="color:rgb(129, 162, 190)">Galterio</span>     <span style="color:rgb(129, 162, 190)">W10</span> 1903     441<span style="color:rgb(222, 147, 95)">.20</span>     4<span style="color:rgb(222, 147, 95)">.12</span><span style="color:rgb(222, 147, 95)">.0</span><span style="color:rgb(222, 147, 95)">.086</span> <span style="color:rgb(129, 162, 190)">x64</span>     <span style="color:rgb(129, 162, 190)">NA</span>        5 <span style="color:rgb(129, 162, 190)">minutes</span> 1<span style="color:rgb(222, 147, 95)">.95</span> <span style="color:rgb(129, 162, 190)">seconds</span>     290<span style="color:rgb(222, 147, 95)">.227</span>           06<span style="color:rgb(222, 147, 95)">.202</span>             11<span style="color:rgb(222, 147, 95)">.723</span>

    So far not a single time posted with the new update is as fast as times posted before. Here Jason's 2070 Super has a device render time of 290 seconds, compared to 339. That's a full 50 seconds slower between updates! His driver was right before this stuff started happening. Granted, Jason's time seems to be abnormally fast as it came really close to a 2080 Super. Now obviously things can change and what an iteration may be can change. But I just don't believe that Iray would change the work done in an iteration that much. If you ask me, Iray performance has been gimped by recent drivers. Like I said before, I was able to get my speed back by rolling back drivers and using a previous version of Daz Studio I still had. When I updated drivers, I lost speed, even in that same previous version of Studio. That points to a driver issue, and everybody so far has seen a drop in render speed.

    Also, when it comes to Iray, I have seen the times it benches to be extremely similar to each other. It is very rare for my PC to suddenly drop or gain any percentage of time over different bench runs, even 8%. Nearly all my bench runs are just a few seconds apart at most, until the recent drivers. It could just be that the Titan for whatever reason is less effected than other cards, it is a different chipset from everybody else that has tested since the new drivers and Daz updates.

    There is another driver update, so I'll try that.

    Which driver version did you roll back to that seemingly restored performance even on the latest Iray build? (aka the latest DS release.) I'll try rolling back to it with my Titan RTX and see if I get similar results.

    That's the problem, no driver I have tried restores performance in the latest build because they are not compatible, I tried to make that clear. But I can restore performance in previous versions of Daz Studio with drivers that predate the ones required for 109+.

    To be clear: with any newer driver (likely 442.28 and up) I lose performance even when I try older versions of Daz. I even tried a version of 4.11, and found my render speeds were down from what I had before when using new drivers. When I roll back drivers to say, 436, I can run 4.11 and some 4.12 at my original render speeds. Obviously that driver does not function for 4.12.1.109 and up. The ones I can run all run faster with driver 436.

    BTW, I ran a new bench with the latest game driver, 446.14, and got basically identical results to the other new drivers. So I don't think I need to post that.

    But this should prove it is a driver problem, not a Iray problem. 4.11 uses an older version of Iray that predates RTX, and it is effected the same way that Iray RTX 2019 is by these drivers with a loss of speed. So I do not believe that Iray 2020 is the problem here, but since it requires newer drivers, it is effected. It is not a CPU security patch, and it is not a Windows update.

    The trouble here is that Daz Studio has new features, so this makes for a difficult choice, do you want to use the new features or sacrifice them for faster rendering? This is also going to heavily skew results. If this is not fixed by Nvidia, it could impact any new GPU releases and make them look less powerful if people compare them to older versions of Daz in this bench. I was hoping some more people would test out the new drivers. I would like to see one of the many 2080ti owners bench these new drivers.

    As a gamer, this also stings, because the newer drivers added a long requested feature, a frame rate limiter. You can find 3rd party software to do this, but it is much better to have this option in the GPU settings proper. There are some games that have no way to limit frame rates, and uncapped frame rates in certain games can be trouble, while also running your GPU very hot. (Far hotter than Iray ever would.) Adding a cap can stabilize performance. So while this is not related to Daz or Iray, that is a feature I'd like to keep.

    However I can imagine a lot of people would be willing to roll back and use older Daz Studios to keep that performance...assuming they have the previous versions since Daz refuses to keep any previous build active.

  • NVLinked result!
    (https://www.daz3d.com/forums/discussion/comment/5701921/#Comment_5701921)

    System Configuration
    System/Motherboard: ASRock X299 Extreme4
    CPU: Intel i7-7800X @ Stock
    GPU: GPU1 INNO3D RTX2080TI iCHILL Frostbite @ stock, GPU2 INNO3D RTX2080TI iCHILL Frostbite @ stock
    System Memory: 64GB Crucial Ballistix Elite DDR4 @ 2400
    OS Drive: Samsung 960 EVO 250GB NVMe SSD
    Asset Drive: 2x Crucial MX500 480GB (JBOD)
    Operating System: Windows 10 Pro x64 Build 1909
    Nvidia Drivers Version: SRD 442.92 WDDM
    Daz Studio Version: 4.12.1.117
    Optix Prime Acceleration: N/A

    Benchmark Results
    DAZ_STATS
    2020-06-05 20:30:56.807 Finished Rendering
    2020-06-05 20:30:56.864 Total Rendering Time: 2 minutes 22.21 seconds

    IRAY_STATS
    2020-06-05 20:33:36.187 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:
    2020-06-05 20:33:36.187 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 2080 Ti): 896 iterations, 3.909s init, 131.943s render
    2020-06-05 20:33:36.187 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (GeForce RTX 2080 Ti): 904 iterations, 3.843s init, 131.820s render

    Rendering Performance: [(896 + 904) / 131.943] = 13.642 iterations per second
    Loading Time: [(0 * 3600 + 2 * 60 + 22.21)] - 131.943 =  10.267 seconds

Sign In or Register to comment.