Daz Studio Iray - Rendering Hardware Benchmarking

1111214161745

Comments

  • cajhincajhin Posts: 154

    quick, unscientific datapoint:

    Render time went from 5:44 to 4:38 with the latest DAZ 4.14.
    (Rtx 2070S, Ryzen 3700x)

    5:44 is with DAZ 4.12, AND I had the exact same 5:44 back in May.
    Since then I've updated Windows and nVidia drivers, so apparently the speedup is tied only to DAZ/iRay version.

     

  • nonesuch00nonesuch00 Posts: 17,890
    edited January 2021
    cajhin said:

    quick, unscientific datapoint:

    Render time went from 5:44 to 4:38 with the latest DAZ 4.14.
    (Rtx 2070S, Ryzen 3700x)

    5:44 is with DAZ 4.12, AND I had the exact same 5:44 back in May.
    Since then I've updated Windows and nVidia drivers, so apparently the speedup is tied only to DAZ/iRay version.

     

    Thanks. I was wondering if the nVidia driver / Windows updates I had also done this week were also influential in my DAZ iRay renders speedups.

    Post edited by nonesuch00 on
  • RayDAntRayDAnt Posts: 1,120
    edited November 2020

    Holy crap, it went from 37 to over 48 iterations per second, that is a monstrous increase. The amount of performance you gained is more than a lot of people's total performance. Just think about that. to put that in perspective, a 3080 ran the bench at 12 iterations. This performance gain is like adding a 5th GPU...a 3080 class GPU...to your rig. This is seriously impressive, and the 4 GPU setup was already impressive!

    Since you have access, would you mind running a single 2080 Super so we can compare it to its previous time? This would also be helpful for looking at scaling. Since we have a single 3090 on 4.14, we would be able to calculate a theoretical peak, and compare that number to what you actually scored to see how many iterations may be lost. For all we know, the scaling might just have improved as well.

    We are going to have to draw a hard line on the benchmark chart now. I think we should carefully note all benchmarks that come after 4.14 because the gap is so big that it may confuse people who compare a 4.14 bench to a 4.12 bench, they are simply not comparable anymore. Obviously benches from one version to the next are generally not comparable, but in practice they really have been consistent over time, with only minor differences aside from when Iray switched to OptiX and RTX support.

    I am surprised that Daz has not advertised this speed upgrade. Advertising a speed increase in the update would get a lot of people on board the choo choo train.

    Just checked, and am also seeing a significant increase in rendering performance on my Titan RTX with the latest Daz Studio and the same drivers (DS 4.14.0.008 beta/non-beta and Nvidia driver version 456.38) over what was seen with the previous beta release (4.12.2.060) also under the same version of Windows. What's interesting here is that both 4.12.0.060 and 4.14.0.008 include the exact same version of Iray (version "Iray RTX 2020.1.1, build 334300.4226, 29 Sep 2020, nt-x86-64" as per the DS changelog - for a full breakdown of which DS versions contain what Iray version, see here.) Meaning that this speedup has to be rooted in updates made to Daz Studio itself (a first at the very least ni the lifetime of this benchmarking thread.) My best guess as to what is actually at the root of making the difference is this from the DS changelog:

    Private Beta (4.14.0.8)

    • Source maintenance

    • Made use of bump and normal maps together with the NVIDIA Iray renderer more efficient

    • Fixed a potential ID related crash in Shader Mixer bricks

    DAZ Studio : Incremented build number to 4.12.2.61

     

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

    I remember seeing that now, and not thinking much of it at the time, that was well before 4.14 was released to us. I thought maybe that would impact VRAM and not much else, again, that was my thought back then. Looking at the results, it is indeed possible that may be what is going on, because I have done some testing with normal maps.. I recently did some testing of HD morphs versus normal maps and found that the HD shape without normals even at a high subdivision was faster than rendering with a low subdivision and normal maps. I had to crank up the subd to like 5 to get the render times to match. The whole point of using normal maps is that they are supposed make it easier to render objects that look more geometrically complex. What is the point of using a normal map if the HD morph renders faster? At subd 3 it even used less VRAM. I had to crank it up to subd 4 to start seeing more VRAM than what I got at subD 2 with normal maps. So if Daz has managed to make normal maps more performant, that could be a big speed gain.

    I would really like to see exactly what the patch note means. How is it more efficient than before? The statement is kind of vague.

    So perhaps I can revisit that test and see how the results go this time. If the speed of 4.14 is indeed tied to efficiency of these maps, then if we disable normal maps and/or bumps then the render speed should be the same between both versions of Daz Studio. So that should be an easy thing to test.

  • AalaAala Posts: 140

    I remember when we did the precursor to this thread with, when I did the first Cornell box and noticed that render times slowed down significantly when I added those clothes with normal maps on. Even a simple ball with a leather texture that covered a really small part of the overall image had a huge impact on the scene. Maybe they finally fixed that?

  • nonesuch00nonesuch00 Posts: 17,890
    RayDAnt said:

    Holy crap, it went from 37 to over 48 iterations per second, that is a monstrous increase. The amount of performance you gained is more than a lot of people's total performance. Just think about that. to put that in perspective, a 3080 ran the bench at 12 iterations. This performance gain is like adding a 5th GPU...a 3080 class GPU...to your rig. This is seriously impressive, and the 4 GPU setup was already impressive!

    Since you have access, would you mind running a single 2080 Super so we can compare it to its previous time? This would also be helpful for looking at scaling. Since we have a single 3090 on 4.14, we would be able to calculate a theoretical peak, and compare that number to what you actually scored to see how many iterations may be lost. For all we know, the scaling might just have improved as well.

    We are going to have to draw a hard line on the benchmark chart now. I think we should carefully note all benchmarks that come after 4.14 because the gap is so big that it may confuse people who compare a 4.14 bench to a 4.12 bench, they are simply not comparable anymore. Obviously benches from one version to the next are generally not comparable, but in practice they really have been consistent over time, with only minor differences aside from when Iray switched to OptiX and RTX support.

    I am surprised that Daz has not advertised this speed upgrade. Advertising a speed increase in the update would get a lot of people on board the choo choo train.

    Just checked, and am also seeing a significant increase in rendering performance on my Titan RTX with the latest Daz Studio and the same drivers (DS 4.14.0.008 beta/non-beta and Nvidia driver version 456.38) over what was seen with the previous beta release (4.12.2.060) also under the same version of Windows. What's interesting here is that both 4.12.0.060 and 4.14.0.008 include the exact same version of Iray (version "Iray RTX 2020.1.1, build 334300.4226, 29 Sep 2020, nt-x86-64" as per the DS changelog - for a full breakdown of which DS versions contain what Iray version, see here.) Meaning that this speedup has to be rooted in updates made to Daz Studio itself (a first at the very least ni the lifetime of this benchmarking thread.) My best guess as to what is actually at the root of making the difference is this from the DS changelog:

    Private Beta (4.14.0.8)

    • Source maintenance

    • Made use of bump and normal maps together with the NVIDIA Iray renderer more efficient

    • Fixed a potential ID related crash in Shader Mixer bricks

    DAZ Studio : Incremented build number to 4.12.2.61

     

    Well seeing this highlighted caused me to go back & test render prior scenes to see if I could visually detect the difference, and the answer is yes. I think a lot of people will now be adjusting the strength of their bump & normal maps to be less; while others might find that the new look is what they were expecting but didn't get in the 1st place. 

  • outrider42outrider42 Posts: 3,679
    edited November 2020

    Alright, I did a new batch of testing, and I think I can comfortably confirm that the normal/bump maps are at the heart of this new new speed. But there may be a catch that comes with this speed.

    I have the general 4.12.0.86 as well as the new 4.14.0.08 beta, so I compared my same test scene from a thread I posted back in July. (Actually, given the timing of this, I cannot help but wonder if my thread got somebody's attention to investigate normal map data usage, but I digress.)

    I ran the scene 3 different ways in both versions using Zelara rendered at a 5000 pixel image resolution. 

    1- Default Zelara, which is subD 2 with normals, no HD morphs applied

    2- Increased subD to 3 and added her HD morph, kept normals on.

    3- SubD 3 with HD morphs again, but without normals.

    In test 1, 4.12 took 6 minutes and 28 seconds to render. But in 4.14 it took only 3 minutes and 30 seconds!

    In test 2, 4.12 took 6 minutes 50 seconds. 4.14 took 3 minutes and 50 seconds. In both tests, I saved 3 minutes off the render, the differences are astonishing.

    In test 3, it gets interesting. Without normals at subD3, 4.12 took 4 minutes 10 seconds. 4.14 took 3 minutes and 45 seconds.

    So you can see that without normal maps, the two versions are much closer in performance. But there's more, this test ran to default convergence. I noticed that 4.12 was also taking more iterations to reach its default value of 95% convergence. While 4.14 was hitting just over 900-950 iterations, 4.12 was over 1000 iterations. It did roughly 50 to 75 more iterations. So when you factor this into test 3, the render times are even closer than they first appear. If I had ran the test with a cap on iterations instead of convergence, they probably would have been a draw. At this point, it becomes quite clear that the way Daz handles normal maps is the prime reason for this new speed in 4.14. In fact, looking at my test, the speed difference is even greater than that of the benchmark scene, which makes sense. All of the spheres lack normal maps. The character in the scene is the only object with normals, her clothes might have some, but the character is not the focus of the scene. My scene has no walls or objects that lack normals. Thus the more your scenes depend on normal maps, the faster 4.14 will be for you!

    Also, the difference between turning normal maps off was extremely minor on performance. In 4.14, I only saved a pitiful 5 seconds turning off the normals at HD3! In 4.12, this was not true at all, turning off the normals offered a massive leap in performance, which is exactly what I reported back in July. In fact the numbers I ran today in 4.12 are very very close to the ones I ran back in July. 

    HOWEVER!!! There is a catch!

    It is not all sunshine and rainbows. As nonesuch00 pointed out, I can see some minor differences in the renders between 4.12 and 4.14. I stress they are very minor! But they are there. Basically, what I see is that the normal maps in 4.12 appear to be slightly stronger than in 4.14. This is really not a big deal to me. I had to directly compare the images up close, zoomed in a fair amount to really see these differences, and this was with a 5000 pixel render. Granted this image is a full body, not a close up. But I think all anybody needs to do if they want that is to simply turn up the normal value a bit. Maybe 25-50% higher...that's a guess. In my test without normal maps, the images were identical. So the normals are indeed the difference.

    BUT that is not the bad part. It looks like 4.14 is using more VRAM than 4.12. A lot more! My Zelara render used a solid 500MB more in 4.14 than in 4.12 in all three tests. For some people that may not be much, but this a single character taking up an additional half GIG of precious VRAM than it used to. Many of us already battle VRAM in a scene creation. Even the no normal test used 500MB more, so this result is quite shocking to me. I didn't have any other apps open during testing, besides notepad, so no other apps should have been running to skew results. Additionally, I have 2 GPUs, and one is not connected to the display. This means that my second GPU is not effected by any apps, as it only gets used with Daz. So the numbers it reports are pretty accurate. Both of my GPUs used 500MB more in 4.14, so it is clear that this is from 4.14, and not some rouge app taking up resources.

    So ultimately this is sort of a mixed bag. There is no doubt that the speed boost is fantastic, as I am sure most of us probably use normal maps. In fact, 4.14 seems to largely negate any performance penalty for using normal maps, as I only gained 5 seconds when I turned them off. This speed boost could translate to many, many hours saved in the long run. But on the flip side the VRAM could be something that catches up to some of us. I am sure that rendering at lower resolutions would not impact it as much, I don't know how many of you render at 5000 pixels, but it is something to be aware of. If you happen to try rendering a larger scene that you previously rendered in 4.12 that was close to your VRAM capacity, there is a possibility that it might not fit your VRAM in 4.14.

    But IMO, I believe the speed is worth that trade. If it was a higher amount of VRAM, maybe I would be more hesitant, but I can work around VRAM limits by optimizing if needed. The performance gains I have seen are just too good to pass on. The benchmark scene does not even do it justice! This is where Iray needs to be to compete with growing competition. Though it is quite interesting to see Daz essentially hedge their bets with the addition of Filament, as finding an alternate render engine for animation is also a topic I and others have discussed at length.

    Post edited by outrider42 on
  • 3DMinh3DMinh Posts: 218

    2 RTX 3090

    System Configuration
    System/Motherboard: Gigabyte Aorus Master X570
    CPU: Ryzen 9 5950x @ stock
    GPU: GPU1: MSI RTX 3090 Ventus 3x OC GPU2: Asus RTX 3090 TUF Gaming OC @ Stock
    System Memory: Gkill Trident Z Neo 64gb @ 3600 MHz Timing 18 - 22 - 22 - 42
    OS Drive: Samsung 960 Pro 1TB
    Asset Drive: Same
    Operating System: Windows 10 pro 20H2
    Nvidia Drivers Version: 456.71 (Studio Driver)
    Daz Studio Version: 4.14.0.8
    Optix Prime Acceleration: STATE (Daz Studio 4.12.1.086 or earlier only)

    Benchmark Results
    DAZ_STATS

    2020-11-17 15:59:29.687 Finished Rendering
    2020-11-17 15:59:29.712 Total Rendering Time: 53.28 seconds

    IRAY_STATS

    2020-11-17 15:59:36.106 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:
    2020-11-17 15:59:36.106 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (GeForce RTX 3090):      917 iterations, 1.678s init, 49.484s render
    2020-11-17 15:59:36.106 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090):      883 iterations, 1.522s init, 48.730s render


    Iteration Rate: 1800 / 49.484s = 36.375 iterations per second
    Loading Time:  53.28s - 49.484s = 3.8 seconds

  • AalaAala Posts: 140

     

    3DMinh said:

    2 RTX 3090

    System Configuration
    System/Motherboard: Gigabyte Aorus Master X570
    CPU: Ryzen 9 5950x @ stock
    GPU: GPU1: MSI RTX 3090 Ventus 3x OC GPU2: Asus RTX 3090 TUF Gaming OC @ Stock
    System Memory: Gkill Trident Z Neo 64gb @ 3600 MHz Timing 18 - 22 - 22 - 42
    OS Drive: Samsung 960 Pro 1TB
    Asset Drive: Same
    Operating System: Windows 10 pro 20H2
    Nvidia Drivers Version: 456.71 (Studio Driver)
    Daz Studio Version: 4.14.0.8
    Optix Prime Acceleration: STATE (Daz Studio 4.12.1.086 or earlier only)

    Benchmark Results
    DAZ_STATS

    2020-11-17 15:59:29.687 Finished Rendering
    2020-11-17 15:59:29.712 Total Rendering Time: 53.28 seconds

    IRAY_STATS

    2020-11-17 15:59:36.106 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:
    2020-11-17 15:59:36.106 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (GeForce RTX 3090):      917 iterations, 1.678s init, 49.484s render
    2020-11-17 15:59:36.106 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090):      883 iterations, 1.522s init, 48.730s render


    Iteration Rate: 1800 / 49.484s = 36.375 iterations per second
    Loading Time:  53.28s - 49.484s = 3.8 seconds

     

    Yeah I just installed my 3090, and I'm also getting a render time in 98 seconds on 4.14 compared to 245 seconds on 2x2080 Ti's in 4.12. Quite a significant improvement.

  • 3DMinh3DMinh Posts: 218
    Aala said:

     

    3DMinh said:

    2 RTX 3090

    System Configuration
    System/Motherboard: Gigabyte Aorus Master X570
    CPU: Ryzen 9 5950x @ stock
    GPU: GPU1: MSI RTX 3090 Ventus 3x OC GPU2: Asus RTX 3090 TUF Gaming OC @ Stock
    System Memory: Gkill Trident Z Neo 64gb @ 3600 MHz Timing 18 - 22 - 22 - 42
    OS Drive: Samsung 960 Pro 1TB
    Asset Drive: Same
    Operating System: Windows 10 pro 20H2
    Nvidia Drivers Version: 456.71 (Studio Driver)
    Daz Studio Version: 4.14.0.8
    Optix Prime Acceleration: STATE (Daz Studio 4.12.1.086 or earlier only)

    Benchmark Results
    DAZ_STATS

    2020-11-17 15:59:29.687 Finished Rendering
    2020-11-17 15:59:29.712 Total Rendering Time: 53.28 seconds

    IRAY_STATS

    2020-11-17 15:59:36.106 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:
    2020-11-17 15:59:36.106 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (GeForce RTX 3090):      917 iterations, 1.678s init, 49.484s render
    2020-11-17 15:59:36.106 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090):      883 iterations, 1.522s init, 48.730s render


    Iteration Rate: 1800 / 49.484s = 36.375 iterations per second
    Loading Time:  53.28s - 49.484s = 3.8 seconds

     

    Yeah I just installed my 3090, and I'm also getting a render time in 98 seconds on 4.14 compared to 245 seconds on 2x2080 Ti's in 4.12. Quite a significant improvement.

    Yeah this generation of RTX cards are amazing. Expensive, but well worth the premium price.

  • chrislbchrislb Posts: 95
    edited November 2020

    I was curious to see how far I could push a 3090 with a stable GPU and Memory overclock and raised power limits.  These settings were also stable for a 40+ minute render.

    The memory overclock is actually only about 500 MHz above DDR6X specifications since the DDR6X underclocked from design specs on most of the 3080 and 3090 cards.  Also, even with the overclocking, the card came nowhere near the 500 watt power limit.

     

    System Configuration:

    System/Motherboard: MSI MEG X570 ACE

    CPU: AMD R9 3950X @ Stock with PBO +200

    GPU: EVGA GeForce RTX 3090 FTW3 ULTRA (24G-P5-3987-KR) @ +169MHz core overclock speed, 1006MHz(x2 = 2012MHz effective) memory overclock speed, and 500 watt XOC BIOS max power limits

    System Memory: Corsair Vengeance RGB Pro 64 GB @ 3600 MHz CAS18

    OS Drive: 1TB Sabrent Rocket NVMe 4.0 SB-ROCKET-NVMe4-1TB

    Asset Drive: XPG SX 8100 NVMe SSD

    Operating System: Windows 10 Pro version 2004 Build 19041.450

    Nvidia Drivers Version: Version 457.30

    Daz Studio Version:  4.14.0.8

     

    Benchmark Results - One Overclocked EVGA RTX 3090 FTW3 Ultra card only no CPU rendering

     

    2020-11-18 01:57:14.499 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend progr: Maximum number of samples reached.

    2020-11-18 01:57:15.046 Finished Rendering

    2020-11-18 01:57:15.093 Total Rendering Time: 1 minutes 31.47 seconds

    2020-11-18 01:57:18.515 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2020-11-18 01:57:18.515 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 1.307s init, 87.493s render 

     

    Iteration Rate: (1800/87.493) = 20.5730 iterations per second

    Loading Time: ((91.47 seconds) - 87.493) = 3.977 seconds

    I think the only way I can get better single card performance is to use water cooling so that it will sustain higher clock speeds for a longer time.

     

    EDIT: I had the wrong Daz version number originally.

    Post edited by chrislb on
  • Since you have access, would you mind running a single 2080 Super so we can compare it to its previous time? This would also be helpful for looking at scaling. Since we have a single 3090 on 4.14, we would be able to calculate a theoretical peak, and compare that number to what you actually scored to see how many iterations may be lost. For all we know, the scaling might just have improved as well.

    I'll run more tests later this week with the new version of Daz on one and two 2080 Supers.

  • AalaAala Posts: 140

    @chrislb

    Yeah these Ampere cards are already at their limits, unlike previous generations. OC'ing them doesn't seem to give meaningful results.

  • outrider42outrider42 Posts: 3,679
    With Iray there is probably little a memory overclock will ever do, since Iray does not swap data in and out of VRAM like a game does. It basically loads scene data once, and keeps it in memory for the duration of that render. It might have a very small effect with multiGPU or CPU+GPU systems that might trade more data, but I can't imagine it doing anything for a single GPU system. This is also why Iray does not make a GPU as hot as many video games do. It is because how video games can peg the VRAM, this creates heat over those components. Since Iray does not make as much use of VRAM, the components around the VRAM do not get as warm. To compare, the PS5 only keeps about 1 second worth of data in VRAM. It is constantly working data in and out during a game session. Many systems only keep about 20 to 30 seconds of data in VRAM at most.

    It has been found that one of the reasons the 3080 and 3090 get as hot as they do is because of their GDDR6X. If you look, neither card even uses the full speed that GDDR6X is capable of, which is why the VRAM is so easy to overclock for most users. Many people believe that Ampere is not using the full speed because of the high temperatures it creates. I would be very careful with overclocking the VRAM in these cards if you plan on playing any games, or use a gaming engine to render. Not many GPUs have sensors on the VRAM components to report how hot they are. So depending on the model, you may not know how hot those components are running.
  • Aala said:

    @chrislb

    Yeah these Ampere cards are already at their limits, unlike previous generations. OC'ing them doesn't seem to give meaningful results.

    19.365 vs 20.573 iterations per second.  That's a ~6% increase in iterations per second.

     

    It has been found that one of the reasons the 3080 and 3090 get as hot as they do is because of their GDDR6X. If you look, neither card even uses the full speed that GDDR6X is capable of, which is why the VRAM is so easy to overclock for most users. Many people believe that Ampere is not using the full speed because of the high temperatures it creates. I would be very careful with overclocking the VRAM in these cards if you plan on playing any games, or use a gaming engine to render. Not many GPUs have sensors on the VRAM components to report how hot they are. So depending on the model, you may not know how hot those components are running.

    Luckily, the FTW3 cards have multiple VRAM temperature sensors.  The front VRM is cooled by the same heatsink as the CPU and the back VRAM is cooled with the backplate.  There are two heatpipes in the backplate to trasfer heat away from the VRAM and down the backplate.

    I'm still curious about the Iray rendering performance RTX 3080 cards and the upcoming 3080 ti.  THe 3080 ti looks to be a 20 GB version of the 3090 for $500 less.

  • billyben_0077a25354billyben_0077a25354 Posts: 771
    edited November 2020

    Decided to throw my hat in the ring  with my major system update (new MB, CPU, & memory).  Still waiting on the right GPU (hopefully a 3070 16GB) to be released.  This is with my GTX 1070 with the updated system build and the latest released Geforce Studio driver.

    System Configuration
    System/Motherboard: Gigabyte X570 Aorus PRO WIFI
    CPU: AMD Ryzen 7 3700X @ 3.6 (boosts to 4.3)
    GPU: EVGA GTX 1070 GAMING @ 1506 MHz
    System Memory: G.Skill 32 GB Trident Z Neo DDR-4 3200 memory @ SPEED
    OS Drive: Samsung 860 EVO 1TB SSD
    Asset Drive: 2 x Western Digital Blue WD20EZRZ 2TB HDD in mirrored array
    Operating System: Windows 10 Professional 64 1909
    Nvidia Drivers Version: 456.38 Studio Driver
    Daz Studio Version: 4.12.1.118
    Optix Prime Acceleration: N/A

    Benchmark Results
    2020-10-14 09:34:06.371 Finished Rendering
    2020-10-14 09:34:06.403 Total Rendering Time: 14 minutes 37.83 seconds
    2020-10-14 09:37:43.976 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:
    2020-10-14 09:37:43.976 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce GTX 1070):                 1800 iterations, 2.427s init, 872.484s render
    Iteration Rate: 2.0630750822 iterations per second
    Loading Time: ((0 + 840 + 37.83) - 872.484)=4.982 seconds

     

    Just updated to 4.14.0.8 earlier today and boy is it faster.  Went from a total render time of 14 minutes 37.93 seconds to 10 minutes 19.36 seconds.

    System Configuration
    System/Motherboard: Gigabyte X570 Aorus PRO WIFI
    CPU: AMD Ryzen 7 3700X @ 3.6 (boosts to 4.3)
    GPU: EVGA GTX 1070 GAMING @ 1506 MHz
    System Memory: G.Skill 32 GB Trident Z Neo DDR-4 3200 memory @ SPEED
    OS Drive: Samsung 860 EVO 1TB SSD
    Asset Drive: 2 x Western Digital Blue WD20EZRZ 2TB HDD in mirrored array
    Operating System: Windows 10 Professional 64 1909
    Nvidia Drivers Version: 456.38 Studio Driver
    Daz Studio Version: 4.14.0.8
    Optix Prime Acceleration: N/A

    Benchmark Results
    2020-11-18 13:31:00.863 Finished Rendering
    2020-11-18 13:31:00.889 Total Rendering Time: 10 minutes 19.36 seconds

    No itenerations line given at the end of the render but found this new informaiton at the start of the render.

    2020-11-18 14:23:24.063 Iray [INFO] - IRAY:RENDER ::   1.2   IRAY   rend info : CUDA device 0 (GeForce GTX 1070): Scene processed in 1.536s
    2020-11-18 14:23:24.065 Iray [INFO] - IRAY:RENDER ::   1.2   IRAY   rend info : CUDA device 0 (GeForce GTX 1070): Allocated 15.450 MiB for frame buffer
    2020-11-18 14:23:24.130 Iray [INFO] - IRAY:RENDER ::   1.2   IRAY   rend info : CUDA device 0 (GeForce GTX 1070): Allocated 1.688 GiB of work space (2048k active samples in 0.061s)
    2020-11-18 14:23:24.130 Iray [INFO] - IRAY:RENDER ::   1.2   IRAY   rend info : CUDA device 0 (GeForce GTX 1070): Used for display, optimizing for interactive usage (performance could be sacrificed)

    I did come up with an iteneraation rate of 2.9213001733 itenerations per second so that is about a 45% increase in itenerations just updating from 4.12 to 4.14.  I am a happy camper and will be much more happy once I can get a decent RTX 3000 series card.

    Post edited by billyben_0077a25354 on
  • chrislbchrislb Posts: 95
    edited November 2020

    System Configuration:

     

    System/Motherboard: MSI MEG X570 ACE

    CPU: AMD R9 3950X @ Stock with PBO +200

    GPU: EVGA GeForce Two EVGA RTX 2080 Super cards

    System Memory: Corsair Vengeance RGB Pro 64 GB @ 3600 MHz CAS18

    OS Drive: 1TB Sabrent Rocket NVMe 4.0 SB-ROCKET-NVMe4-1TB

    Asset Drive: XPG SX 8100 NVMe SSD

    Operating System: Windows 10 Pro version 2004 Build 19041.450

    Nvidia Drivers Version: Version 457.30

    Daz Studio Version: 4.14.0.8

     

    Benchmark Results - Two EVGA RTX RTX 2080 Super cards only no CPU rendering

     

    2020-11-19 18:01:07.983 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend progr: Maximum number of samples reached.

    2020-11-19 18:01:08.610 Total Rendering Time: 2 minutes 16.46 seconds

    2020-11-19 18:01:11.529 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2020-11-19 18:01:11.529 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 2080 SUPER): 821 iterations, 1.309s init, 132.084s render

    2020-11-19 18:01:11.529 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (GeForce RTX 2080 SUPER): 979 iterations, 1.299s init, 131.904s render

     

    Iteration Rate: (1800/132.084) = 13.6276 iterations per second

    Loading Time: ((136.46 seconds) - 132.084) = 4.376 seconds

     

    Benchmark Results - One EVGA RTX RTX 2080 Super card only no CPU rendering

     

    2020-11-19 18:13:07.264 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend progr: Maximum number of samples reached.

    2020-11-19 18:13:07.888 Total Rendering Time: 4 minutes 44.12 seconds

    2020-11-19 18:13:17.251 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2020-11-19 18:13:17.251 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 2080 SUPER): 1800 iterations, 1.172s init, 280.223s render

     

    Iteration Rate: (1800/280.223) = 6.4234 iterations per second

    Loading Time: ((284.12 seconds) - 280.223) = 3.897 seconds

     

    For some reason two cards had a combined iteration that was more than double the single card result.

     

    EDIT: Rebooting the PC and rerunning the single card benchmark again gave the same result.  The difference was less than 0.2 secodns in render time between the two benchmarks.

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

     

    chrislb said:

    System Configuration:

     

    System/Motherboard: MSI MEG X570 ACE

    CPU: AMD R9 3950X @ Stock with PBO +200

    GPU: EVGA GeForce Two EVGA RTX 2080 Super cards

    System Memory: Corsair Vengeance RGB Pro 64 GB @ 3600 MHz CAS18

    OS Drive: 1TB Sabrent Rocket NVMe 4.0 SB-ROCKET-NVMe4-1TB

    Asset Drive: XPG SX 8100 NVMe SSD

    Operating System: Windows 10 Pro version 2004 Build 19041.450

    Nvidia Drivers Version: Version 457.30

    Daz Studio Version: 4.14.0.8

     

    Benchmark Results - Two EVGA RTX RTX 2080 Super cards only no CPU rendering

     

    2020-11-19 18:01:07.983 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend progr: Maximum number of samples reached.

    2020-11-19 18:01:08.610 Total Rendering Time: 2 minutes 16.46 seconds

    2020-11-19 18:01:11.529 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2020-11-19 18:01:11.529 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 2080 SUPER): 821 iterations, 1.309s init, 132.084s render

    2020-11-19 18:01:11.529 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (GeForce RTX 2080 SUPER): 979 iterations, 1.299s init, 131.904s render

     

    Iteration Rate: (1800/132.084) = 13.6276 iterations per second

    Loading Time: ((136.46 seconds) - 132.084) = 4.376 seconds

     

    Benchmark Results - One EVGA RTX RTX 2080 Super card only no CPU rendering

     

    2020-11-19 18:13:07.264 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend progr: Maximum number of samples reached.

    2020-11-19 18:13:07.888 Total Rendering Time: 4 minutes 44.12 seconds

    2020-11-19 18:13:17.251 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2020-11-19 18:13:17.251 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 2080 SUPER): 1800 iterations, 1.172s init, 280.223s render

     

    Iteration Rate: (1800/280.223) = 6.4234 iterations per second

    Loading Time: ((284.12 seconds) - 280.223) = 3.897 seconds

     

    For some reason two cards had a combined iteration that was more than double the single card result.

     

    EDIT: Rebooting the PC and rerunning the single card benchmark again gave the same result.  The difference was less than 0.2 secodns in render time between the two benchmarks.

    I see the issue. Your second 2080 Super ran 979 iterations, while the 1st one only ran 821. That is a gap of 158 iterations! That is a surprisingly large gap in performance between two cards that should be much closer than that. To compare, look my 1080tis. One ran 906 and the other 894. 

    So if you are running your solo 2080 Super bench with that slower card, there is your answer. If possible, try the other 2080 Super instead, I bet the difference will make much more sense.

    The real question here is what is slowing this 2080 Super down so much compared to the other one? Certainly running Windows is going to tax it slightly, but 158 iterations just seems too much. This could be skewing your benchmark results, too. Unfortunately there is only one other 2080 Super on the bench list, and it is rather old now.

  •  

    I see the issue. Your second 2080 Super ran 979 iterations, while the 1st one only ran 821. That is a gap of 158 iterations! That is a surprisingly large gap in performance between two cards that should be much closer than that. To compare, look my 1080tis. One ran 906 and the other 894. 

    So if you are running your solo 2080 Super bench with that slower card, there is your answer. If possible, try the other 2080 Super instead, I bet the difference will make much more sense.

    The real question here is what is slowing this 2080 Super down so much compared to the other one? Certainly running Windows is going to tax it slightly, but 158 iterations just seems too much. This could be skewing your benchmark results, too. Unfortunately there is only one other 2080 Super on the bench list, and it is rather old now.

     

    For some reason 2080 Super one was always using more VRAM than the other when both were rendring.  I'm not sure why.

     

  • outrider42outrider42 Posts: 3,679
    chrislb said:

     

    I see the issue. Your second 2080 Super ran 979 iterations, while the 1st one only ran 821. That is a gap of 158 iterations! That is a surprisingly large gap in performance between two cards that should be much closer than that. To compare, look my 1080tis. One ran 906 and the other 894. 

    So if you are running your solo 2080 Super bench with that slower card, there is your answer. If possible, try the other 2080 Super instead, I bet the difference will make much more sense.

    The real question here is what is slowing this 2080 Super down so much compared to the other one? Certainly running Windows is going to tax it slightly, but 158 iterations just seems too much. This could be skewing your benchmark results, too. Unfortunately there is only one other 2080 Super on the bench list, and it is rather old now.

     

    For some reason 2080 Super one was always using more VRAM than the other when both were rendring.  I'm not sure why.

     

    Is it the display GPU? Or it it could be the GPU driving the viewport in Daz. Even so, I wouldn't expect this. Just looking at my own, one of my 1080tis is driving the display, and it does indeed use a bit more VRAM no matter what thanks to Windows. I will add that my 1080tis are not the same card, not even the same brand. So my situation is a bit different from many people who buy the same cards. It is also a bit different in that I have GPU #2 as my display driver. Both my slots are full pcie x16, so it doesn't matter (at least for gaming) which I use as the main card. It is kind of interesting. GPU #1 is actually clocked a bit higher, however GPU2 is in a much better cooling situation being on the bottom and having a beefy cooler. I can't swap them because the size of the 2nd card is too fat, it would choke the other card if I swapped them. Anyway, the result is that GPU2 ends up running much cooler. I just ran a render, and it stayed at 67C, meanwhile GPU1 ran at 82C. Because of how Pascal steps down at certain temps, the end result is that GPU1 was running at 1911 mhz, while GPU2 was at 1932 mhz.

    So with going by the frequency, you would expect GPU2 to do more iterations, right? Wrong! GPU2 gets beat by GPU1 every single time, and the length of time the render runs doesn't change this either. I've checked my iteration counts on hour long renders, even then GPU2 with its faster sustained clockspeed still cannot beat the 1st GPU in iteration count. Remember, GPU2 is my display, so clearly this impacts its performance even when it is running cooler and faster than GPU1. But in the end my numbers still seem to line up with others in spite of this.

    I took a peak at couple other multiGPU scores. One with a pair of 2080tis saw one do 904 and the other 896. Another user ran two 1080s with doing 905 and the other 895. Then another user with a different pair of 1080tis ran 904 to 896.

    So I think we can see the pattern, most of the GPU pairs put up are very close to each other, often within 15 iterations, or even 10. But yours are over 150 iterations apart, so I wonder if something weird is going on. Maybe Windows is doing something in background, or maybe it could be a power limit being hit somehow. Is this still the monster rig test or have the 3090s been removed? Maybe there is an issue with power delivery.

  • chrislbchrislb Posts: 95
    edited November 2020

    The previous single card 2080 Super test was GPU0, which was also driving the display.  I removed the 3090 and only had the 2080 Super cards in the system for the test.  I'm not sure if something weird was going on with the drivers, so I uninstalled the video card drivers with DDU in safe mode, rebooted, and reinstalled the drivers then rebooted again and reran the tests.

     

    System Configuration:

    System/Motherboard: MSI MEG X570 ACE

    CPU: AMD R9 3950X @ Stock with PBO +200

    GPU: EVGA GeForce Two EVGA RTX 2080 Super cards

    System Memory: Corsair Vengeance RGB Pro 64 GB @ 3600 MHz CAS18

    OS Drive: 1TB Sabrent Rocket NVMe 4.0 SB-ROCKET-NVMe4-1TB

    Asset Drive: XPG SX 8100 NVMe SSD

    Operating System: Windows 10 Pro version 2004 Build 19041.450

    Nvidia Drivers Version: Version 457.30

    Daz Studio Version: 4.14.0.8

     

    Benchmark Results - Two EVGA RTX RTX 2080 Super cards only no CPU rendering

     

    2020-11-22 15:03:35.242 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend progr: Received update to 01800 iterations after 131.121s.

    2020-11-22 15:03:35.250 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend progr: Maximum number of samples reached.

    2020-11-22 15:03:35.820 Finished Rendering

    2020-11-22 15:03:35.865 Total Rendering Time: 2 minutes 14.9 seconds

    2020-11-22 15:03:39.100 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2020-11-22 15:03:39.100 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 2080 SUPER): 892 iterations, 1.286s init, 129.267s render

    2020-11-22 15:03:39.100 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (GeForce RTX 2080 SUPER): 908 iterations, 1.317s init, 129.903s render

     

    Iteration Rate: (1800/129.903) = 13.856 iterations per second

    Loading Time: ((134.9 seconds) - 129.903) = 4.997 seconds

     

    Benchmark Results - One EVGA RTX RTX 2080 Super card only no CPU rendering GPU0

     

    2020-11-22 15:32:10.503 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend progr: Received update to 01800 iterations after 252.480s.

    2020-11-22 15:32:10.511 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend progr: Maximum number of samples reached.

    2020-11-22 15:32:11.075 Finished Rendering

    2020-11-22 15:32:11.123 Total Rendering Time: 4 minutes 15.36 seconds

    2020-11-22 15:32:12.901 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2020-11-22 15:32:12.901 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 2080 SUPER): 1800 iterations, 1.148s init, 251.522s render

     

    Iteration Rate: (1800/251.522) = 7.1564 iterations per second

    Loading Time: ((255.36 seconds) - 251.522) = 3.838 seconds

     

    Benchmark Results - One EVGA RTX RTX 2080 Super card only no CPU rendering GPU1

     

    2020-11-22 15:37:53.420 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend progr: Received update to 01800 iterations after 249.759s.

    2020-11-22 15:37:53.421 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend progr: Maximum number of samples reached.

    2020-11-22 15:37:54.039 Total Rendering Time: 4 minutes 12.66 seconds

    2020-11-22 15:38:11.447 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2020-11-22 15:38:11.447 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (GeForce RTX 2080 SUPER): 1800 iterations, 1.150s init, 248.813s render

     

    Iteration Rate: (1800/248.813) = 7.2343 iterations per second

    Loading Time: ((252.66 seconds) - 248.813) = 3.847 seconds

     

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

    That seems to fix it. Now the difference is 892-908, much more like the others who have 2 cards. The dual card bench also lines up now, as the two cards combined are slightly less than their solo iteration counts combined, no funny magic. Interesting how a driver may have caused this. But it has happened before. I recall back when some gamers were claiming that older cards like Pascal had been 'nerfed' by Nvidia after Turing arrived. It turns out that such claims were simply the result of a driver just not taking well, and were not reproduced by other users. Sometimes you just have to try installing again, even if you had reinstalled already. That is probably what happened here, though I will say that Iray performance has been acting funny for a while up until the latest drivers. For several months and numerous driver updates some of us lost some render speed and we never figured out why. It did not effect my gaming at all, just Iray. Whatever, I am glad the new driver fixed it. I just wish I knew what actually happened. Like I said before, I still have 4.12 and so I was able to confirm this. 4.12 now runs like it used to before that particular driver, and then 4.14 improved rendering itself, so for me, I experienced a huge bump in performance.

  • volpler11volpler11 Posts: 29
    edited November 2020

    Got my EVGA 3090 FTW3 Ultra today. Decided to do some benchmarking make sure the cards works.

    System Configuration

    System/Motherboard: Asus X570-F

    CPU: AMD 3900X @ PBO

    GPU: EVGA 3090 FTW3 ULTRA @ stock normal bios 

    System Memory: 2x32GB @ 3600 CL18

    OS Drive: Samsung 960 EVO 1TB

    Asset Drive: western digital SN750 1TB

    Operating System: Windows 10 2004

    Nvidia Drivers Version: 457.30

    Daz Studio Version: 4.14.0.8 Beta

     

    Benchmark Results: Single 3090 with no CPU. Fresh reboot, no other program running, 1440p ultrawide monitor.

    2020-11-23 14:12:57.332 Total Rendering Time: 1 minutes 39.41 seconds

    2020-11-23 14:13:29.853 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 1.913s init, 95.029s render

     

    2020-11-23 14:15:09.346 Total Rendering Time: 1 minutes 38.33 seconds

    2020-11-23 14:15:17.442 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 1.436s init, 95.018s render

     

    2020-11-23 14:16:57.120 Total Rendering Time: 1 minutes 38.41 seconds

    2020-11-23 14:17:04.765 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 1.438s init, 95.109s render

     

    2020-11-23 14:18:44.229 Total Rendering Time: 1 minutes 38.41 seconds

    2020-11-23 14:19:14.095 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 1.423s init, 95.114s render

     

    2020-11-23 14:20:53.867 Total Rendering Time: 1 minutes 38.42 seconds

    2020-11-23 14:21:03.023 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 1.434s init, 95.117s render

     

    2020-11-23 14:22:42.775 Total Rendering Time: 1 minutes 38.51 seconds

    2020-11-23 14:22:45.041 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 1.431s init, 95.217s render

     

    2020-11-23 14:24:54.028 Total Rendering Time: 1 minutes 38.43 seconds

    2020-11-23 14:25:17.938 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 1.478s init, 95.118s render

     

    2020-11-23 14:26:58.032 Total Rendering Time: 1 minutes 38.39 seconds

    2020-11-23 14:27:04.391 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 1.445s init, 95.088s render

     

    2020-11-23 14:28:43.953 Total Rendering Time: 1 minutes 38.54 seconds

    2020-11-23 14:30:11.129 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 1.424s init, 95.255s render

     

    2020-11-23 14:32:01.885 Total Rendering Time: 1 minutes 38.35 seconds

    2020-11-23 14:41:21.131 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 1.415s init, 95.122s render

     

    2020-11-23 14:43:00.263 Total Rendering Time: 1 minutes 38.43 seconds

    2020-11-23 14:43:02.348 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 1.435s init, 95.132s render

     

    Iteration Rate: 18.92 iterations per second, SD: 0.0137

    Post edited by volpler11 on
  • volpler11volpler11 Posts: 29
    edited November 2020

    Some additional testing, see my other post for system spec.

     

    With 1440p ultrawide + 1440p normal monitor (light browsing on 2nd monitor running chrome, excel, notepad)

    2020-11-23 15:24:04.732 Total Rendering Time: 1 minutes 42.57 seconds

    2020-11-23 15:25:59.144 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 1.440s init, 99.256s render

     

    2020-11-23 15:27:42.228 Total Rendering Time: 1 minutes 41.63 seconds

    2020-11-23 15:27:52.940 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 1.419s init, 98.297s render

     

    2020-11-23 15:29:54.157 Total Rendering Time: 1 minutes 44.59 seconds

    2020-11-23 15:30:09.038 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 1.451s init, 101.264s render

     

    2020-11-23 15:36:51.232 Total Rendering Time: 1 minutes 44.55 seconds

    2020-11-23 15:36:57.426 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 1.406s init, 101.300s render

     

    Iteration Rate: 18.00 iterations per second, SD: 0.27

    04.87% slower than running single 1440p ultrawide monitor.

     

     

    With 1440p ultrawide + chrome, excel, notepad running in background, 2nd monitor turned off.

     

    2020-11-23 15:48:18.616 Total Rendering Time: 1 minutes 38.70 seconds

    2020-11-23 15:48:45.583 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 1.401s init, 95.469s render

     

    2020-11-23 15:50:26.189 Total Rendering Time: 1 minutes 38.69 seconds

    2020-11-23 15:51:00.210 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 1.441s init, 95.344s render

     

    2020-11-23 15:52:41.956 Total Rendering Time: 1 minutes 38.72 seconds

    2020-11-23 15:53:26.054 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 1.431s init, 95.409s render

     

    Iteration Rate: 18.87 iterations per second, SD: 0.0124

    0.28% slower than running single 1440p ultrawide monitor.

     

    With 1440p ultrawide + light browsing on chrome, excel, notepad running in background, 2nd monitor turned off.

    2020-11-23 15:59:31.278 Total Rendering Time: 1 minutes 42.39 seconds

    2020-11-23 16:01:20.044 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 1.407s init, 99.138s render

     

    2020-11-23 16:03:46.456 Total Rendering Time: 1 minutes 41.22 seconds

    2020-11-23 16:04:22.897 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 1.412s init, 97.971s render

     

    2020-11-23 16:06:08.634 Total Rendering Time: 1 minutes 42.63 seconds

    2020-11-23 16:07:48.882 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 1.582s init, 99.137s render

     

    Iteration Rate: 18.23 iterations per second, SD: 0.0125

    3.65% slower than running single 1440p ultrawide monitor.

    Post edited by volpler11 on
  • How do the CPUs affect the general performance of DAZ's UI if you're only using the GPUs to render?

    I've got a 5900X in the mail and finally got a preorder set up for my new RTX 3090. Do Intel CPUs edge out AMD in single core basic operation, or have people had a good experience with Ryzen being responsive, or even more responsive in this case?

  • ArtAngelArtAngel Posts: 1,484
    edited November 2020
    I don't know how many of you render at 5000 pixels, but it is something to be aware of. If you happen to try rendering a larger scene that you previously rendered in 4.12 that was close to your VRAM capacity, there is a possibility that it might not fit your VRAM in 4.14.

     

    But IMO, I believe the speed is worth that trade. If it was a higher amount of VRAM, maybe I would be more hesitant, but I can work around VRAM limits by optimizing if needed. The performance gains I have seen are just too good to pass on. The benchmark scene does not even do it justice! This is where Iray needs to be to compete with growing competition. Though it is quite interesting to see Daz essentially hedge their bets with the addition of Filament, as finding an alternate render engine for animation is also a topic I and others have discussed at length.

    I render at 10,000, 7680, 5120 and other time near 4k pixels. However I do use 2048 in some cases and none of those other sizes are needed for web  videos or images. Not sure if 4.14 is an option for me but if it was, maybe as a beta, but how many plug-ins does one lose with the beta or the Pro version? What would be the gain other than a live OpenGl preview option that surpasses the texture view yet falls short of Iray realism? My understanding is 4.14 is 4.12 with a filamont option to view and render scenes without Iray, by utilizing the CPU not to mention the texture issues regarding blonde balding hair. Also 4.12 broke a few things like IK chains, Strand Based Hair, and Iray Section Nodes, which are still broken in 4.14, but what else might be broken in 4.14? What is the gain vs the cost for those of us who don't have Kepler or Maxwell cards? Some people using 4.14 say it works well while others say it crashes. After reading 150 pages on various forums it seems that some bugs and crashes are avoided and 4.14 runs mush much smoother with CPU enabled. Wondering what your thoughts are.

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

    The only thing downloading the beta would do is overwrite the current beta you have, if you don't save the previous one somewhere else or install the new one to a different location. It is certainly possible to have multiple installs of Daz, it is easier if you install on different drives, just don't let DIM delete the old install (its a check box in the settings). I have several versions of Daz on my PC across my drives. I keep a 4.10 and 4.11, a couple 4.12's and now 4.14. Some are betas some are not, but point being I've got like 5 or 6 versions of Daz on one PC.

    It is possible to get those plugins working in the betas. Richard wrote something about it, I actually forget where. I just use the full version for those times when I want to use a plugin I can't in the beta. Most of the plugins I use work in the betas anyway.

    Going from 4.10 and 4.11, I think a lot of things have changed. Dforce in particular has seen the most changes. I also think the Daz app is a bit snappier when working on dense scenes, actually, the difference is striking for my PC. I had some scenes with a lot of foliage that were making Daz 4.11 slow down to a crawl. It was borderline unusable. But these same scenes did not drag down 4.12+. Though my CPU is not as good as either of your machines, so maybe you haven't ran into that as much. But the app performance can be pretty dramatic on my end.

    And this new 4.14 renders so much faster. It is a noticeable difference, and I think you will really notice it, too, if you use similar scenes that you made before. It is a difference that is big enough that IMO I am willing to ignore other faults. If I need something that only works in the full version, then I'll save and open the scene in that version. No big deal.

     

    How do the CPUs affect the general performance of DAZ's UI if you're only using the GPUs to render?

    I've got a 5900X in the mail and finally got a preorder set up for my new RTX 3090. Do Intel CPUs edge out AMD in single core basic operation, or have people had a good experience with Ryzen being responsive, or even more responsive in this case?

    Daz is single threaded. So single threaded performance is the prime factor in how well it may run. However Daz is not super heavy on CPU unless you build very dense scenes with lots of geometry. Like say you toss in a bunch of HD level 5 models, that could slow any PC down. Otherwise, pretty much any CPU is going to function very similar.

    Some aspects that might improve are instances of smoothing. If you enable smoothing on numerous items and/or have high smoothing/collision counts, then the CPU load grows a lot. If you move the camera around and notice a little progress bar at the bottom of Daz, this is the CPU computing the smoothing calculations or something in the viewport. It has to do that every time you pose and move the things around. So a faster CPU can improve this.

    Since Daz is single threaded, there can still be instances of slowdown even on a very powerful CPU. You can look at Ryzen reviews, AMD pretty much took the single core performance crown in most situations. I think Intel may have squeaked out a couple wins in some specific Intel friendly apps, but those were very small wins while AMD dominated most everything else. I don't know which CPU Daz would favor, that isn't something I have ever seen tested that way. But I don't think you would ever notice the difference if there is one. So I would not worry in the least about this, a 5900X isn't going to have any issues.

  • ArtAngelArtAngel Posts: 1,484

    The only thing downloading the beta would do is overwrite the current beta you have, if you don't save the previous one somewhere else or install the new one to a different location. It is certainly possible to have multiple installs of Daz, it is easier if you install on different drives, just don't let DIM delete the old install (its a check box in the settings). I have several versions of Daz on my PC across my drives. I keep a 4.10 and 4.11, a couple 4.12's and now 4.14. Some are betas some are not, but point being I've got like 5 or 6 versions of Daz on one PC.

    It is possible to get those plugins working in the betas. Richard wrote something about it, I actually forget where. I just use the full version for those times when I want to use a plugin I can't in the beta. Most of the plugins I use work in the betas anyway.

    Going from 4.10 and 4.11, I think a lot of things have changed. Dforce in particular has seen the most changes. I also think the Daz app is a bit snappier when working on dense scenes, actually, the difference is striking for my PC. I had some scenes with a lot of foliage that were making Daz 4.11 slow down to a crawl. It was borderline unusable. But these same scenes did not drag down 4.12+. Though my CPU is not as good as either of your machines, so maybe you haven't ran into that as much. But the app performance can be pretty dramatic on my end.

    And this new 4.14 renders so much faster. It is a noticeable difference, and I think you will really notice it, too, if you use similar scenes that you made before. It is a difference that is big enough that IMO I am willing to ignore other faults. If I need something that only works in the full version, then I'll save and open the scene in that version. No big deal.

     

    Thanks for your helpful advice. I have been hopping between computers to use certain products which are version reliant. I currently have five pro versions on five devices and your way makes a lot more sense. Guess what I'm going to be doing this Thanksgiving? Trying to hunt through external hard drives for saved zip versions. After TG I'll also hunt down the benchmark .duf and give it a go for 4.14 on 2080x2. I have several old store bought systems, the newest of the older desktops is one (from costco years ago) upgraded with a higher power  supply  and a GTX 210 (it refused a  970 so it's limited to 3.3 OpenGL)  that has 4.9 on it, and at times it does come in handy for compiling scene portions, using old plug-ins that have been dropped in newer versions. So, thanks to you I can unclutter my workspace and stop acting like a rabbit. 

  • outrider42outrider42 Posts: 3,679

    Just make sure to backup your old installs in case you accidentally mess up. I forgot to uncheck the option to remove the install in DIM and lost a version or two from that. It is extremely easy to miss and nothing is more frustrating.

  • System Configuration
    System/Motherboard: TUF GAMING X570-PLUS 
    CPU: AMD Ryzen 9 3900X 12-Core Processor 3.79GHz (stock)
    GPU: NVIDIA GeForce RTX 2070 SUPER 8GB / stock
    GPU: NVIDIA GeForce RTX 2070 SUPER 8GB / stock [new]
    System Memory: Corsair Vengence LPX 32GB kit (4x16GB = 64GB) DDR4 @ 3200 MHz
    OS Drive: Samsung SSD 970 EVO Plus 1TB M.2 NVMe (disk 0)
    DAZ3D Drive: Samsung SSD 970 EVO Plus 1TB M.2 NVMe (disk 1)
    Asset Drive: AMD-RAID0 (Samsung SSD 860 EVO 1TB x 2)
    Operating System: Windows 10 Home 10.0.18363 Build 18636
    Nvidia Drivers Version: 27.21.14.5167 (2020.07.05)
    Daz Studio Version: 4.12.1.118 Pro Edition 64bit
    Optix Prime Acceleration: N/A (Daz Studio 4.12.1.086 or earlier only)


    Benchmark Results (GPU x1)
    DAZ_STATS
    2020-07-16 21:37:49.845 Finished Rendering
    2020-07-16 21:37:49.876 Total Rendering Time: 5 minutes 51.51 seconds
    IRAY_STATS
    2020-07-16 21:38:08.433 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 2070 SUPER):   1800 iterations, 2.358s init, 346.371s render

    Iteration Rate: (1800 / 346.371) =  5.197 iterations per second
    Loading Time: ((0 * 3600 + 5 * 60 + 51.51) - 346.371) =  5.139 seconds


    Benchmark Results (CPU+GPU x1)
    DAZ_STATS
    2020-07-16 18:46:25.543 Finished Rendering
    2020-07-16 18:46:25.567 Total Rendering Time: 4 minutes 58.22 seconds
    IRAY_STATS
    2020-07-16 18:46:52.008 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 2070 SUPER):   1509 iterations, 1.939s init, 293.483s render
    2020-07-16 18:46:52.008 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CPU:   291 iterations, 1.595s init, 293.221s render

    Iteration Rate: (1509 +291/ 293.483) =  6.133 iterations per second
    Loading Time: ((0 * 3600 + 4 * 60 + 58.22) - 293.483) =  5.737 seconds


    [NEW BENCHMARK]

    Nvidia Drivers Version: 27.21.14.5671 (2020.09.30)
    Daz Studio Version: 4.14.0.8 Pro Edition (64 Bit)

    Benchmark Results (CPU+GPU+GPU)
    DAZ_STATS
    2020-11-29 02:42:01.188 Finished Rendering
    2020-11-29 02:42:01.225 Total Rendering Time: 3 minutes 42.6 seconds
    IRAY_STATS
    2020-11-29 02:42:15.989 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (GeForce RTX 2070 SUPER): 799 iterations, 2.800s init, 216.428s render
    2020-11-29 02:42:15.989 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 2070 SUPER): 810 iterations, 2.318s init, 216.444s render
    2020-11-29 02:42:15.989 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CPU: 191 iterations, 1.939s init, 216.804s render

    Iteration Rate: (799+810+191/216.428) =  8.317 iterations per second
    Loading Time: ((0 * 3600 + 3 * 60 + 42.6) - 216.444) =  6.156 seconds


    Benchmark Results (GPU+GPU)
    DAZ_STATS
    2020-11-29 02:31:06.979 Finished Rendering
    2020-11-29 02:31:07.015 Total Rendering Time: 2 minutes 24.20 seconds
    IRAY_STATS
    2020-11-29 02:32:00.968 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (GeForce RTX 2070 SUPER): 902 iterations, 1.860s init, 139.457s render
    2020-11-29 02:32:00.968 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 2070 SUPER): 898 iterations, 1.754s init, 138.989s render

    Iteration Rate: (902+898/ 139.457) =  12.907 iterations per second
    Loading Time: ((0 * 3600 + 2 * 60 + 24.20) - 139.457) =  4.743 seconds

Sign In or Register to comment.