Daz Studio Iray - Rendering Hardware Benchmarking

1171819202123»

Comments

  • System Configuration
    System/Motherboard: Aorus X570 Elite
    CPU: AMD Ryzen 9 3900X 12-Core @ stock
    GPU: MSI Gaming X GeForce GTX 1080 Ti @ stock
    System Memory: G.Skill Trident Z 64 GB @ 3200
    OS Drive: Samsung SSD 970
    Asset Drive: Seagate HDD 4TB
    Operating System: Windows 10 Pro 20H2
    Nvidia Drivers Version: 461.40
    Daz Studio Version: 4.12.1.117
    Optix Prime Acceleration: 

    Benchmark Results
    2021-05-15 06:37:02.150 Finished Rendering
    2021-05-15 06:37:02.180 Total Rendering Time: 8 minutes 40.85 seconds
    2021-05-15 06:39:15.285 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2021-05-15 06:39:15.285 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce GTX 1080 Ti): 1800 iterations, 1.465s init, 517.472s render


    Iteration Rate: 3.478 iterations per second
    Loading Time: 3.378 seconds

    Iray_RenderBench.png
    900 x 900 - 1M
  • kraftwerkdkraftwerkd Posts: 2
    edited May 15

    Hope I got this right.

    System Configuration
    System/Motherboard: ASUS ROG Strix B350-I Gaming
    CPU: AMD Ryzen 5 1500X @ 3.6GHz
    GPU: Gigabyte GeForce GTX 1660 Ti (1500MHz default, boost clock OC'd to 1900MHz)
    System Memory: ADATA 16GB DDR4 @ 1200MHz
    OS Drive: ADATA HV620S SSD
    Asset Drive: WD Blue 1TB SSD (BitLocker drive encryption enabled)
    Operating System: Windows 10 Pro, version 20H2, build 19042.928
    Nvidia Drivers Version: 466.11 WHQL (non-Studio Driver)
    Daz Studio Version: 4.15.0.2
    Optix Prime Acceleration: n/a

    Benchmark Results
     

    2021-05-14 18:21:22.909 Finished Rendering<br /> 2021-05-14 18:21:22.965 Total Rendering Time: 9 minutes 29.21 seconds
     

    2021-05-14 18:21:50.667 Iray [INFO] - IRAY:RENDER ::&nbsp; &nbsp;1.0&nbsp; &nbsp;IRAY&nbsp; &nbsp;rend info : Device statistics:2021-05-14 18:21:50.667 Iray [INFO] - IRAY:RENDER ::&nbsp; &nbsp;1.0&nbsp; &nbsp;IRAY&nbsp; &nbsp;rend info : CUDA device 0 (NVIDIA GeForce GTX 1660 Ti): 1800 iterations, 3.293s init, 562.806s render


    Iteration Rate: (1800 / 562.806) = 3.198 Iterations per second
    Loading Time: ((0 * 3600 + 9 * 60 + 29.21) - 562.506) 6.704 seconds

    Post edited by kraftwerkd on
  • RayDAntRayDAnt Posts: 880

    kraftwerkd said:

    Hope I got this right.

    System Configuration
    System/Motherboard: ASUS ROG Strix B350-I Gaming
    CPU: AMD Ryzen 5 1500X @ 3.6GHz
    GPU: Gigabyte GeForce GTX 1660 Ti (1500MHz default, boost clock OC'd to 1900MHz)
    System Memory: ADATA 16GB DDR4 @ 1200MHz
    OS Drive: ADATA HV620S SSD
    Asset Drive: WD Blue 1TB SSD (BitLocker drive encryption enabled)
    Operating System: Windows 10 Pro, version 20H2, build 19042.928
    Nvidia Drivers Version: 466.11 WHQL (non-Studio Driver)
    Daz Studio Version: 4.15.0.2
    Optix Prime Acceleration: n/a

    Benchmark Results
     

    2021-05-14 18:21:22.909 Finished Rendering
    2021-05-14 18:21:22.965 Total Rendering Time: 9 minutes 29.21 seconds

     

    2021-05-14 18:21:50.667 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:2021-05-14 18:21:50.667 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (NVIDIA GeForce GTX 1660 Ti): 1800 iterations, 3.293s init, 562.806s render


    Iteration Rate: (1800 / 562.806) = 3.198 Iterations per second
    Loading Time: ((0 * 3600 + 9 * 60 + 29.21) - 562.506) 6.704 seconds

    yes

  • skyeshotsskyeshots Posts: 45

    System/Motherboard: Gigabyte X299X
    CPU: I9-10980XE @ 3.00 GHZ
    GPU: PNY RTX A6000 x4
    System Memory: 96 GB Corsair Dominator Platinum DDR4-3466
    OS Drive: Intel 670p M.2 1 TB NVMe
    Asset Drive: Intel Optane 380 GB (905p)
    Operating System: Win 10 Pro, 20H2
    Nvidia Drivers Version: 466.11 DCH
    Daz Studio Version: 4.15

    2021-05-30 13:37:29.483 Finished Rendering
    2021-05-30 13:37:29.519 Total Rendering Time: 28.67 seconds
    2021-05-30 13:37:38.293 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:
    2021-05-30 13:37:38.293 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (NVIDIA RTX A6000):      441 iterations, 1.804s init, 23.293s render
    2021-05-30 13:37:38.293 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (NVIDIA RTX A6000):      442 iterations, 1.678s init, 23.125s render
    2021-05-30 13:37:38.294 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 2 (NVIDIA RTX A6000):      456 iterations, 1.720s init, 24.011s render
    2021-05-30 13:37:38.294 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 3 (NVIDIA RTX A6000):      461 iterations, 1.702s init, 24.087s render
    Loading Time: 4.583
    Rendering Performance: 1800/24.087s = 74.73 iterations per second


    Moved the OS to a dedicated 1 TB NVMe, with the Optane drive for Daz assets. Also, pulled the NVLink bridges (for now) and switched the 3rd card down to a vertical mount w/GPU riser cable, allowing the Optane's heat spreader greatly improved airflow. I didn't think it was going to make a big difference as many single card setups tend to run a bit hotter in vertical. Avg. drop was about 10 degrees C for the GPUs and substantial drop for the Optane (perhaps 15 degrees cooler) with the 3rd GPU in vertical position.

    These are all turbine cooled cards, but something to think about if you have the option to vertical mount one of your GPUs and possibly help ventilate your NVMe. Depending on your build, it may help drop temps, especially versus pancake stacked GPUs.

     

    PXL_20210525_020942048.jpg
    867 x 850 - 248K
    PXL_20210525_021239768.jpg
    650 x 566 - 115K
  • rinkuchalrinkuchal Posts: 37
    edited July 9

    Finally got an upgrade.

    System Configuration
    System/Motherboard: Alienware m17 r4 OC Profile Off
    CPU: Intel(R) Core(TM) i7-10870H CPU @ 2.20GHz (16 Threads), ~2.2GHz
    GPU: NVIDIA GeForce RTX 3070 Laptop GPU 8GB @ stock
    System Memory: 32GB RAM
    OS Drive: Kingston NV1  2TB NVMe
    Asset Drive: Same
    Operating System: Win 10 Home 20H2
    Nvidia Drivers Version: 461.66 DCH
    Daz Studio Version: 4.15

    Benchmark Results - GPU Only
    DAZ_STATS

    2021-06-18 17:36:22.219 Finished Rendering
    2021-06-18 17:36:22.283 Total Rendering Time: 2 minutes 45.74 seconds

    IRAY_STATS

    2021-06-18 17:39:33.794 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:
    2021-06-18 17:39:33.794 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (GeForce RTX 3070 Laptop GPU): 1800 iterations, 4.106s init, 155.692s render

    Iteration Rate: (DEVICE_ITERATION_COUNT / DEVICE_RENDER_TIME) 11.56
    Loading Time: ((TRT_HOURS * 3600 + TRT_MINUTES * 60 + TRT_SECONDS) - DEVICE_RENDER_TIME) 10.04

    Added GPU details, and Alienware OC Profile.

    [TEST RENDER] Benchmark 2m45s.png
    900 x 900 - 1M
    Post edited by rinkuchal on
  • chrislbchrislb Posts: 55
    edited June 22

    I decided to try some experiments with some new hardware.  One of the issues with a RTX 3090 is when you have a high enough power limit, the power limit of the card isn't what's prevents the card from rendering faster.  The limiting factor then becomes the voltage is limiting the GPU from maintaining a higher clock speed.  Most RTX 3090 cards have a GPU voltage limit of about 1.089 volts.

    I ran the benchmark on a RTX 3090 Kingpin, which has a 450 watt or 520 watt power limit(and an optional 1000 watt power limit) along with software that allows you to modify the voltage for the GPU and memory along with adjusting the voltage changes under load(Load Line Calibration).  

    The card is designed for extreme overclocking and cooling well below ambient temperatures.  Its not uncommon to use liquid nitrogen to cool the GPU on the card down to -120C to -160C to achieve high GPU speeds and break computer benchmark records.  That's why it has software to control the voltages and LLC.

    For testing, I ran a 3090 Kingpin with its factory hybrid cooling system, which includes its own closed loop liquid cooling system with a 360mm radiator and a separate fan and heatsinks that cool the VRMs and other components.  I also didn't use the 520 watt BIOS and instead used one of the other factory BIOS versions with a ~450 watt power limit.  The GPU temperature was kept between 37C and 43C during the render benchmarks.  I used 1.25 volts for the GPU, the maximum safe voltage for ambient cooling, and adjusted the other GPU and VRAM voltages accordingly.

     

    System Configuration:

    System/Motherboard: MSI MEG ACE x570

    CPU: AMD Ryzen R9 5950X at stock speed

    GPU: EVGA RTX 3090 Kingpin Hybrid

    System Memory: 32 GB of DDR4 3666 MHz G.Skill Trident Z Neo CAS 16(Underclocked from 4000 MHz)

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

    Asset Drive: XPG SX 8100 4TB NVMe SSD

    Operating System: Windows 10 pro Build 19043.1055

    Nvidia Drivers Version: 466.77

    Daz Studio Version: 4.15.02

     

    All tests were GPU only rendering.

     

    Benchmark Results:

    I ran the card with the stock VRAM and GPU clock settings and a maximum of 1.25 volts for the GPU.

    5950X 3090 Kingpin 1.25v Stock GPU speed Stock VRAM speed(9752*2 = 19502)

    2021-06-22 00:10:41.058 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend progr: Received update to 01800 iterations after 94.785s.

    2021-06-22 00:10:41.058 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend progr: Maximum number of samples reached.

    2021-06-22 00:10:41.557 Finished Rendering

    2021-06-22 00:10:41.590 Total Rendering Time: 1 minutes 36.90 seconds

    2021-06-22 00:10:48.402 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2021-06-22 00:10:48.402 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (NVIDIA GeForce RTX 3090): 1800 iterations, 0.850s init, 94.040s render

    Iteration Rate: (1800 / 94.040) = 19.141 iterations per second

    Loading Time: ((1 minutes 36.90 seconds) - 94.040) = 2.86 seconds

     

    I ran the card with a GPU overclock of 281 MHz(2,295 peak MHz) and stock VRAM speed and a maximum of 1.25 volts for the GPU.

    5950X 3090 Kingpin 1.25v +281 GPU Stock VRAM(9752*2 = 19502)

    2021-06-21 23:47:30.233 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend progr: Received update to 01800 iterations after 89.174s.

    2021-06-21 23:47:30.239 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend progr: Maximum number of samples reached.

    2021-06-21 23:47:30.716 Finished Rendering

    2021-06-21 23:47:30.752 Total Rendering Time: 1 minutes 31.26 seconds

    2021-06-21 23:47:33.868 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2021-06-21 23:47:33.868 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (NVIDIA GeForce RTX 3090): 1800 iterations, 0.849s init, 88.435s render

    Iteration Rate: (1800 / 88.435) = 20.354 iterations per second

    Loading Time: ((1 minutes 31.26 seconds) - 88.435) = 2.825 seconds

     

    I ran the card with a GPU overclock of 281 MHz(2,295 peak MHz) and a VRAM overclock of 900MHz and a maximum of 1.25 volts for the GPU.

    5950X 3090 Kingpin 1.25v +281 GPU +900 VRAM(10652x2 = 21,304)

    2021-06-21 23:43:03.595 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend progr: Received update to 01800 iterations after 83.352s.

    2021-06-21 23:43:03.599 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend progr: Maximum number of samples reached.

    2021-06-21 23:43:04.074 Finished Rendering

    2021-06-21 23:43:04.110 Total Rendering Time: 1 minutes 25.43 seconds

    2021-06-21 23:43:06.887 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2021-06-21 23:43:06.887 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (NVIDIA GeForce RTX 3090): 1800 iterations, 0.833s init, 82.623s render

    Iteration Rate: (1800 / 82.623) = 21.786 iterations per second

    Loading Time: ((1 minutes 25.43 seconds) - 82.623) = 2.807 seconds

     

    I ran the card with a GPU overclock of 281 MHz(2,295 peak MHz) and a VRAM overclock of 1200MHz and a maximum of 1.25 volts for the GPU.

    5950X 3090 Kingpin 1.25v +281 GPU +1200 VRAM(10952x2 = 21,904)

    2021-06-21 23:59:55.378 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend progr: Received update to 01800 iterations after 81.548s.

    2021-06-21 23:59:55.378 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend progr: Maximum number of samples reached.

    2021-06-21 23:59:55.856 Finished Rendering

    2021-06-21 23:59:55.896 Total Rendering Time: 1 minutes 23.66 seconds

    2021-06-21 23:59:58.914 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2021-06-21 23:59:58.914 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (NVIDIA GeForce RTX 3090): 1800 iterations, 0.868s init, 80.808s render

    Iteration Rate: (1800 / 80.808) = 22.275 iterations per second

    Loading Time: ((1 minutes 23.66 seconds) - 80.808) = 2.852 seconds

     

    Previously, I saw minimal gains in the rendering time reduction by overclocking the VRAM(past a certian speed) on the graphics card.  However, it appear that when the GPU is running at higher clock speeds, the gains from overclocking the VRAM are more noticeable.  Being able to increase the GPU voltage to the maximum safe voltage for ambient cooling saw a reduction in render time of 13.232 seconds(94.040 seconds - 80.808 seconds) from the stock GPU and VRAM speeds.  The issue(other than increased MSRP) is the 3090 Kingpin is the only RTX 3090 on the market with this ability to change voltages with software above 1.10 volts for the GPU.  There are several other RTX 3090 cards where you can solder an external voltage controller onto the card.  However you then need to have the knowledge and skill to solder tiny connections and adjust those voltages without damaging the GPU.

    In reality, you are probably better off buying more 3090s than spending the extra money for a Kingpin 3090 or pair of Kingpin 3090 cards.  MSRP for the RTX 3090 Kingpin Hybrid is about $2100.  The Kingpin Hydro Copper, which requires a seperate open loop cooling system, has a MSRP of over $2300.  Current street price of a Kingpin Hybrid is $3300-$4000 if you are unable to get one direct from the manufacturer.  For a little more than the MSRP cost of a pair of 3090 Kingpin cards, you can buy three Nvidia founders edition RTX 3090 cards(MSRP $1499.99) or three base model cards from another manufacturer and have faster render times than a pair of RTX 3090 Kingpin cards.

    Post edited by chrislb on
  • outrider42outrider42 Posts: 2,909
    My concern would be what impact overclocking the VRAM may have on its longevity. VRAM seems to be one of the failure points found after mining long periods of time. I do not have stats for this, but it does seem logical to me that the VRAM would break down before the GPU core will. Also, numerous launch GPUs had issues with memory, like "Space Invader" glitch. This glitch was aptly named as the artifacts often looked like sprites from old classic.

    Plus the GDDR6X used in the 3090s runs pretty hot out of the box. It has been suggested this is why Ampere does not make full use of the 6X speed. While the Kingpin may be better designed for cooling it, VRAM cooling is often an afterthought in many GPU designs. In some cases you can even find cards that have some or even all of their VRAM not making contact with the cooler.
  • jd641jd641 Posts: 248
    edited June 22

    I've never done a benchmark for the thread before since my secondary card is my rendering card and I was stuck on a 970 for a very long time. Now that I've got a 3060 I thought I'd put up my results.

     

    System Configuration
    System/Motherboard: Gigabyte X299 UD4 Pro
    CPU: Intel(R) Core(TM) i7-9800X CPU @ 3.80GHz
    GPU: eVGA RTX 3060 XC Gaming
    System Memory: G.Skill Ripjaws V Series 128GB
    OS Drive: Samsung 850 Evo 1TB
    Asset Drive: Same
    Operating System: Win10 Pro 20H2 19042.746
    Nvidia Drivers Version: 466.77
    Daz Studio Version: 4.15.0.2 64-bit

    Benchmark Results

    2021-06-22 03:56:01.453 Finished Rendering
    2021-06-22 03:56:01.492 Total Rendering Time: 3 minutes 46.31 seconds

    2021-06-22 04:06:26.733 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:
    2021-06-22 04:06:26.733 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (NVIDIA GeForce RTX 3060):      1800 iterations, 2.170s init, 221.357s render


    Iteration Rate:  8.131 iterations per second?????
    Loading Time: I don't understand wtf this means at all. I suck at math, maybe someone can figure it out for me and I can edit my post later on. crying

    Post edited by jd641 on
  • chrislbchrislb Posts: 55

    outrider42 said:

    My concern would be what impact overclocking the VRAM may have on its longevity. VRAM seems to be one of the failure points found after mining long periods of time. I do not have stats for this, but it does seem logical to me that the VRAM would break down before the GPU core will. Also, numerous launch GPUs had issues with memory, like "Space Invader" glitch. This glitch was aptly named as the artifacts often looked like sprites from old classic.

     

    Plus the GDDR6X used in the 3090s runs pretty hot out of the box. It has been suggested this is why Ampere does not make full use of the 6X speed. While the Kingpin may be better designed for cooling it, VRAM cooling is often an afterthought in many GPU designs. In some cases you can even find cards that have some or even all of their VRAM not making contact with the cooler.

     

    I can understand the concern with the VRAM temperatures on some 3090 models.  Even the Nvidia Founder's Edition cards had inadequate VRAM cooling. I've taken apart the EVGA 3090 FTW3 Ultra cards, the Asus 3090 Strix cards, and the MSI Gaming X Trio and Suprim cards.  All of them had thermal pads for the rear VRAM to transfer heat elsewhere, to a metal back plate, and what appeared to be adequate cooling for the front VRAM.  They also cooled the front VRAM with the GPU heatsink.  The EVGA 3090 FTW3 cards had a thermal sensor for the front and rear VRAM to monitor its temeprature.  I've also seen some DIY solutions using either large heatsinks and thermal adhesive pads or active water cooling backplates to cool the rear of RTX 3090 cards.

    On the 3090 Kingpin Hybrid the front VRAM is cooled by the GPU cooling system with copper plates and thermal pads.  The rear side VRAM has thermal pads and a metal back plate.  The 3090 Kingpin has multiple thermal sensors for most components and the rear side VRAM thermal sensor rarely showed temperatures above 60C in rendering or gaming benchmarks.  The front side VRAM next to the GPU was 40-50C under load.  The VRAM temps dropped below 40C at idle.

    The manufacturer sells a water cooled 3090 Kingpin model and also a conversion kit to convert the hybrid model to open loop water cooling.  That also may help with VRAM temperatures because it removes heat from the GPU and less heat transfers to the rear side of the card to heat the rear VRAM.

    One other thing I noticed is that larger time reductions in this Iray 3D render benchmark don't always lead to larger reductions in render time actual larger renders at higher resolution with multiple objects.

  • outrider42outrider42 Posts: 2,909
    edited June 22

    One other thing I noticed is that larger time reductions in this Iray 3D render benchmark don't always lead to larger reductions in render time actual larger renders at higher resolution with multiple objects.

    I'm quite interested in this last bit here. Is it the bench scene or something else. Ray tracing cores accelerate the process, but are kind of a wild card, as how much they impact performance can vary depending on how the scene is built
    Post edited by outrider42 on
  • chrislbchrislb Posts: 55

    outrider42 said:

     

    One other thing I noticed is that larger time reductions in this Iray 3D render benchmark don't always lead to larger reductions in render time actual larger renders at higher resolution with multiple objects.

    I'm quite interested in this last bit here. Is it the bench scene or something else. Ray tracing cores accelerate the process, but are kind of a wild card, as how much they impact performance can vary depending on how the scene is built

     

    I was curious how the benchmark result correlate to actual renders.  I've noticed the minimal difference in the effect of overclocking in several scenes I created in Daz where I did the same render to the same number of iterations with the same hardware, but different hardware overclock settings.  The overclock settings that gave vastly different results in the benchmark yielded minimal difference in a 40+ minute render even though the temperatures of the GPU and VRAM were kept low and the overclock was maintained for the entire render.    The render time difference was minimal between stock settings and overclocked GPU and VRAM settings, less than a 20 seconds difference in a 40+ minute render for one of the scenes.  EVen though the benhmark may show a difference of 5-10% in render time results, the "real world" renders showed less than a 1% difference with the same overclock settings.

    One scene was several different cars of different color and paint finishes in an airplane hangar with sunlight coming in through the open doors.  I forget what the other scenes were that I tested.  I think one might have been a face in a mirror with DOF enabled.  No CPU rendering was used.

  • outrider42outrider42 Posts: 2,909

     

    My thinking is that the VRAM overclock effects only one aspect of the render, and this aspect presents itself more easily in a simpler scene like our bench. In my own testing, my 1080tis, which have no RT cores, also benefitted greatly by overclocking the VRAM. But I generally stopped there. I chose not to overclock my VRAM in my regular renders because these cards are now 4 years old and I don't want to rock the boat. I feel my results would likely mirror yours, though. I could probably give it a test to see.

    There may also be a limit to how much the VRAM overclock helps as more data is used in the memory. It may even be a limit in how Iray works, a software speed limit if you will.

    What might be interesting is if we can test out geometry VS shading. These are two different tasks for the GPU to calculate, and the ray tracing cores impact the geometry the most. So if we could make a scene that is geometrically complex but simple shaded, and another scene that has a lot of complex shaders but less geometry, I think the results would be very telling about how the cards perform these tasks and how a VRAM overclock impacts it.

  • nonesuch00nonesuch00 Posts: 15,363

    Supposedly the video card supply is returning to normal so maybe it's not out of the question I get to buy a RTX 3080 or RTX 3090 by end of March 2022.

  • the-brianthe-brian Posts: 1
    edited July 2

    System Configuration
    System/Motherboard: Asrock B450 Pro4
    CPU: AMD Ryzen 7 3700X @ stock (if left at defaults)
    GPU: NVIDIA RTX3060 @ stock (if left at defaults)
    System Memory: Corsair Vengeance @ 1330MHz / 2660MHz
    OS Drive:
    Asset Drive:
    Operating System: Win 10
    Nvidia Drivers Version: 471.11
    Daz Studio Version: 4.15.0.2


    Benchmark Results
    2021-07-02 19:59:10.104 Total Rendering Time: 3 minutes 36.99 seconds
    2021-07-02 19:59:21.034 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (NVIDIA GeForce RTX 3060):      166 iterations, 1.922s init, 212.764s render
    2021-07-02 19:59:21.034 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CPU:      138 iterations, 1.686s init, 212.867s render
    Iteration Rate: 7.83 iterations per second
    Loading Time: 4.226 seconds

    Somehow it used CPU + GPU. But the stats of 7.8 render/s are correct

    Post edited by the-brian on
  • PerttiAPerttiA Posts: 2,260

    System Configuration
    System/Motherboard: MSI X99A SLI PLUS
    CPU:  Intel i7-5820K @ 3.30GHz
    GPU: Asus RTX 2070 Super @ stock
    System Memory: Kingston 4x16GB DDR4 3200-CL16 @ stock
    OS Drive: Kingston 500GB SSD
    Asset Drive: 4 x Kingston 900GB SSD
    Operating System:  Windows 7 Ultimate SP1
    Nvidia Drivers Version: 456.38
    Daz Studio Version: 4.15.0.2

    Benchmark Results
    Total Rendering Time: 4 minutes 37.83 seconds
    IRAY   rend info : CUDA device 0 (GeForce RTX 2070 SUPER): 1800 iterations, 9.483s init, 261.482s render
    Iteration Rate: 6.884 iterations per second
    Loading Time: 16.35 seconds

  • outrider42outrider42 Posts: 2,909

    PerttiA said:

    System Configuration
    System/Motherboard: MSI X99A SLI PLUS
    CPU:  Intel i7-5820K @ 3.30GHz
    GPU: Asus RTX 2070 Super @ stock
    System Memory: Kingston 4x16GB DDR4 3200-CL16 @ stock
    OS Drive: Kingston 500GB SSD
    Asset Drive: 4 x Kingston 900GB SSD
    Operating System:  Windows 7 Ultimate SP1
    Nvidia Drivers Version: 456.38
    Daz Studio Version: 4.15.0.2

    Benchmark Results
    Total Rendering Time: 4 minutes 37.83 seconds
    IRAY   rend info : CUDA device 0 (GeForce RTX 2070 SUPER): 1800 iterations, 9.483s init, 261.482s render
    Iteration Rate: 6.884 iterations per second
    Loading Time: 16.35 seconds

    Thanks for posting! We haven't had a 2070 super in a while. Though a lot of focus is on the newest parts we still could use more times for previous gen cards. There are a lot of cards that have not had updated benches from the latest Daz. If your time is accurate, that 2070 Super gained a nice chunk of performance in Iray.

    If anybody reading this has a previous gen card, it would great if you took a few minutes to run the test scene.

  • GoOutlierGoOutlier Posts: 13

    System Configuration
    System/Motherboard: ASRock z370 Extreme 4
    CPU: Intel 8700k @ 4.8Ghz
    GPU: MSI 1070 Armor; EVGA 3080ti FTW3 Ultra
    System Memory: GSkill 2x16GB DDR4 3200
    OS Drive: Samsung 850 EVO 500GB
    Asset Drive: Samsung 870 EVO 2TB
    Operating System: Windows 10 Home Ver. 2004 Build 19041.1052
    Nvidia Drivers Version: 471.11 Studio
    Daz Studio Version: 4.15.0.2

    MSI GTX 1070 8GB Armor @ Stock

    Benchmark Results
    Total Rendering Time: 9 minutes 34.46 seconds
    CUDA device 0 (NVIDIA GeForce GTX 1070): 1800 iterations, 2.177s init, 569.997s render
    Iteration Rate: 3.157 iterations per second
    Loading Time: 4.463 seconds

    EVGA RTX 3080ti FTW3 ULTRA 12GB @ Stock

    Benchmark Results
    Total Rendering Time: 1 minutes 36.42 seconds
    CUDA device 0 (NVIDIA GeForce RTX 3080 Ti): 1800 iterations, 2.015s init, 92.312s render
    Iteration Rate: 19.499 iterations per second
    Loading Time: 4.108 seconds

     

    I've been experimenting with the clocks on the 3080ti and will post some results when I have more data.

     

  • System Configuration
    System/Motherboard: ASUS Z170-A
    CPU: Intel Core i7 6700 @ stock
    GPU: Nvidia RTX 3070Ti @ stock
    System Memory: Kingston Hyper X Fury 32GB DDR4 Dual Channel @ 1069MHz
    OS Drive: WD SN750 NVMe PCIe M.2 1 TB
    Asset Drive: Seagate Barracuda 4TB
    Operating System: Windows 10 Pro 20H2 Build 19042.1052
    Nvidia Drivers Version: 471.11 Studio
    Daz Studio Version: VERSION BITS
    Optix Prime Acceleration: n/a

    Benchmark Results
    Total Rendering Time: 2 minutes 21.47 seconds

    CUDA device 0 (NVIDIA GeForce RTX 3070 Ti):      1800 iterations, 2.866s init, 135.662s render

    Iteration Rate: 13.268 iterations per second
    Loading Time: 5.808 seconds

     

    Noob here!  I didn't notice any other 3070Ti results so here's my first crack.

  • System Configuration
    System/Motherboard: Aorus X570 Master
    CPU: AMD Ryzen 7 2700 8-Core @ stock
    GPU0: MSI RTX 3060 Gaming 12G  @ stock
    GPU1: Zotac RTX 3090 @ stock

    System Memory: Team Group 16GB @ 2400 + Patriot Viper 32GB @2666 (48GB total)
    OS Drive: WD Black SN750 NVMe 1TB
    Asset Drive: WD Blue NVMe 2TB
    Operating System: Windows 10 Pro 20H2
    Nvidia Drivers Version: GeForce 471.11
    Daz Studio Version: 4.15.0.2
    Optix Prime Acceleration: None

    Benchmark Results
    2021-07-07 22:01:29.461 Finished Rendering
    2021-07-07 22:01:29.569 Total Rendering Time: 1 minutes 12.84 seconds

    2021-07-07 22:01:40.738 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2021-07-07 22:01:40.738 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (NVIDIA GeForce RTX 3090): 1285 iterations, 3.370s init, 65.872s render

    2021-07-07 22:01:40.738 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (NVIDIA GeForce RTX 3060): 515 iterations, 3.299s init, 65.948s render

    Iteration Rate: (1800 iterations/65.948s) 27.29 iterations per second
    Loading Time: (Total render time - Iray render) 6.332 seconds

  • Dim ReaperDim Reaper Posts: 647

    I finally managed to get hold of a 3090.  The day before it arrived, I did a few runs of the benchtest scene with the current cards in order to give a comparison using the same version of DS, same drivers etc.  Been a while since I posted results here, so hopefully there are no mistakes in the calculations.

    System Configuration
    System/Motherboard: ASUS X99-S
    CPU: Intel i7 5960X @3GHz / 3.5GHz boost
    GPU1: 2080Ti
    GPU 2: 1080Ti

    (New GPU 1: 3090)

    System Memory: 32GB KINGSTON HYPER-X PREDATOR QUAD-DDR4
    OS Drive: Samsung M.2 SSD 960 EVO 250GB
    Asset Drive: 1.8TB HDD
    Operating System: Windows 10 Pro version 2004 OS build 19041.985
    Nvidia Drivers Version: 471.11
    Daz Studio Version: 4.15.0.2
    Optix Prime Acceleration: n/a

     

    1080Ti only:

    2021-07-10 10:09:22.838 Total Rendering Time: 6 minutes 30.23 seconds

    2021-07-10 10:09:38.472 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2021-07-10 10:09:38.472 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (NVIDIA GeForce GTX 1080 Ti): 1800 iterations, 2.827s init, 385.096s render

    Iteration Rate: (1800/385.1) = 4.674 iterations per second

    Loading time: ((0+360+30.2)-385.1) = 5.100 seconds

     

    2080Ti only:

    2021-07-10 10:01:15.360 Total Rendering Time: 3 minutes 36.44 seconds

    2021-07-10 10:02:09.966 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2021-07-10 10:02:09.966 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (NVIDIA GeForce RTX 2080 Ti): 1800 iterations, 3.299s init, 209.680s render

    Iteration Rate: (1800/209.67) = 8.585 iterations per second

    Loading time: ((0+180+36.44)-209.67) = 6.770 seconds

     

    1080Ti + 2080Ti

    2021-07-10 10:12:37.004 Total Rendering Time: 2 minutes 23.70 seconds

    2021-07-10 10:12:48.120 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2021-07-10 10:12:48.121 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (NVIDIA GeForce RTX 2080 Ti): 1176 iterations, 2.792s init, 138.616s render

    2021-07-10 10:12:48.121 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (NVIDIA GeForce GTX 1080 Ti): 624 iterations, 2.786s init, 138.289s render

    Iteration Rate: (1800/138.6) = 12.987 iterations per second

    Loading time: ((0+120+23.7)-138.6) = 5.100 seconds

     

    3090

    2021-07-11 16:14:59.295 Total Rendering Time: 1 minutes 44.91 seconds

    2021-07-11 16:15:13.341 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2021-07-11 16:15:13.341 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (NVIDIA GeForce RTX 3090): 1800 iterations, 9.414s init, 92.093s render

    Iteration Rate: (1800/92.1) = 19.544 iterations per second

    Loading time: ((0+60+44.91)-92.1) = 12.810 seconds

     

    I do plan to put the 2080Ti back in for iray rendering, but unfortunately the support bracket that came with the 3090 takes up all of the other slots, so I need to find another anti-sag solution first.  For now though, I'm pleased that the new card is much faster than the other two together.

  • Dim ReaperDim Reaper Posts: 647
    edited July 17

    I found an alternative support bracket and so was able to install the 2080Ti.  The iteration rate is about where you would expect given the linear scaling of iray.

    System Configuration
    System/Motherboard: ASUS X99-S
    CPU: Intel i7 5960X @3GHz / 3.5GHz boost
    GPU1: 3090
    GPU 2: 2080Ti

    System Memory: 32GB KINGSTON HYPER-X PREDATOR QUAD-DDR4
    OS Drive: Samsung M.2 SSD 960 EVO 250GB
    Asset Drive: 1.8TB HDD
    Operating System: Windows 10 Pro version 2004 OS build 19041.985
    Nvidia Drivers Version: 471.11
    Daz Studio Version: 4.15.0.2
    Optix Prime Acceleration: n/a

    RTX 3090 + RTX 2080Ti

    2021-07-17 14:51:00.720 Total Rendering Time: 1 minutes 11.75 seconds

    2021-07-17 14:51:09.811 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2021-07-17 14:51:09.811 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 1 (NVIDIA GeForce RTX 2080 Ti): 547 iterations, 2.875s init, 65.364s render

    2021-07-17 14:51:09.811 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (NVIDIA GeForce RTX 3090): 1253 iterations, 2.743s init, 64.928s render

    Iteration Rate: (1800/65.15) = 27.629 iterations per second

    Loading time: ((0+60+12.76)-65.15) = 7.610 seconds

    I'm a bit confused as to why the loading time is much lower than it was with just the 3090 alone.

    In case the information is useful to anyone wanting to upgrade and still keep their older card in the system, I am running a 1200W PSU.  Measuring the power draw of the whole machine using a meter at the plug socket, over several different renders the maximum power draw from the plug socket was 767W.

    Post edited by Dim Reaper on
  • RayDAntRayDAnt Posts: 880

    I'm a bit confused as to why the loading time is much lower than it was with just the 3090 alone.

    If this was your very first use of the 3090 in DS/Iray after making any sort of system hardware config adjustments, could be that the time increase was just a one-time thing due to first-run compiling at the driver level. The way to tell would be to run just the 3090 on the benchmark a couple times in a row and see if the loading time is consistent or not.

  • Dim ReaperDim Reaper Posts: 647
    edited July 17

    RayDAnt said:

    Dim Reaper said:

    I'm a bit confused as to why the loading time is much lower than it was with just the 3090 alone.

    If this was your very first use of the 3090 in DS/Iray after making any sort of system hardware config adjustments, could be that the time increase was just a one-time thing due to first-run compiling at the driver level. The way to tell would be to run just the 3090 on the benchmark a couple times in a row and see if the loading time is consistent or not.

     

    Thanks for the suggestion.  I've just run the benhcmark scene 4 times on only the 3090.  The first time was 4-5 seconds longer than any subsequent runs - even after shutting DS and reloading.  I think you are correct about it compiling something, but I hadn't changed anything before doing these 4 runs.

     

    just loaded:

    2021-07-17 18:43:15.640 Total Rendering Time: 1 minutes 44.90 seconds

    2021-07-17 18:43:58.155 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2021-07-17 18:43:58.155 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (NVIDIA GeForce RTX 3090): 1800 iterations, 8.729s init, 92.900s render

     

    Run once more

    2021-07-17 18:46:39.486 Total Rendering Time: 1 minutes 39.49 seconds

    2021-07-17 18:46:44.802 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2021-07-17 18:46:44.802 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (NVIDIA GeForce RTX 3090): 1800 iterations, 2.579s init, 94.649s render

     

    Re-load

    2021-07-17 18:52:32.666 Total Rendering Time: 1 minutes 39.4 seconds

    2021-07-17 18:54:21.808 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2021-07-17 18:54:21.808 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (NVIDIA GeForce RTX 3090): 1800 iterations, 2.867s init, 93.066s render

     

    Run again

    2021-07-17 18:57:53.225 Total Rendering Time: 1 minutes 37.92 seconds

    2021-07-17 18:58:15.376 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:

    2021-07-17 18:58:15.376 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (NVIDIA GeForce RTX 3090): 1800 iterations, 2.618s init, 92.999s render

     

    EDIT:

    Just out of interest, I left the scene loaded after the last run with the 3090 and then enabled the 2080ti.  For the first run, the load time was 11.3 sec, but went down to 5.3 seconds for the second run.

    Post edited by Dim Reaper on
  • JamesJABJamesJAB Posts: 1,736

    Just got and installed my shiny new RTX A5000 24GB card!

    System Configuration
    System/Motherboard: Dell Precision T7610
    CPU: Dual Intel Xeon E5-2650 V2 @ 2.60GHz
    GPU: PNY Quadro RTX A5000 24GB
    System Memory: 64GB quad chanel 1600MHz Reg ECC
    OS Drive: 1TB WD SATA SSD - WDS100T2B0B-00YS70
    Asset Drive: 4TB WD RE4 - WD40EZRX-00S
    Operating System: Windows 10 Pro 21H1
    Nvidia Drivers Version: 471.11
    Daz Studio Version: 4.15.0.14 Public Build

    Benchmark Results
    2021-07-20 18:52:23.964 Finished Rendering
    2021-07-20 18:52:24.064 Total Rendering Time: 2 minutes 15.7 seconds
    2021-07-20 18:55:04.599 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : Device statistics:
    2021-07-20 18:55:04.599 Iray [INFO] - IRAY:RENDER ::   1.0   IRAY   rend info : CUDA device 0 (NVIDIA RTX A5000): 1800 iterations, 15.520s init, 113.449s render

    Iteration Rate: (1800 / 113.449) 15.8
    Loading Time: ((0 + 120 + 15.7) - 113.5) = 22.2 seconds

    I'd sat that's a pretty good result for a single 8 pin power connector card that tops out at 230W.

Sign In or Register to comment.