Daz 3D is part of 
Connect
DAZ Productions, Inc.
7533 S Center View Ct #4664
West Jordan, UT 84084
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2024 Daz Productions Inc. All Rights Reserved.
Comments
Well, I did grab a new card over the Black Friday/Cyber Monday weekend, so I can now pretend I'm sort of contributing to the thread, as even if this is the baby card of the 1600 series, I believe it's a card that has not yet been benchmarked for Iray.
System/Motherboard: Hewlett Packard IPM87-MP
CPU: Intel Core i7 4790 @ stock MHz (100mV undervolt)
GPU: Gigabyte GTX 1650 LP OC @ 1695 MHz boost (Factory overclock, Reference 1665 Mhz)
System Memory: 2x Micron PC3-12800U-11-11-B1 8GB DDR3 @ 1600 MHz
OS Drive: Samsung 850 EVO 250GB mSATA
Asset Drive: Seagate Barracuda Green ST2000DL003 2 TB
Operating System: Windows 8.1 Home 64-bit, Version 6.3, Build 9600
Nvidia Drivers Version: 441.41
Daz Studio Version: Daz Studio 4.12.0.086
Optix Prime Acceleration: On
Benchmark Results:
2019-12-05 13:52:48.268 Total Rendering Time: 19 minutes 33.25 seconds
2019-12-05 13:52:51.192 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (GeForce GTX 1650): 1800 iterations, 8.574s init, 1161.116s render
Iteration Rate: 1.550 iterations per second
Loading Time: 12.134 seconds
I have to say, I'm pretty happy. Based on the gaming benchmarks vs. my old 1050 Ti, I was expecting the 1650 to be a 30% speed improvement and hoping for 40%, so a 50% increase is very welcome. It's a shame that the 1650 Super turned out to be a 100 Watt card that won't end up in a low profile version, as that could have been particularly awesome.
(Okay, a 50% performance increase probably wouldn't be worth chasing with more expensive cards, but with the relatively low cost of cards at this level and the fact that good low profile cards retain their value pretty well, the exercise of upgrading will have only set me back about £20-30 once I sell on the 1050 Ti, and this does again put me on the top low profile card, so I'd say reasonably worthwhile all in all).
Other than a driver update, the only other difference to note versus the system I ran before is that I have now undervolted the CPU by 100 mV in order to get the system a bit cooler and quieter, but as both were GPU-only renders and the system wasn't throttling before, this won't really have had much impact.
System Configuration
System/Motherboard: Custom Build/Asus Crosshair V Formula Z
CPU: AMD FX-8370 @ Stock
GPU: nVidia Geforce RTX 2060 @ Stock, : nVidia Geforce RTX 2060 @ Stock
System Memory: Corsair DDR3-2400 16GB
OS Drive: Samsung SSD 850 EVO 250GB
Asset Drive: WDC 100GB
Operating System: Windows 10 Pro 64bit Build 18362.476
Nvidia Drivers Version: 441.41
Daz Studio Version: Daz Studio 4.12.0.086
Optix Prime Acceleration: On and Off
2 GPU Optix On
2019-12-04 23:07:40.306 Total Rendering Time: 4 minutes 13.3 seconds
2019-12-04 23:07:48.896 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (GeForce RTX 2060): 896 iterations, 6.504s init, 241.227s render
2019-12-04 23:07:48.896 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 1 (GeForce RTX 2060): 904 iterations, 6.693s init, 240.680s render
Iteration Rate: (1800 / 241.227) = 7.462 iterations per second
Loading Time: ((0 + 240 + 13.3) - 241.227) = (253.3 - 241.227) = 12.073 seconds
1 GPU Optix On-drives monitor
2019-12-04 23:18:58.485 Total Rendering Time: 8 minutes 2.15 seconds
2019-12-04 23:19:04.776 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (GeForce RTX 2060): 1800 iterations, 4.935s init, 473.345s render
Iteration Rate: (1800 / 473.345) = 3.803 iterations per second
Loading Time: ((0 + 480 + 2.15) - 473.345) = (482.15 - 473.345) = 8.805 seconds
1 GPU second card Optix On
2019-12-04 23:28:09.079 Total Rendering Time: 7 minutes 58.81 seconds
2019-12-04 23:28:13.070 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 1 (GeForce RTX 2060): 1800 iterations, 5.011s init, 469.870s render
Iteration Rate: (1800 /469.870 ) = 3.831 iterations per second
Loading Time: ((0 + 420 +58.81 ) - 469.870) = (478.81 - 469.870) = 8.94 seconds
1 GPU second card Optix Off
2019-12-04 23:38:58.433 Total Rendering Time: 7 minutes 58.65 seconds
2019-12-04 23:39:03.312 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 1 (GeForce RTX 2060): 1800 iterations, 4.995s init, 469.746s render
Iteration Rate: (1800 /469.746 ) = 3.832 iterations per second
Loading Time: ((0 + 420 + 58.65) - 469.746) = 8.904 seconds
CPU + 2 GPU Optix On
2019-12-04 23:45:34.584 Total Rendering Time: 4 minutes 28.38 seconds
2019-12-04 23:45:39.905 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (GeForce RTX 2060): 885 iterations, 6.019s init, 256.602s render
2019-12-04 23:45:39.905 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 1 (GeForce RTX 2060): 883 iterations, 6.412s init, 256.266s render
2019-12-04 23:45:39.905 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CPU: 32 iterations, 5.840s init, 258.582s render
Iteration Rate: (1800 /258.582 ) = 6.961 iterations per second
Loading Time: (0 + 240 +28.38 ) -258.582 ) = 9.798 seconds
Thanks @LenioTG! Honestly it’s bugged me for years how little consistent info there was on this stuff. And since I recently came into posession of a formal education in data gathering/analysis of this very sort, I figured why not put my idle moments to good use?
Been very happy with the results so far. The level of consistency between people’s results has been especially impressive considering we’ve got people from likely all over the world reporting in in all sorts of different testing environments. There are lots of reasons to hate on Nvidia, but hardware quality clearly isn’t (currently) one of them.
The secret to finding IRAY_STATS in the log file is to first completely close down Daz Studio and then open/refresh the log file. Device Statistics are only reported at the close of an Iray instance. And since Iray Instances persist in active memory even after initial completion conditions have been met (to facilitate render resumption), the only surefire way to see those stats is by closing DS and potentially waiting an extra bit for the process itself to fully shutdown in the background. And then check the log file.
I’d caution against making any direct comparison between the numbers in this thread and the results you got without those Iray statistics. The “1800 iterations after XXX.XXXs” line you did find confounds max device Init time with max device Render time as found in the elusive Iray Device Statistics. Meaning that it is always going to undereport true device rendering performance (in putting together this thread I ended up having to study the relationships between pretty much every time based statistic given in the log file to determine which ones were most reliable/relevant to performance metering. This stat turned out to be one of the ones to avoid.)
System Configuration
System/Motherboard: Gigabyte x470 Aorus Gaming 5 Wifi (Bios F50a latest)
CPU: Ryzen 7 2700X/stock
GPU: Asus RTX 2080ti Turbo
System Memory: Ripjaws V DDR4-3600 4x16GB (64GB) Manually Clocked to 3200 for stability
OS Drive: Samsung SSD 950 PRO 512GB
Asset Drive: WD My Passport 0827 USB Device 3TB
Operating System: Windows 10 Pro Version 1809 OS build 17763.529
Nvidia Drivers Version: 26.21.14.4120(NVIDIA 441.20)
Daz Studio Version: 4.12.1.40
GPU Only
DAZ_STATS
2019-12-07 14:35:12.481 Finished Rendering
2019-12-07 14:35:12.532 Total Rendering Time: 4 minutes 7.37 seconds
IRAY_STATS
2019-12-07 14:36:06.593 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2019-12-07 14:36:06.593 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (GeForce RTX 2080 Ti): 1800 iterations, 2.420s init, 240.813s render
Iteration Rate: (1800 / 240.813) 7.474679523115446
Loading Time: ((0 + 4 * 60 + 7.37) - 240.813) 6.557
CPU Only
DAZ_STATS
2019-12-07 17:21:17.653 Finished Rendering
2019-12-07 17:21:17.713 Total Rendering Time: 49 minutes 3.10 seconds
IRAY_STATS
2019-12-07 17:21:28.671 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2019-12-07 17:21:28.671 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CPU: 1800 iterations, 2.092s init, 2937.642s render
Iteration Rate: (1800 / 2937.642) 0.6127363375115143
Loading Time: ((0 + 49 * 60 + 3.1) - 2937.642) 5.458
System Configuration
System/Motherboard: Gigabyte B450M AORUS M
CPU: Ryzen 5 3600 @ Stock speed
GPU: RTX 2070 SUPER MSI Ventus @ Stock speed
System Memory: 32Gb (4x8) HyperX Fury @ 3000Mhz
OS Drive: Samsung 970 Evo 250Gb
Asset Drive: Samsung 860 1TB
Operating System: Windows 10 build 18362
Nvidia Drivers Version: 441.66
Daz Studio Version: 4.12.0.86 Pro Edition (64-bit)
Optix Prime Acceleration: OFF (Daz Studio 4.12.1.086 or earlier only)
Benchmark Results
DAZ_STATS
2019-12-17 19:56:14.964 Finished Rendering
2019-12-17 19:56:14.993 Total Rendering Time: 5 minutes 15.19 seconds
IRAY_STATS
I still can't find this, sorry. I've closed Daz right away this time and then I've searched for the log.
2019-12-17 19:56:14.359 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Received update to 01800 iterations after 312.266s.
2019-12-17 19:56:14.359 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Maximum number of samples reached.
Iteration Rate: 1800 iterations/309,224s=5,82
Loading Time: 2019-12-17 19:51:05.148 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Received update to 00001 iterations after 3.042s.
Close and reopen Notepad. Then use it to navigate to "c:\Users\[your username here]\AppData\Roaming\DAZ 3D\Studio4\log.txt" You should now find the missing Iray statistics.
System Configuration
System/Motherboard: ASUS ROG Maximus IX CODE
CPU: Intel I7-7700K BRAND MODEL @ 4.2 GHz
GPU: EVGA RTX 2080 + EVGA GTX 1080 (neither overclocked)
System Memory: Corsair Dominator Platinum CMD32GX4M2B3000C15 Total 64GB Dual Channel DRAM Frequency 1071 MHz, NB Frequency 4119 MHz
OS Drive: Samsung 850 Pro 512GB
Asset Drive: Samsung 860 EVO 1TB
Operating System: Windows 10 64-bit Version 1909 Build 18363.535
Nvidia Drivers Version: 441.28 Driver type Standard
Daz Studio Version: 4.12.1.40 Public Build (BETA) 64-bit
Optix Prime Acceleration: OFF
Benchmark Results
2020-01-05 08:33:22.319 Finished Rendering
2020-01-05 08:33:22.350 Total Rendering Time: 3 minutes 43.88 seconds
2020-01-05 08:34:43.624 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2020-01-05 08:34:43.639 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (GeForce RTX 2080): 1266 iterations, 3.945s init, 217.087s render
2020-01-05 08:34:43.639 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 1 (GeForce GTX 1080): 534 iterations, 6.455s init, 214.569s render
Iteration Rate - RTX 2080: 5.65 iterations/second
Iteration Rate - GTX 1080 : 2.38 iterations/second
Loading Time - RTX 2080 - 6.8 sec
Loading Time - GTX 1080 - 9.31 sec
FYI updated all tables to reflect latest results.
Hey, thanks for this thread! Not sure why it's not stickied in the Daz Studio area or something. There is enough questions about this all over the forums that it seems it should be consolidated.
Heard some comments about GTX cards recently, so went searching. I am looking at a 1660ti versus a RTX 2060 in a new laptop. I am running a GTX960M in my laptop now (probably not worth benchmarking since it is 6 years old, and there is already results on a 960) I had seen this note on a post: "The RTX card will render anywhere from 20% to 50% faster than the 1660ti", which leans me towards the RTX, of course. Then I found this thread, and it looks like about a +/-30% increase for the RTX, and it will be a 75% or so increase in speed versus my 960, woohoo!
The gaming speed seems to be about 14% increase on average, so that solidifies it for me. (https://youtu.be/X59_-ol8H88) I'll post a benchmark when I get it.
Again, awesome thread!
@3dOutlaw A few notes first:
RTX is a little different than past generations in the performance does not scale exactly the same. With pure CUDA the performance generally scales exactly the same across all of the different scenes you may make. And the 2060 has a major CUDA advantage over 960M, it has more and they are faster on top of that. But then you have the RT cores, and these are a bit different. The performance boost you get from RT cores will vary depending on the content of your scene. So you might not always get the exact same speed boost, it might be a bit less or a bit more than what this test does. But it is basically like this: The more geometrically complex your scenes are, the more of a benefit the RT cores become over pure CUDA rendering. So the more stuff you have in your scene, the better the investment in RTX becomes! That is also why I generally recommend RTX to anyone, those RT cores are the big game changers for Iray rendering.
Next up, your calculated performance gains are off, you are actually underestimating them, and by a lot. That is because the laptop versions of GPUs are a little different from the desktop. How much of a difference has changed since the 960M. The 960M is a TOTALLY different GPU than a full GTX 960. They have different CUDA counts, clocks, and are built on different dies, and you can guess the 960M's specs are vastly lower in each. The 960M is a pale imitation of the full 960. The desktop 960 can range from 30% to a whopping 80% faster than the 960M depending on the task. They really are not in the same league at all.
The script changes a bit with newer laptops, but still depends on which you buy. Today Nvidia actually uses the same chips in laptops that they do for desktops, so that is an obvious plus as you now get the same CUDA counts that desktops do. However they are still downclocked compared to the desktop (in the 2060's case, it is clocked about 30% less than desktop). The result is that while they perform slower than the desktop version, they are not as horribly handicapped like the 960M Vs the 960. At least I think. Looking at Notebookcheck gives me pause because those numbers are way less than a desktop 2060. https://www.notebookcheck.net/NVIDIA-GeForce-RTX-2060-Laptop-Graphics-Card.384946.0.html
According to that link, the desktop 1660ti actually beats the laptop 2060. So this again points out the differences in desktop and laptop. BUT that is not Iray, and the 1660ti has no RT cores like the 2060 does. So I am sure that once you factor RT cores in, the laptop 2060 would win. Still...laptops...
There was going to be what they call a "Max Q" version of the 2060, but this was cancelled. That's good! LOL. The Max-Q versions cut the clockspeeds drastically to save on power for like super thin notebooks. But this obviously reduces the performance a lot as well. That is not good for Iray.
Even though the Max-Q 2060 is cancelled, I still wanted to make a note of this here in case anybody happens to be reading this thread and is looking at the other Max-Q models of the 2070 or 2080. Of course I personally would recommend not buying a laptop at all, but if one absolutely must buy a laptop for rendering, I would steer clear of Max-Q versions of anything. The website I linked is a great source for comparing them, too bad they don't bench Iray.
Now, since we do not actually have a 960M benchmark, I would like to ask you if you are willing to try it. It would be nice to have that data point, plus it would give you a better idea of what performance to expect. Though these are desktop versions, and again the laptop version will be a little slower. There is one 2070 laptop benched, but that's a 2070. I will warn you, this bench might take a while for your 960M to complete. It took a regular 960 nearly 30 minutes, so it could take yours 40 to 45 or maybe even more. So I can understand if you don't want to run it.
Meanwhile, desktop 2060's can run this test in about 8 minutes. A laptop version may be 15-30% slower than that mark. But that would still be a massive gain over a 960M.
Just updating for the latest nVidia/IRAY/Daz software updates. Nothing major changed. Slightly slower times, by a fraction of a few seconds. Major instability with Daz/IRAY, trying to render anything substantial. (Beta version of Daz.)
No sense doing TCC, Added CPU... (CPU was a bit faster, due to Embree. Shaved 5 min off a 30 min render.)
System Configuration
System/Motherboard: ASRock x299 OC Formula (PCIe 8x/8x/8x/8x speeds)
CPU: Intel i9-7980XE 3.80Ghz, 18 core (36 threads), 44 PCIe lanes @ Stock (Water cooled)
GPU: Nvidia Titan V @ stock (5120 cuda cores)
GPU: Nvidia Titan V @ stock (5120 cuda cores)
GPU: Nvidia Titan Xp Collectors Edition @ stock (3840 cuda cores)
GPU: Nvidia Titan Xp Collectors Edition @ stock (3840 cuda cores)
System Memory: Corsair Dominator Platinum 64GB DDR4 @ 2133Mhz
OS Drive: Samsung 960 PRO 2TB NVMe SSD M.2
Asset Drive: Samsung 850 EVO 4TB V-NAND SSD
Operating System: Windows 10 Home (64-bit) version 1903 build 18362.535
Nvidia Drivers Version: 441.66 STUDIO DRIVERS
Single card tests
Benchmark Results - Titan-V #1 (WDDM Mode)
Daz Studio Version: 4.12.1.40 Beta x64
Optix: OptiX Prime version 5.0.1
Total Rendering Time: 3 minutes 56.63 seconds
CUDA device 1 (TITAN V): 1800 iterations, 2.673s init, 230.437s render
Iteration Rate: (1800 / 230.437) = 7.811 iterations per second
Loading Time: (236.63 - 230.437) = 6.193 seconds
Benchmark Results - Titan-Xp Collectors Edition #1 (WDDM Mode)
Daz Studio Version: 4.12.1.40 Beta x64
Optix: OptiX Prime version 5.0.1
Total Rendering Time: 6 minutes 57.28 seconds
CUDA device 2 (TITAN Xp Collectors Edition): 1800 iterations, 2.521s init, 412.073s render
Iteration Rate: (1800 / 412.073) = 4.368 iterations per second
Loading Time: (417.28 - 412.073) = 5.207 seconds
Benchmark Results - Titan-V #2 (WDDM Mode)
Daz Studio Version: 4.12.1.40 Beta x64
Optix: OptiX Prime version 5.0.1
Total Rendering Time: 3 minutes 55.74 seconds
CUDA device 0 (TITAN V): 1800 iterations, 2.141s init, 230.851s render
Iteration Rate: (1800 / 230.851) = 7.797 iterations per second
Loading Time: (235.74 - 230.851) = 4.889 seconds
Benchmark Results - Titan-Xp Collectors Edition #2 (WDDM Mode)
Daz Studio Version: 4.12.1.40 Beta x64
Optix: OptiX Prime version 5.0.1
Total Rendering Time: 6 minutes 55.46 seconds
CUDA device 3 (TITAN Xp Collectors Edition): 1800 iterations, 2.169s init, 410.585s render
Iteration Rate: (1800 / 410.585) = 4.384 iterations per second
Loading Time: (415.46 - 410.585) = 4.875 seconds
MULTI-CARD
Benchmark Results - 2x Titan-V & 2x Titan-Xp Collectors Edition (WDDM Mode)
Daz Studio Version: 4.12.1.40 Beta x64
Optix: OptiX Prime version 5.0.1
Total Rendering Time: 1 minutes 24.84 seconds
CUDA device 1 (TITAN V): 579 iterations, 3.343s init, 78.731s render
CUDA device 2 (TITAN Xp Collectors Edition): 316 iterations, 2.547s init, 79.174s render
CUDA device 0 (TITAN V): 591 iterations, 2.406s init, 79.333s render
CUDA device 3 (TITAN Xp Collectors Edition): 314 iterations, 3.068s init, 78.709s render
Iteration Rate: (1800 / 78.986) = 22.779 iterations per second
Loading Time: (88.84 - 78.986) = 9.854 seconds
CPU
Benchmark Results - CPU (i9-7980XE 3.80Ghz, 18 core (36 threads))
Daz Studio Version: 4.12.1.40 Beta x64
Renderer: Using Embree 2.8.0
Total Rendering Time: 25 minutes 43.52 seconds
CPU: 1800 iterations, 2.038s init, 1538.749s render
Iteration Rate: (1800 / 1538.749) = 1.170 iterations per second
Loading Time: (1543.52 - 1538.749) = 4.771 seconds
System/Motherboard: ASUSTeK H87M-PLUS Rev X.0x
CPU: Intel Core i7-4770 CPU @ 3.40GHz/stock
GPU: Gainward GeForce GTX 980 Ti/stock
System Memory: 32 GB DDR3 1600 MHZ DIMM (Don't know brand)
OS Drive: Samsung SSD 850 EVO 500GB
Asset Drive: ST2000DX001-1NS164
Operating System: Windows 10, Build 9200, No Service Pack Installed
Nvidia Drivers Version: 441.66 (26.21.14.4166)
Daz Studio Version: 4.12.1.40 (BETA)
Optix Prime Acceleration:
Benchmark Results
2020-01-16 01:49:31.150 Finished Rendering
2020-01-16 01:49:31.181 Total Rendering Time: 10 minutes 33.8 seconds
2020-01-16 01:56:37.742 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2020-01-16 01:56:37.742 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (GeForce GTX 980 Ti): 1800 iterations, 4.715s init, 624.731s render
Iteration Rate: (1800 / 624.731) = 2.8812
Loading Time: (633.8 - 624.731) = 9.069
System Configuration (CPU AND GPU)
System/Motherboard: Asus ROG STRIX TRX40-E Gaming
CPU: AMD Threadripper 3970x @ Stock
GPU: GPU1 Asus RTX 2080 Super @ stock , GPU2 Asus RTX 2080 Super @ stock
System Memory: GSKILL Trident Z Neo 64GB/Quad Channel @ 3600MHz
OS Drive: Corsair MP600 NVMe Gen4 1TB
Asset Drive: Samsung 970 EVO Plus 1TB
Operating System: Windows 10 Pro 64bit 1909
Nvidia Drivers Version: Studio driver version 441.87
Daz Studio Version: 4.12.0.86 64bit
Optix Prime Acceleration: ON
Benchmark Results
Total Rendering Time: 2 minutes 27.12 seconds
CUDA device 1 (GeForce RTX 2080 SUPER): 720 iterations, 3.022s init, 141.039s render
CUDA device 0 (GeForce RTX 2080 SUPER): 721 iterations, 2.698s init, 141.340s render
CPU: 359 iterations, 1.615s init, 141.832s render
Rendering Performance: 1800 / 141.832 iterations per second
Loading Time: 0 + 120 + 27.12 - 141.832 seconds = 5.288
Hey xionis!
If you don't mind, I'd love to see you run this test with just the Threadripper 3970X. It'd be nice to know what kind of numbers it can put up vs. the 1950X result we have in this thread already!
This is relevant for those times that a render might go to CPU only, due to a large scene or maybe if your GPUs are tied up doing something else. You probably should optimize at that point to squeeze the scene into your GPUs, but inquiring minds want to know!
I did run it last night as well as the 2 cards alone with and without SLI enabled and also removing the nvlink to see if theres a difference, but I was just so tired when I made that post that I didn't feel like posting all the results. However if i recall correctly it did the render in 11min 30sec, but ill try to post all my results tonight.
System Configuration (CPU ONLY)
System/Motherboard: Asus ROG STRIX TRX40-E Gaming
CPU: AMD Threadripper 3970x @ Stock
GPU: GPU1 Asus RTX 2080 Super @ stock , GPU2 Asus RTX 2080 Super @ stock
System Memory: GSKILL Trident Z Neo 64GB/Quad Channel @ 3600MHz
OS Drive: Corsair MP600 NVMe Gen4 1TB
Asset Drive: Samsung 970 EVO Plus 1TB
Operating System: Windows 10 Pro 64bit 1909
Nvidia Drivers Version: Studio driver version 441.87
Daz Studio Version: 4.12.0.86 64bit
Optix Prime Acceleration: OFF
Benchmark Results
Total Rendering Time: 11 minutes 30.8 seconds
CPU: 1800 iterations, 1.686s init, 686.133s render
Rendering Performance: 1800 / 686.133 iterations per second
Loading Time: 0 + 660 + 30.8 - 686.133 seconds = 4.667
System Configuration (2 GPU ONLY)
System/Motherboard: Asus ROG STRIX TRX40-E Gaming
CPU: AMD Threadripper 3970x @ Stock
GPU: GPU1 Asus RTX 2080 Super @ stock , GPU2 Asus RTX 2080 Super @ stock
System Memory: GSKILL Trident Z Neo 64GB/Quad Channel @ 3600MHz
OS Drive: Corsair MP600 NVMe Gen4 1TB
Asset Drive: Samsung 970 EVO Plus 1TB
Operating System: Windows 10 Pro 64bit 1909
Nvidia Drivers Version: Studio driver version 441.87
Daz Studio Version: 4.12.0.86 64bit
Optix Prime Acceleration: OFF
Benchmark Results
Total Rendering Time: 2 minutes 35.43 seconds
CUDA device 0 (GeForce RTX 2080 SUPER): 895 iterations, 1.814s init, 151.382s render
CUDA device 1 (GeForce RTX 2080 SUPER): 905 iterations, 1.926s init, 151.065s render
Rendering Performance: 1800 / 151.382 iterations per second
Loading Time: 0 + 120 + 35.43 - 151.382 seconds = 4.048
Well the 3970x is certainly the fastest CPU only bench posted by quite a margin so far. 11.5 minutes places it in the range of the 1070ti/1080 and 1660ti GPUs, which is brand new territory for CPU performance in Iray. Of course, that is a $2000 CPU compared to those GPUs, but a CPU would not be limited by their VRAM. Kyotokid is somewhere contemplating how he can manipulate the lottery in his favor.
The other downside is that in spite of its obvious speed, when using two new GPUs it barely added to the render speed. The iteration speed dropped a bit, but not that much, when rendering with the 2080 Supers. It shows just how fast the RTX cards are. Just think, it was 3 years ago that the 1080 was one of the fastest GPUs in the world. The 3970x can match this GPU is rendering speed in Iray, which itself is a very impressive feat. But the 2080 Supers are SO much faster than the previous generation GPUs thank to RT cores. And when you have two of them, well, you have these results.
All around these are exciting results, because it does show that somebody can pursue Iray with just CPU and still get a result matching a high end (if last generation) GPU. That is not a statement we could have made a couple years ago.
Now start thinking about the 3990X 64 core Threadripper that is supposed to launch in the next few days. While I'm sure heat throttling will keep if from fully halving the 3970X render times, somewhere between 6-8 minutes should be easily doable. That puts the 3990X within the range of CURRENT GEN Nvidia GPUs, assuming that Daz Studio can handle it.
Plus, as you noted, the (V)RAM limit should be almost a non issue as you'll probably install at least 64 GB of ram in said 3990X TRX 40 system... of course at $4 grand, you can buy a few 2080 Ti's at that price. But this is less than the cost of Nvidia's high end Quadro cards, The latest top end Quadro cards are sitting at $4K for the RTX 6000, and $5.5k for the RTX 8000. The RTX 8000 does have 48GB of VRAM, and would be faster, but if you are using a CPU based rendering engine like 3Delight, it won't be of much help. The RTX Titan is probably 'better bang for the buck' vs. the RTX 6000 at $2.5K of course.
With the RTX Titan clocking in at around 4 minutes, well if the 3990X can manage to pull off a 7 minute result, and then consider everything else you could use that Threadripper for... well these are pretty exciting times to be living in hardware wise, that's for sure! Thanks AMD for continuing to be a disruptive force in the market!
Along those lines, and extrapolating further from the result above, while I don't think anyone will be rendering with dual 7742 EPYC Rome cpus and their 128 cores/256 threads around here, said system could theoretically hit parity with the RTX titan, assuming you don't run into NUMA issues with the two sockets.
I don't see a lot of Daz users lining up to buy the top end Threadrippers though. Nvidia pretty much pwns most of us around here thanks to Iray.
In any case, thanks Xionis for sharing your 32 core CPU only result. Now I'm pondering the possible config variations of my next build again...
I imagine we can get an idea of the 3990x performance from the reviews when they compare it to the 3970x. It is indeed doubtful it will be linear, but it certainly will be faster. And yeah, if you compare it to Quadro, that does alter the value proposition.
We also need to keep in mind that this is only one benchmark scene. Things may be a little different in other scenes. I'd like to see the 3970x tackle that one bench that had dforce strand hair in it that seemed to really favor RTX. Would the 3970x hold up in a scene like that? Or would it perform even better?
It would be pretty cool to see one of these chips run 3DL. I have said that if Ryzen existed in 2015 when Iray hit Daz Studio, things might have turned out very different here. Ryzen could have turned the tables for 3DL. But that is another topic.
I'd be happy to benchmark any scene you'd like as long as it not something huge that's going to take hours. I was thinking about running the benchmark again because it occured to me I usually leave my computer rendering while i'm at work or asleep using manfriday's render queue. So when I ran that benchmark it had been going non-stop for a few days and I wonder if it will be faster if I allow it to rest/cooldown for a hour first. I also kept it at stock settings, so I could tweak some settings using AMD master tool, but it's probably best to get a real world speed using the base clock settings.
I'd be curious to know what one of those would be like as well. However the 2.9Ghz speed makes me wonder how much faster it would be even with double the core and thread count i'm thinking maybe only a 20-30% speed increase before that thing starts to thermal throttle. Speaking of, i forgot to mention my threadripper is AIR cooled using the Coolermaster Wraith Ripper inside a Lian Li O11 Dynamic XL case with 10 120mm fans (Bottom and side are intake, top and rear are exhaust). After spending so much on the CPU, motherboard, ram, SSDs, video cards, and everything else I couldn't afford to do a full custom loop as well, but I may do it this summer because this thing gets HOT! I started off with a AIO liquid cooler but it didn't completely cover the cpu, so I opted for the wraith ripper which surprisingly dropped my cpu about 10-20 degrees under load compared to the AIO I was running (Asus Ryujin 360)
Had to go digging. This thread and its bench didn't get a lot of traction, but what results were posted showed that RTX really made huge gains in this particular scenario compared to non RTX GPUs. But some CPU times were really interesting. It should only take a few minutes to run. https://www.daz3d.com/forums/discussion/344451/rtx-benchmark-thread-show-me-the-power/p1
Then there is the original benchmark by sickleyield. It is old and perhaps outdated, and the info is all over the place. But it is still cool to see what new hardware can do with it as it has benchmarks going back to 2015. This one should only take a couple minutes as well. On GPU it should be less than a minute with a 2080 Super. https://www.daz3d.com/forums/discussion/53771/iray-starter-scene-post-your-benchmarks/p1
If you want to, not that you have to, but it would be cool to see these with the 3970x by itself, and then a test with the 2080 Supers.
Starting from a colder temp can effect the time, since it might prevent some throttling. With how short these benches are, you can possibly finish them before hitting the thermal max. However, I think a 'real world' scenario makes sense. We are more likely to be using the hardware a lot, so starting from a cold temp isn't much of a reality. And that advantage is pretty small anyway, after a few minutes you should be reaching max temps in normal use. In the scope of a long render, its probably not going to factor much.
Fwiw I actually have both those threads linked at the top of this one for easy access here. And yeah - I too am very curious to see how performance scales on the 3970X with that Strand hair scene (my guess is: not very well. But you never know.)
PS: All performance tables are now updated.
I ran the stand based hair "show me the power" test on only the threadripper 3970x
2020-02-02 00:45:38.548 Total Rendering Time: 1 minutes 26.67 seconds
Also I had shut down my system to allow it to cool for an hour and ran your benchmark test again to see if there would be a difference since the first time I ran it it had been running for a few days straight and it finished in 11 minutes 21.34 seconds, so about a 9 second performance gain there.
I swear, like half the posts in that show me the power thread are for 2080ti's or multiple card combos. I think we have enough 2080ti posts, folks, LOL. Kind of hard to get a reference point in a sea of 2080tis. I saw a 1080ti and it was actually slower than the 3970x, but that was for 4.11. Not sure about 4.12. I have two 1080tis, I thought I had run that test but I didn't see a post. So I'll do that later. I'm tired and the numbers are starting to blend together.
And indeed, saving 9 seconds on a 11 minute render is peanuts in the scope of things. That doesn't scale either, since that savings is only at the very start and dropped off after a couple minutes once the CPU hits its normal operating temp. It would only make a difference if you kept it cool the entire time with extreme liquid cooling. Even then, it would be a small boost overall.
It just occurred to me that the 3970x ran slower when you ran it plus the two 2080 Supers. That had to be because of temps. When you ran it alone, well, it was the only thing running. But when you ran all 3 devices, all 3 were kicking out heat, warming up that case for sure. That probably caused the 3970x to throttle some, which is why its iteration rate dropped a little. The heat may also have effected the 2080 Supers, because they ran nearly as fast when you turned OFF the 3970x. The difference was just 8 seconds. Indeed, their iterations per second was way lower when the 3970x was running with them. So that would seem to indicate some throttling. The 3970x throws out so much heat it may be hurting your performance. Did you happen to notice the GPU temps when you ran these tests? That might explain why you thought the speed was slower initially. You may have been throttling during those renders and did not realize it.
With all the performance you have, if you do not have a hardware monitor, you need to get one ASAP. One that monitors not only the temps, but can also control the fans. An aggressive fan curve may help with temps. It might get loud, but it will help you run faster and also help keep the hardware from baking. Just using a fan curve drops my temps by over 10 degrees.
System Configuration
System/Motherboard: ASRock TRX40 Creator
CPU: BRAND MODEL @ AMD 3960x @3800
GPU: BRAND MODEL @ Nvidia RTX 2080 Ti FE
GPU: BRAND MODEL @ Nvidia RTX Quadro 4000
System Memory: G.Skill Tridentz Z RGB @ 3000 (64 GB)
OS Drive: Samsung M.2 SSD
Asset Drive: Samsung M.2 SSD
Operating System: Windows 10 Pro 1909
Nvidia Drivers Version: 432
Daz Studio Version: 4.12.1.55 Beta
GPUs:
Benchmark Results
DAZ_STATS
2020-02-09 08:02:17.823 Total Rendering Time: 2 minutes 59.35 seconds
IRAY_STATS
2020-02-09 08:04:06.004 Iray [INFO] - CUDA device 0 (GeForce RTX 2080 Ti): 1237 iterations, 2.519s init, 174.192s render
2020-02-09 08:04:06.004 Iray [INFO] - CUDA device 1 (Quadro RTX 4000): 563 iterations, 2.711s init, 173.583s render
Iteration Rate (2080 Ti): 7.1
Iteration Rate (Quadro 4000): 3.2
GPU (RTX Quadro 4000 only):
Benchmark Results
DAZ_STATS
2020-02-09 11:31:51.275 Total Rendering Time: 8 minutes 49.63 seconds
IRAY_STATS
2020-02-09 11:31:58.003 Iray [INFO] - CUDA device 1 (Quadro RTX 4000): 1800 iterations, 1.929s init, 525.004s render
Iteration Rate (Quadro 4000): 3.4
CPU only:
Benchmark Results
DAZ_STATS
2020-02-09 10:18:38.647 Total Rendering Time: 14 minutes 16.52 seconds
IRAY_STATS
2020-02-09 10:18:48.317 Iray [INFO] - CPU: 1800 iterations, 1.635s init, 852.239s render
Iteration Rate (AMD 3960x): 1.91
@RayDAnt
data for the Quadro is added to my previous post.
System Configuration
System/Motherboard: Gigabyte Z370 Aorus Gaming 7
CPU: Intel i7-8700K @ stock (MCE enabled)
GPU: Nvidia Titan RTX @ stock (watercooled)
System Memory: Corsair Vengeance LPX 32GB DDR4 @ 3000Mhz
OS Drive: Samsung Pro 970 512GB NVME SSD
Asset Drive: Sandisk Extreme Portable SSD 1TB
Operating System: Windows 10 Pro 1903
Nvidia Drivers Version: 442.19
Daz Studio Version: 4.12.1.076 Beta x64
Benchmark Results: Titan RTX (TCC)
Total Rendering Time: 3 minutes 48.81 seconds
CUDA device 0 (TITAN RTX): 1800 iterations, 2.199s init, 223.914s render
Iteration Rate: (1800 / 223.914) = 8.039 iterations per second
Loading Time: ((0 + 180 + 48.81) - 223.914) = 4.896 seconds
Benchmark Results: Titan RTX (WDDM)
Total Rendering Time: 3 minutes 54.34 seconds
CUDA device 0 (TITAN RTX): 1800 iterations, 2.789s init, 228.870s render
Iteration Rate: (1800 / 228.870) = 7.865 iterations per second
Loading Time: ((0 + 180 + 54.34) - 228.870) = 5.470 seconds
Benchmark Results: Titan RTX (WDDM, used for display)
Total Rendering Time: 3 minutes 57.5 seconds
CUDA device 0 (TITAN RTX): 1800 iterations, 4.205s init, 230.028s render
Iteration Rate: (1800 / 230.028) = 7.825 iterations per second
Loading Time: ((0 + 180 + 57.5) - 230.028) = 7.472 seconds
Benchmark Results: Titan RTX (WDDM, used for display) + i7-8700K
Total Rendering Time: 3 minutes 56.33 seconds
CUDA device 0 (TITAN RTX): 1700 iterations, 2.547s init, 229.869s render
CPU: 100 iterations, 2.365s init, 231.289s render
Iteration Rate: (1800 / 231.289) = 7.783 iterations per second
Loading Time: ((0 + 180 + 56.33) - 231.289) = 5.041 second
Benchmark Results: Titan RTX (WDDM) + i7-8700K (used for display)
Total Rendering Time: 3 minutes 57.16 seconds
CUDA device 0 (TITAN RTX): 1697 iterations, 2.552s init, 231.700s render
CPU: 103 iterations, 2.346s init, 232.265s render
Iteration Rate: (1800 / 232.265) = 7.750 iterations per second
Loading Time: ((0 + 180 + 57.16) - 232.265) = 4.895 seconds
Benchmark Results: Titan RTX (TCC) + i7-8700K (used for display)
Total Rendering Time: 3 minutes 59.48 seconds
CUDA device 0 (TITAN RTX): 1695 iterations, 2.268s init, 232.776s render
CPU: 105 iterations, 2.351s init, 234.572s render
Iteration Rate: (1800 / 234.572) = 7.674 iterations per second
Loading Time: ((0 + 180 + 59.48) - 234.572) = 4.908 seconds
Benchmark Results: i7-8700K
Total Rendering Time: 1 hour 2 minutes 14.97 seconds
CPU: 1800 iterations, 2.580s init, 3729.654s render
Iteration Rate: (1800 / 3729.654) = 0.483 iterations per second
Loading Time: ((3600 + 120 + 14.97) - 3729.654) = 5.316 seconds
Benchmark Results: i7-8700K (used for display)
Total Rendering Time: 1 hour 4 minutes 47.33 seconds
CPU: 1800 iterations, 2.385s init, 3882.360s render
Iteration Rate: (1800 / 3882.360) = 0.464 iterations per second
Loading Time: ((3600 + 240 + 47.33) - 3882.360) = 4.970 seconds
System Configuration
System/Motherboard: Microsoft Surface Book 2
CPU: Intel i7-8650U @ stock
GPU: Nvidia GTX 1050 2GB @ stock
System Memory: 16GB DDR3 @ 1867Mhz
OS Drive: Samsung OEM 512GB NVME SSD
Asset Drive: Sandisk Extreme 1TB External SSD
Operating System: W10 1909
Nvidia Drivers Version: 442.19
Daz Studio Version: 4.12.1.076 Beta x64
Benchmark Results: GTX 1050 2GB
Total Rendering Time: 38 minutes 37.87 seconds
CUDA device 0 (GeForce GTX 1050): 1800 iterations, 5.870s init, 2307.063s render
Iteration Rate: (1800 / 2307.063) = 0.780 iterations per second
Loading Time: ((0 + 2280 + 41.83) - 2307.063) = 14.767 seconds