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
I've undervolted my 3090 slightly since my cooling situation is suboptimal and I actually get higher boost clocks by lowering the voltage since it results in less heat. Power draw is about 320W while rendering.
System Configuration
System/Motherboard: Asus Prime X570-Pro
CPU: AMD Ryzen 3800X @ Stock
GPU: Geforce 3090 Founder's Edition, undervolted slightly
System Memory: 2x16GB G.Skill Trident Z Neo DDR4 3600Mhz @ XMP Stock, so 3600mhz CL16
OS Drive: Samsung 950 PRO 256GB
Asset Drive: Corsair MX500 2TB
Operating System: Windows 10 Pro, Version 2004, build 19041.572
Nvidia Drivers Version: Geforce Game Ready Driver 456.71
Daz Studio Version: Daz Public Beta 4.12.2.51
Benchmark Results
2020-10-15 11:13:36.690 Finished Rendering
2020-10-15 11:13:36.714 Total Rendering Time: 2 minutes 11.75 seconds
2020-10-15 11:13:52.925 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2020-10-15 11:13:52.925 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 2.093s init, 126.927s render
Iteration Rate: 14.176
Loading Time: 4.82 seconds
Good to know! What undervolt did you use?
Question for those of you who are versed in running this benchmark with multiple cards available--if I am able to upgrade my GTX Titan X (Pascal) 12GB card to a 3090 in the next few months (my Titan gets an iteration rate around 4 iterations/sec with the benchmark which is in between a 1080Ti and Titan Xp on this list), would it make sense to keep my Titan X in my #4 PCIe slot and set it to TCC mode and run renders with both the 3090 and Titan X available for scenes requiring less than 12 GB VRAM, or would the benefits be too low due to the fact that the 3090 is 3.5x faster than the Titan? It looks like the 3090 gets 13-14 iterations/sec on this benchmark depending on the vendor, so I was wondering if I would see around a 10% boost from having both cards available (back of the envelope estimate on my part)? I did see somewhere a post of a 3090 + 2080Ti + a threadripper CPU at 25 iterations/sec, but can't find the thread to double check.
Maybe someone could run a test with 3090 + 1080Ti if they have both cards available and post?
I ran this test a couple of days ago while testing my 3090; did one along with my 1080Ti. However! The 1080Ti is currently on a X1 PCIe riser, and even by itself seems to be getting a bit lower hashrates than I might have expected, perhaps due to this. This is a temporary solution until I get a Zen3/my X570 board comes in. Also, I don't know anything about TCC mode, or how this might be useful. Of note, going by a performance/watt, adding the pascal card isn't worth it. My total system power usage at the wall is around 610 Watts with both rendering iray.
System Configuration
System/Motherboard: Asrock B450m-pro
CPU: AMD Ryzen 7 1700x @stock
GPU: NVidia GeForce RTX 3090 Founder's edition @ stock; EVGA GTX 1080Ti Gaming @ stock; on X1 riser.
System Memory: 48GB (Various; DDR4 2400)
OS Drive: ADATA SX8200NP NVME
Asset Drive: WDC WD10EZEX
Operating System: Windows 10 Pro Build 1909
Nvidia Drivers Version: 456.71
Daz Studio Version: DAZ Public Beta 4.12.2.51
Benchmark Results
2020-10-13 21:17:26.814 Finished Rendering
2020-10-13 21:17:26.849 Total Rendering Time: 1 minutes 54.42 seconds
2020-10-13 21:18:03.689 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2020-10-13 21:18:03.689 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 1 (GeForce GTX 1080 Ti): 352 iterations, 3.461s init, 107.697s render
2020-10-13 21:18:03.696 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (GeForce RTX 3090): 1448 iterations, 3.390s init, 108.079s render
Iteration Rate: 1080Ti: 3.268 iterations/s; 3090: 13.398 iterations/s; combined (cumulative): 16.666 iterations/s
Loading Time: 1080Ti: 6.723s; 3090: 6.341s
Thank you very much for this information--it looks about what I was thinking (13-14 iterations/sec from the 3090 and about 3 iterations/sec from the 1080Ti). This is about a 15-20% improvement from the 3090 alone, but at the cost of pulling an additional 200W or so on the power supply. FYI--TCC mode is an option for Titan class cards (it stands for Tesla Compute Compatible) which allows you to run an alternate driver which disables the display ports on the card, making the full VRAM available for use by IRAY. For my Titan X, using the standard WDDM driver Windows reserves around 900 MB of VRAM for the system to use, leaving 11.1 GB available for IRAY renders. TCC allows to use the full 12 GB, but at the cost of not being able to use the card for monitor display (therfore it is only useful as a secondary card in a system if you don't have a graphics adapter built in to your CPU).
No problem; maybe someone else can use their 1080ti to see if mine is being significantly impaired due to the slower pcie bus. I'll certainly revisit this once I have my new setup in a fewish weeks. Thanks for the info regarding TCC; I did notice that the 1080Ti only has 100MB reserved per Task Manager, apparently, and perhaps just a bit more per Hwinfo. I have attached some screenshots; perhaps it will serve as a nice bonus to negate the need to have TCC while running older non Titan cards. Unfortunately, having 24GB at one's disposal means that you get greedy quickly, quickly blowing past the almost 11GB limit anyway..
System Configuration
System/Motherboard: ASRock X570 Phantom Gaming ITX
CPU: Ryzen 3600X
GPU: Zotac Trinity RTX 3090
System Memory: Kingston HyperX DDR4 64GB 3200
OS Drive: Samsung 870QVO 1TB
Asset Drive: Same
Operating System: Windows 10 Pro x64 (2004)
Nvidia Drivers Version: GRD 456.71
Daz Studio Version: 4.12.2.51 BETA
Benchmark Results
2020-10-15 21:47:54.051 Finished Rendering
2020-10-15 21:47:54.075 Total Rendering Time: 2 minutes 17.12 seconds
2020-10-15 21:53:16.089 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2020-10-15 21:53:16.089 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 2.271s init, 131.299s render
Iteration Rate: 13.71 iterations per second
Loading Time: 5.82 seconds
Can someone with two cards (one of them being 3090), test something?
Since you cannot switch 3090 to TCC mode maybe changing this Windows setting in Advanced Display Settings would improve performance and allow using all VRAM on 3090?
I would test myself but I don't have a second card or a free slot at the moment.
System Configuration
System/Motherboard: ASROCK B450M-HDV R4.0
CPU: Ryzen 3 1200 @ 3.69Ghz
GPU: GTX 1660 Super @ 1530 Mhz
System Memory: 16GB HyperX @ 3200 Mhz
OS Drive: 1TB WD 7200 RPM
Asset Drive: BRAND MODEL CAPACITY/Same (if same as OS)
Operating System: Windows 10
Nvidia Drivers Version: Studio Driver 456.38
Daz Studio Version: 4.12
Optix Prime Acceleration: STATE (Daz Studio 4.12.1.086 or earlier only)
Benchmark Results
2020-10-16 03:13:05.105 Finished Rendering
2020-10-16 03:13:05.149 Total Rendering Time: 11 minutes 32.21 seconds
2020-10-16 03:13:17.899 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2020-10-16 03:13:17.899 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (GeForce GTX 1660 SUPER): 1716 iterations, 5.319s init, 676.429s render
2020-10-16 03:13:17.906 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CPU: 84 iterations, 5.034s init, 682.768s render
Iteration Rate: 2.6 iterations per second
Loading Time: 15.781 seconds
I had already made this benchmark with the same GPU, but with different hardware, in December 2019. Let's see if drivers/Daz Studio/Windows 10 improved anything.
I've tested my GPU, my CPU and my GPU+CPU, both in the stable and the public build. I haven't repeated the CPU one for the stable build because it would have taken an eternity.
RTX 2060 (Beta)
System Configuration
System/Motherboard: ASUS TUF Gaming X570-Plus Wi-Fi
CPU: Ryzen 5 3600 @ stock
GPU: RTX 2060 EVGA Xc @ stock
System Memory: Crucial Ballistix 2x16GB @ 3600MHz
OS Drive: Sabrent Rocket NVMe 250GB
Asset Drive: Sabrent Rocket Q NVMe 1TB
Operating System: Windows 10 Pro 2004
Nvidia Drivers Version: Studio Drivers 456.38
Daz Studio Version: 4.12.2.51 64-bit Public Build
Benchmark Results
DAZ_STATS
2020-10-16 10:27:33.016 Finished Rendering
2020-10-16 10:27:33.043 Total Rendering Time: 7 minutes 58.51 seconds
IRAY_STATS
2020-10-16 10:27:56.462 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2020-10-16 10:27:56.462 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (GeForce RTX 2060): 1800 iterations, 2.184s init, 473.759s render
Iteration Rate: 3.80 iterations per second
Loading Time: 4.75 seconds
RTX 2060 + Ryzen 5 3600 (Beta)
System Configuration
System/Motherboard: ASUS TUF Gaming X570-Plus Wi-Fi
CPU: Ryzen 5 3600 @ stock
GPU: RTX 2060 EVGA Xc @ stock
System Memory: Crucial Ballistix 2x16GB @ 3600MHz
OS Drive: Sabrent Rocket NVMe 250GB
Asset Drive: Sabrent Rocket Q NVMe 1TB
Operating System: Windows 10 Pro 2004
Nvidia Drivers Version: Studio Drivers 456.38
Daz Studio Version: 4.12.2.51 64-bit Public Build
Benchmark Results
DAZ_STATS
2020-10-16 10:01:15.262 Finished Rendering
2020-10-16 10:01:15.298 Total Rendering Time: 7 minutes 32.70 seconds
IRAY_STATS
2020-10-16 10:01:33.213 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2020-10-16 10:01:33.213 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (GeForce RTX 2060): 1603 iterations, 2.827s init, 446.552s render
2020-10-16 10:01:33.213 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CPU: 197 iterations, 2.308s init, 447.406s render
Iteration Rate: 4.02 iterations per second
Loading Time: 5.29 seconds
Ryzen 5 3600 (Beta)
System Configuration
System/Motherboard: ASUS TUF Gaming X570-Plus Wi-Fi
CPU: Ryzen 5 3600 @ stock
GPU: RTX 2060 EVGA Xc @ stock
System Memory: Crucial Ballistix 2x16GB @ 3600MHz
OS Drive: Sabrent Rocket NVMe 250GB
Asset Drive: Sabrent Rocket Q NVMe 1TB
Operating System: Windows 10 Pro 2004
Nvidia Drivers Version: Studio Drivers 456.38
Daz Studio Version: 4.12.2.51 64-bit Public Build
Benchmark Results
DAZ_STATS
2020-10-16 11:36:07.556 Finished Rendering
2020-10-16 11:36:07.598 Total Rendering Time: 1 hours 1 minutes 38.31 seconds
IRAY_STATS
2020-10-16 11:36:51.117 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2020-10-16 11:36:51.117 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CPU: 1800 iterations, 2.173s init, 3693.519s render
Iteration Rate: 0.49 iterations per second
Loading Time: 4.79 seconds
RTX 2060 (Stable)
System Configuration
System/Motherboard: ASUS TUF Gaming X570-Plus Wi-Fi
CPU: Ryzen 5 3600 @ stock
GPU: RTX 2060 EVGA Xc @ stock
System Memory: Crucial Ballistix 2x16GB @ 3600MHz
OS Drive: Sabrent Rocket NVMe 250GB
Asset Drive: Sabrent Rocket Q NVMe 1TB
Operating System: Windows 10 Pro 2004
Nvidia Drivers Version: Studio Drivers 456.38
Daz Studio Version: 4.12.1.118 64-bit Stable Build
Benchmark Results
DAZ_STATS
2020-10-16 12:05:45.912 Finished Rendering
2020-10-16 12:05:45.937 Total Rendering Time: 8 minutes 28.78 seconds
IRAY_STATS
2020-10-16 12:05:53.504 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2020-10-16 12:05:53.504 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (GeForce RTX 2060): 1800 iterations, 2.168s init, 503.865s render
Iteration Rate: 3.57 iterations per second
Loading Time: 4.92 seconds
RTX 2060 + Ryzen 5 3600 (Stable)
System Configuration
System/Motherboard: ASUS TUF Gaming X570-Plus Wi-Fi
CPU: Ryzen 5 3600 @ stock
GPU: RTX 2060 EVGA Xc @ stock
System Memory: Crucial Ballistix 2x16GB @ 3600MHz
OS Drive: Sabrent Rocket NVMe 250GB
Asset Drive: Sabrent Rocket Q NVMe 1TB
Operating System: Windows 10 Pro 2004
Nvidia Drivers Version: Studio Drivers 456.38
Daz Studio Version: 4.12.1.118 64-bit Stable Build
Benchmark Results
DAZ_STATS
2020-10-16 11:50:14.208 Finished Rendering
2020-10-16 11:50:14.234 Total Rendering Time: 7 minutes 47.95 seconds
IRAY_STATS
2020-10-16 11:51:56.526 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2020-10-16 11:51:56.526 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (GeForce RTX 2060): 1587 iterations, 2.729s init, 461.791s render
2020-10-16 11:51:56.526 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CPU: 213 iterations, 2.082s init, 463.088s render
Iteration Rate: 3.89 iterations per second
Loading Time: 4.86 seconds
CONCLUSIONS
In December 2019 I had a slightly worse motherboard, RAM and Asset Drive, with older Daz/Windows/drivers.
The results I had were 3.75 for the Stable build, and 3.71 for the Beta.
It's strange to see that the Stable build performance dropped a lot: from 3.75 to 3.57 (-4.8%).
On the other hand, the Beta performance increased significantly: from 3.71 to 3.80 (+2.4%).
Even stranger is that the current Beta is much faster than the current Stable build, for me: 3.57 vs 3.80 (+6.4%).
What about the RTX 3070? Its not in the Nvidia GPU Benchmark Results list (3.1).
System Configuration
System/Motherboard: ASUS ROG STRIX X570-F GAMING
CPU: AMD RYZEN 3700X @ stock (Auto OC enabled)
GPU: EVGA RTX 2070 SUPER XC GAMING @ stock
System Memory: G.SKILL RIPJAWS 4X16 GB @ 3200 Mhz
OS Drive: Adata XPG 8200 Pro 1 TB NVMe SSD
Asset Drive: Seagate Barracuda 2 TB 7200 RPM SATA HDD
Operating System: Windows 10 Enterprise 2004 19041.572
Nvidia Drivers Version: 456.38 SD
Daz Studio Version: 4.12.2.51
Benchmark Results
Total Rendering Time: 5 minutes 15.55 seconds
CUDA device 0 (GeForce RTX 2070 SUPER): 1800 iterations, 2.400s init, 309.607s render
Iteration Rate: 5,81 iterations per second
Loading Time: 5,943 seconds
Not released until10/29/2020. The scuttlebut is that Nvidia is holding off on releasing the 3070 until after the Big NAVI launch because if Big NAVI is going to actually be as big as people are saying it will be then Nvidia can drop the RTX 3070 card price to be competitive.
I'm currently undervolting to 0.887mV, down from 1.075mV on default. I just realized that my undervolting probably wasn't applied when I posed my benchmark. When I reran it now with the benchmark it's running at 250W peak when undervolted, a good 50-55W under the default. The performance difference is negligible, so definitely worthwhile if noise/heat are a concern.
System Configuration
System/Motherboard: Gigabyte Aorus Extreme TRX 40
CPU: Threadripper 3960X in Precision Boost Overdrive mode
GPU: Inno3d iChill x3 3090 (stock settings and cooler) + Galax 1-Click OC 2080 Ti (stock settings and cooler)
System Memory: 48Gb various G.Skill modules at 3200 XMP (2x 16Gb and 2x 8Gb in quad channel)
OS Drive: Samsung 960 Pro NVMe SSD 1Tb
Asset Drive: Same
Operating System: Windows 10 64bit
Nvidia Drivers Version: Studio Driver v 456.71 (Updated)
Daz Studio Version: 4.12.2.54 (64-bit) Public Build (Updated)
Optix Prime Acceleration: Nope
Benchmark Results - 3090 ONLY
2020-10-21 19:48:49.884 Finished Rendering
2020-10-21 19:48:49.925 Total Rendering Time: 2 minutes 10.6 seconds
2020-10-21 19:49:03.744 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2020-10-21 19:49:03.744 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 1.703s init, 125.812s render
Iteration Rate: 14.31 iterations per second
Loading Time: 4.79 seconds
This is reasonably faster than the first bench, and I tried to keep everything else (background processes etc) equal. Interestingly, the driver studio release notes specifically mention Daz 3d, and talk about support for 3000 series GPU coming "later this month", which is great for general release users.
I am getting a second 3090 next week and look forward to taking it for a spin.
May I ask where you are getting your second 3090 from--I'm still trying to find somewhere that isn't out of stock for my first 3090 (and I refuse to buy from scalpers)?
Yeah, no problem.
I live in Australia and supply here has been pretty dire. Luckily, I found a small retailer who managed to get hold of a reasonable number of cards from the less popular brands. The first card I bought is Inno3d, the second is a Colorful.
It seems that even those who are aware of this smaller retailer would rather wait for Asus, Gigabyte, MSI etc than accept a less popular brand.
I, too, would never buy from a scalper.
Good on you for that--I am interested to see what a benchmark with 2x 3090 would look like at this stage. I am personally looking for a 3090 with 3 8-pin power connectors since I will only be able to fit one card in my current system and I have read some technical concerns about the cards with only 2 8-pin connectors being able to handle max power draws without having stability issues (unless the user undervolts the card to counteract this). I have some projects I am working on that have very large scenes (up to 32 G3/G8 characters) that I currently have to render using composites on my Titan X GPU and I am really hoping that the 3090 with 24 GB of VRAM will greatly speed up my workflow.
For what it's worth, the Inno3d I have is a reference board with 2x 8-pin power connectors, and I have had absolutely no stability issues whatsoever during stress tests, rendering or extended gaming. I haven't undervolted mine. From what I gather, the early concerns around stability (and the various theories as to the causes) were resolved with the first driver update after release.
Iray is actually less demanding than many video games. If you run video games with uncapped frame rates, you will easily use more power and heat. Iray does not function the same way, because of how Iray does not constantly swap data in and out of VRAM like a game would. The scene loads once, and that is it, even with Nvlink.
Thus I think most dual 8 pin 3090s will be perfectly fine for Iray. They might be just a few seconds slower, but that would only be because of clock speeds. And as one user has shown, they actually undervolted their 3090 and got better performance. So I am not so sure that 3x8 pins is going to be all that great.
I'd also like to see if Nvlink works on the 3090 properly for Iray, how much VRAM does Iray's log report being available.
System Configuration
System/Motherboard: Gigabyte Aorus Extreme TRX 40
CPU: Threadripper 3960X in Precision Boost Overdrive mode
GPU: Inno3d iChill x3 3090 (stock settings and cooler) + Colorful iGame OC Advanced 3090 (stock setting and cooler)
System Memory: 48Gb various G.Skill modules at 3200 XMP (2x 16Gb and 2x 8Gb in quad channel)
OS Drive: Samsung 960 Pro NVMe SSD 1Tb
Asset Drive: Same
Operating System: Windows 10 64bit
Nvidia Drivers Version: Studio Driver v 456.71
Daz Studio Version: 4.12.2.60 (64-bit) Public Build (Updated)
Optix Prime Acceleration: Nope
Benchmark Results - 2x 3090
2020-10-29 18:07:37.524 Finished Rendering
2020-10-29 18:07:37.569 Total Rendering Time: 1 minutes 9.57 seconds
2020-10-29 18:07:45.428 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2020-10-29 18:07:45.428 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 1 (GeForce RTX 3090): 897 iterations, 2.279s init, 64.442s render
2020-10-29 18:07:45.428 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (GeForce RTX 3090): 903 iterations, 1.792s init, 64.719s render
Iteration Rate: 27.81 iterations per second
Loading Time: 4.851 seconds
I'm pretty happy with those results! The addition of the Threadripper reduced total rendering time by 1 second, so I didn't bother with the calculations as it clearly isn't worth it.
Have you tried an extended render test, to see how toasty they get?
I did a 1500 frame animation render that took 7 hours with the 3090 and 2080 Ti. The room was pretty warm in the morning but the card temps were fine. The coolers are pretty beefy and my case is completely open (Thermaltake Core P7).
Ah ok, yea that will help. Waiting patiently for my blocks to arrive from EK..
Nice! I plan to watercool mine at some point; they can be pretty loud under full load.
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) @ Stock speed and stock 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: WD Red 12 TB
Operating System: Windows 10 Pro version 2004 Build 19041.450
Nvidia Drivers Version: Version 457.09
Daz Studio Version: 4.12.2.60 Public Beta
Benchmark Results - RTX 3090 only no CPU rendering
2020-10-31 00:28:04.972 Finished Rendering
2020-10-31 00:28:05.023 Total Rendering Time: 2 minutes 17.53 seconds
2020-10-31 00:28:13.551 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2020-10-31 00:28:13.551 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (GeForce RTX 3090): 1800 iterations, 6.435s init, 128.464s render
Iteration Rate: (1800/128.464) = 14.011 iterations per second
Loading Time: ((137.53 seconds) - 128.464) = 9.066 seconds
Benchmark Results - RTX 3090 and 3950X CPU
2020-10-31 00:40:44.730 Finished Rendering
2020-10-31 00:40:44.771 Total Rendering Time: 2 minutes 11.22 seconds
2020-10-31 00:40:48.210 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2020-10-31 00:40:48.210 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (GeForce RTX 3090): 1637 iterations, 2.343s init, 126.156s render
2020-10-31 00:40:48.210 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CPU: 163 iterations, 2.007s init, 126.401s render
Iteration Rate: (1800/126.401) = 14.240 iterations per second
Loading Time: ((137.53 seconds) - 126.401) = 4.819 seconds
Too bad there's no 3070 stats here, with it being on the 104 die and all. While the VRAM would make them most likely a poorish choice for iray, it would be interesting. Could be a value king as long you use plenty of scene optimization.
It looks like using the CPU with a pair of 3090's actually increases render time. The 3090 cards I used were the EVGA GTX 3090 FTW3 Ultra cards. https://www.evga.com/products/product.aspx?pn=24G-P5-3987-KR
They have a higher base and boost clock than the Nvidia founder's edition cards and a power limit of 420 per card watts with the stock BIOS. Both cards hit 2000+ MHz during the render.
If you use PX1 or MSI Afterburner, you can increase the stock power limit to 450 watts per card. EVGA also has a BIOS for the cards which can increase the power limit to 500 watts per card. I may be able to get the render time for this benchmark under 1 minute if I use the 500 watt BIOS and the software to raise the power limit to 500 watts. I actually hit the card's 420 watt power limit during the render.
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) @ Stock speed and stock 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: 4.12.2.60 Public Beta
Benchmark Results - Two EVGA RTX 3090 FTW3 Ultra cards only no CPU rendering
2020-11-10 20:45:43.570 Finished Rendering
2020-11-10 20:45:43.619 Total Rendering Time: 1 minutes 8.35 seconds
2020-11-10 20:45:47.129 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2020-11-10 20:45:47.129 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (GeForce RTX 3090): 896 iterations, 1.863s init, 63.361s render
2020-11-10 20:45:47.129 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 1 (GeForce RTX 3090): 904 iterations, 1.812s init, 63.570s render
Iteration Rate: (1800/63.570) = 28.315 iterations per second
Loading Time: ((68.35 seconds) - 63.570) = 4.78 seconds
Benchmark Results - Two EVGA RTX 3090 FTW3 Ultra and 3950X CPU
2020-11-10 20:56:09.682 Finished Rendering
2020-11-10 20:56:09.728 Total Rendering Time: 1 minutes 11.95 seconds
2020-11-10 20:56:12.284 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (GeForce RTX 3090): 868 iterations, 2.582s init, 66.549s render
2020-11-10 20:56:12.284 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 1 (GeForce RTX 3090): 851 iterations, 2.511s init, 65.870s render
2020-11-10 20:56:12.284 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CPU: 81 iterations, 2.107s init, 66.197s render
Iteration Rate: (1800/ 66.549 ) = 27.047 iterations per second
Loading Time: ((71.95 seconds) - 66.549) = 5.401 seconds
Yes - it's a known side-effect of the way Iray manages splitting up a single rendering workload between multiple CPUs/GPUs in the same computer system. A highly technical explanation of it can be found here (see section 3.2.2: "Batch Rendering" starting on page 23.)
The gist of it is that if your CPU's raw rendering performance is less than half that of one or more of the GPUs in your system, enabling it plus one or more of those GPUs for rendering will result in overall slower rendering since the Iray work scheduler will periodically skip the CPUs contributions to the final render (due to them taking too long to accumulate) and re-assign those contributions to be completed over again by one of the faster GPUs.
Are you going to pick up a Nvlink for these?