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
System Configuration:
System/Motherboard: ASROCK Z790 PG Lightning
CPU: Intel Core i7 13700k @ stock (if left at defaults)
GPU: Nvidia RTX4080 @ stock
System Memory: Corsair Vengeance 64GB (2x32GB) DDR5 @ 5600
OS Drive: WD Black SN850x
Asset Drive: SAMSUNG 870 QVO SSD 2TB
Power Supply: EVGA SuperNova G2 850W
Operating System: Windows 11 22H2, 22621.819
Nvidia Drivers Version: Nvidia Studio Driver v527.56
Benchmark Results (DAZ Studio 4.21.0.5, x64):
Total Rendering Time: 1 minutes 55.28 seconds
CUDA device 0 (NVIDIA GeForce RTX 4080): 1800 iterations, 1.109s init, 105.933s render
Iteration Rate: 16.565 iterations per second
Loading Time: 6.619 seconds
Benchmark Results (DAZ Studio 4.21.1.26 Beta x64)
Total Rendering Time: 1 minutes 33.29 Seconds
CUDA device 0 (NVIDIA GeForce RTX 4080): 1800 iterations, 0.572s init, 91.714s render
Iteration Rate: 19.626 iterations per second
Loading Time: 1.576s
I'm seeing a pretty decent performance improvement (18.5%) on the 4080 with the 4.21.1.26 version of Studio. Having said that, the improvement I see might have more to do with Lovelace optmization than with regression fixes.
No, multi-GPU support for Iray rendering using Daz Studio exists comnpletely independent of NVLink activation or support. Although NVLink has its own uses here as well (mostly pooling texture memory between GPUs for very large scenes.)
You totally can with Iray. As long as each GPU is compatible and each GPU can fit your scene in its VRAM capacity they can all be used to render. If the scene exceeds my 3060's 12GB VRAM then it will not run, no memory pooling. But as long as it is under 12GB then the 3060 and 3090 will combine to render faster.
This is not unique to Iray. A lot of rendering engines support this, it just depends on the software. It is even possible to do this with gaming through some DirectX 12 feature, though I believe only Ashes of the Singulariy is capable of it.
Nvlink can be used in supported software to pool VRAM cards of the same GPU model, and Iray supports this. There is a caveat, only the texture data is shared. The mesh data is still mirrored on each card. Still, pooling texture data could potentially save up a lot of memory, just know you do not get a true doubling of VRAM capacity with Nvlink.
I will say that I don't think mixing RTX and GTX is worthwhile due to how different they are. It could be just me, but I was getting a lot of instability when I tried to use my 3090 and 1080ti together. This completely cleared up when I replaced the 1080ti with the 3060. That could just be my hardware, but even so, the GTX cards are so much slower that it really isn't worth it anyway. A 1080ti takes over 8 minutes to do this bench, a 3060 does it in half the time and uses a 100 Watts less, and the 3060 is one of the slower RTX cards.
This piqued my interest - just ran the benchmark and according to HWinfo the max power draw of a stock 4090 was 263w.
That's with the latest non-beta release of DS and so may increase if more of the card's features are used in a future release, I'll do a proper benchmark then but render time was 83s for this version.
Yeah, would be very curious to hear what you or anyone else with a 40XX gen card sees in terms of power draw running the benchmark on DS 4.21.1.026 Beta vs DS 4.21.1.013 Beta (since the former is the first release with Iray specifically optimized for Lovelace.)
I'll see what the latest version that I have access to is tomorrow and see what I can find out. GPU usage was 80s to high 90s, might be limited by CPU or Sys RAM speed.
On a related note I've seen GPU utilization increase over the last couple of versions. I've also seen someone with a Titan complain that they were suddenly experiancing overheating issues, which would be the case with marginal cooling and an increase in GPU use.
The non beta version is not using the new Iray yet. So if you can do the beta, that is the only way to test the new Iray. If you have never tried the Daz beta, you have to do a couple small things to access it first. I am not sure if you still need to "buy" the beta from the store (of course it is free). Then in DIM you need to check the box that shows hidden products, this is very important or the beta will not show up. Once you do that the beta will be shown, and it is clearly marked as beta.
The beta installs to a different path from the main release, so it does not override anything. In fact, everything is separate, as your preferences and window styles will need to be set up again. You may also need to point the beta to your Daz asset location so it sees them. This is good, so you never have to worry about it conflicting with your main version or causing problems.
You can also run instances of the beta and non beta at the same time if your PC can handle it.
I wonder about the power draw. 263 Watts is just so low. I found one of the very few reviews that shows power draw while rendering. It is for Octane, though. Still in this render engine the 4090 is hitting 300 Watts. So almost 40 more Watts than what you are reporting for Iray. Also, notice the 3090 is reaching 350 Watts in this same test, which is weird as well considering all the others are under their TDP.
This site is very new and oddly does not have a 4090 review. It only has a 4080 review which lists power draws for a variety of render engines, but this is the chart that shows a 4090 and other GPUs power use. You can also see how older GPUs use less than their full TDP rating for Octane. But why does the 3090 chug 355W here?
I wonder if it was an AIB 3090 with a higher power limit and not a 350 watt limit founders or reference card? My 1000+ watt/No Power Limit power limit RTX 3090 cards will draw 400-425 watts peak in a Daz render. The lower power limit 3090 cards I've tested draw less power than that in a Daz render, 370-400 watts with a 450 watt power limit.
I haven't recorded power draw in octanebench yet. I'll try that soon.
Had a chance to play with the beta.
Power draw is the figure reported by HWinfo as GPU power and matches the GPU power figure reported by the Nvidia performance overlay. Utilization in the 80s to high 90s.
Card is an Ichill frostbite 4090 with an Alphacool water block. Reference PCB locked to 450w in the bios. 3 x 8 pin adaptor. I'm sure FE cards and AIBs could pull more.
421.1.26 beta
71.960 seconds / 25.01 iterations per second
GPU power 266w
421.0.05
84.018 seconds / 21.42 iterations per second
GPU power 254w
16.75% speed increase in the beta
Edited to add - Tried some more typical renders in the beta which gave the card a chance to warm up a little as well. Hit 282w. Living on the edge now.
I'm rather sure that the Octanebench results posted above are with a 450 watt BIOS card. For a 3090 to be pulling 355 watts on average in Octanebench, it would have to be a higher power limit BIOS than what I tried. I ran Octanebench with the 430 watt BIOS option on my 3090(three position BIOS switch) and got this result with the card pulling 320-340 watts depending on what part of the test it was at:
The power draw makes me think the 3090 they uses was 100 watts higher in power limit than the reference design cards and founders edition cards.
I also ran Octanebench with one of my 3090s that has the no power limit BIOS and got this result with the card pulling 445-475 watts depending on what part of the test it was at:
It was bouncing off the voltage limit with that power draw. GPU temperatures were 38-42C the entire time. I could probably raise the score above 800 points with overclocking and adding mroe voltage.
I did some testing with DS 4.21.1.26. Unfortunately, I did a run with 4.21.0.5 before updating the beta and didn't realise the new beta needed a driver update to work. However, there does seem to be a small increase in speed on 30xx series cards, since it shaved around 11 seconds off the render time on the benchmark. Still not back to the speeds of 4.15, but an improvement.
4.21.0.5 - 517.48 driver
2023-01-01 08:02:38.848 [INFO] :: Total Rendering Time: 2 minutes 11.52 seconds
2023-01-01 08:02:54.747 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (NVIDIA GeForce RTX 3090): 1800 iterations, 1.815s init, 126.648s render
4.21.1.26 - 527.56 driver
2023-01-01 09:14:14.880 [INFO] :: Total Rendering Time: 2 minutes 0.66 seconds
2023-01-01 09:20:30.763 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (NVIDIA GeForce RTX 3090): 1800 iterations, 1.320s init, 117.296s render
4.15.02 - 527.56 driver
2023-01-01 09:25:44.023 Total Rendering Time: 1 minutes 42.62 seconds
2023-01-01 09:25:50.190 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (NVIDIA GeForce RTX 3090): 1800 iterations, 2.925s init, 96.523s render
My suprim x 4090 draws the the same amount of power (around 270w) when rendering with Dforce hair and 2 characters and simple environment, which led me to point out that current Daz3d version ( I run the lastest 4.21 btw) does note take advantage full power of 4090. Devs should take note of this and make change that fully utilize the capacity. The 4090 should draw max power (mine is 520w) during rendering.
3D rendering of static images (what Iray does) is a fundamentally different workload than gaming. Gaming involves having a constant, high-bandwidth flow of data (textures et.) running into the GPU over the PCI-E bus and being written into GPU VRAM in a time-critical fashion. This by itself is a significant source of power draw, that simply isn't applicable to an Iray style render task. Hence why Iray renders always show a significantly lower power draw on a GPU than what's seen with that same GPU when being used for gaming.
Simply put, Iray will never pull the max power a GPU is designed to deliver for gaming workloads because it isn't a gaming workload.
It's difficult to get a 4090 to draw maximum power under any workload. Gaming while using RTX raytracing features, an uncapped frame limit and no upscaling will get close as will some synthetic benchmarks. You'd need to be fully utilizing all the features of the card while not being help back by another part of the system or the built in limits in the driver and bios. Highest I've seen is 440w.
Certainly rendering uses less energy than gaming, but in this case we are talking about 187 Watt reduction from what the card is rated for. That just seems really steep.
Let's take a closer look at those Octane numbers and we will see a pattern.
1080ti: 191 Watts TDP: 250 difference: 59
2080ti: 186 W TDP: 250 difference: 64
2070: 141 W TDP: 175 difference: 34
2070 Super: 149 W TDP: 215 difference: 66
4080: 251 W TDP: 320 difference: 69
3090: 355 W (outlier) TDP: 350 difference: -5
All of these cards show less than a 75 Watt delta from their respective TDP, including the 4080 which is also a Lovelace card. The 4080 is a Founder's model so we know its exact TDP. Sadly we do not know the other models, and it does look like the 3090 has to be some highly overclocked variant, because I know my 3090 doesn't use nearly that much.
But that 75 Watts is important, because that fits right into the amount of power that PCIe can push across its bus. So that is my expectation with the 4090, that it would run around 375 Watts when rendering Iray. The real number is so far below this that it makes me wonder what is going on with it. The only other explanation is if rendering is automatically using a different power profile, one based on 350 Watts instead of 450. After all, if you add 75 to the reported 270 Watts that oddbob stated...you get an even 350 Watts! However, if this was a Lovelace design, then I would expect the 4080 to show a similar large delta, but it doesn't.
For some reason it won't kick into higher gear, card basically won't fully engage and only gets up to 51.2% TDP in GPU-Z.
Earlier today it did kick into higher gear, at which point the TRT was 1m 15.39s, but haven't been able to replicate it. (went up to 350W)
System Configuration
System/Motherboard: Gigabyte Aorus x570 Ultra
CPU: 5800x @ 4.5
GPU: RTX 4090 (inno3d iChill) @ +1499 memclock undervolted at 950mv
System Memory: 32GB DDR4 @ 3200
OS Drive: Samsung 980 pro 1TB
Asset Drive: Same
Power Supply: Corsair RMX850 Watt
Operating System: Windows 10 Pro Build 19045
Nvidia Drivers Version: 527.56
Daz Studio Version: 4.21
Benchmark Results
Total Rendering Time: 1 minutes 18.8 seconds
CUDA device 0 (NVIDIA GeForce RTX 4090): 1800 iterations, 0.770s init, 76.122s render
Iteration Rate: (1800 / 76.122s) 23.64625206904706 iterations per second
Loading Time: ((0 * 3600 + 1* 60 + 18.8) - 76.122s) 2.678 seconds
Board power braw not going above 255W
Earlier result
TRT: 1m 15.39s
1800 iterations, 0.756s init, 73.486s render
IR: 24.49446153008736
LT: 1.904s
Board power draw 350W
second run with beta build 4.21.1.26:
Benchmark Results
TRT: 1m 16.3s
IRAY: 1800 iterations, 7.845s init, 66.439s render
Iteration Rate: 27.09252095907524 iterations per second
Loading Time: 9.861 seconds
Board power draw again didn't go above 255W
(I saw RayDAnt asking earlier in the thread)
Yes, the 4090 will work in Daz 4.16. Several people have tested it. It is faster than it is in 4.20/21 as well, so I'd stick with 4.16. Indeed, I am using 4.16, though I have a 3090.
You can also use 4.21 in beta form, and leave your primary install intact. That is my recommendation to everyone. DO NOT UPDATE Daz 4.16 (or 4.15) if you are lucky enough to have it. Keep this, back it up, and keep that backup safe, back up your backup. You can download the beta and keep that version up to date, allowing you to run 4.16 and 4.21+ simultantiously, the best of both worlds (though one world is clearly significantly more desirable than the other).
There may be a few very small benefits to 4.21 that make it worth it using once in a while. The biggest is VBD support for volumetric fog and smoke. That could come in handy if you wish to use such an effect. But 4.21 manages to break some HDRIs that use certain EXR formats, and OOT hair shaders are broken as well. You can work around by removing a transmap texture, but doing so may slightly alter the look of the hair. Thus I don't consider that to be acceptible as a workaround. These issues may be down to how it reads files, and it might be possible to bypass by resaving the files. Of course even if you get these to work, you are still rendering much slower than 4.16 and have to deal with the long list of other documented issues of 4.20/21.
This is probably not the place to ask, but since you brought it up. Just how do you backup a Daz install? I have the 4.16.03 install .exe, is that what you mean, or is there a way to backup my current installation?
Ideally make a copy of the zips from the Downloads folder - both the application and the Daz-branded plug-ins. Then if you want to reinstall, work with Install Manager offline and place the zips in the Downloads folder - they should then show, after a refresh, in the Ready To Install tab if you have a different version installed.
System/Motherboard: MSI B450 Tomahawk Max II
CPU: AMD Ryzen 2600 @ stock
GPU: MSI Suprium X RTX 4090
System Memory: Crucial 64GB (2x32GB) DDR4 3600
OS Drive: Samsung M.2 970 Evo
Asset Drive: Same
Power Supply: Seasonic Prime GTX 1000 80 Plus Gold
Operating System: Windows 10 Build 19045
Nvidia Drivers Version: Nvidia Studio Driver v527.56
Daz Studio Version(s): 4.16.0.3
Benchmark Results
2023-01-09 15:10:32.706 Total Rendering Time: 1 minutes 2.30 seconds
Loading Time: 2.132 seconds
Iterations Rate Per Second: 31.09
2023-01-09 15:11:16.827 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2023-01-09 15:11:16.828 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (NVIDIA GeForce RTX 4090): 1800 iterations, 2.280s init, 57.893s render
System Configuration
System/Motherboard: MSI MPG B550 GAMING EDGE
CPU: AMD RYZEN 5800X @ STOCK
GPU: MSI GAMING X TRIO RTX 4070 Ti
System Memory: G-SKILL 32GB (4*8GB) DDR4 2600
OS Drive: WD M.2 N750
Asset Drive: SAME
Power Supply: EVGA 750 GQ 80 PLUS GLOD
Operating System: W10 BUILD 19044
Nvidia Drivers Version: NVIDIA GAME DRIVER 528.02
Daz Studio Version: 4.21.1.26
Benchmark Results
Total Rendering Time: 2 minutes 3.13 seconds
Iteration Rate: 14.61
Loading Time: 1.58 seconds
LCJD can you post the time from your help log? It looks like this:
Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (NVIDIA GeForce RTX 3090): 1800 iterations, 0.932s init, 107.757s render
The render time in seconds has your correct render time. The yellow is my render time with 4.21.1.26
This is also important because if what you are writing is correct, then I have NO DOUBT that something is wrong with Daz Iray and the 4000 series. The 4070ti is faster than the 3090 at, well, just about everything. It is 13% faster at gaming, and when it comes to rendering the 4070ti is a fair bit faster than the 3090. There are some rare games where the 3090 barely slips ahead of the 4070ti, but oh so barely.
However, your time is around 2 minutes. If this is true, then your 4070ti is not just slower than my 3090, it is 13 seconds slower, and roughly 12% slower. A double digit loss. This should not be the case. Check out how the 4070ti compares to the 3090 in this Blender Optix test (and by extension, how much faster the entire 4000 series is compared to the 3090.) The crazy part...Iray is based on OptiX.
In this bench, higher is better, as it measures ''samples per minute''.
The 4070ti is not just faster than the 3090, it is significantly faster than the 3090. When you consider that the 4070ti beats the 3090 so easily and at so many things there is no way that Iray would be the only software where the 3090 can beat the 4070ti at all. Much less by such a large gap.
These results run counter to every other generation of GPU, where the new generation saw a bigger uplift with Iray than it did for video games. Then you add the wildly reduced power draws and I think the evidence is really starting to mount that something is very, very wrong with Iray and the RTX 4000 series.
If anybody else has a 4070ti, please take the time to do this bench so we can verify if this result an outlier or the norm.
System Configuration
System/Motherboard: Gigabyte X570 AORUS MASTER
CPU: AMD RYZEN 5 3600 @ stock
GPU: Inno3D RTX 4070 Ti X3 @ stock
System Memory: Corsair Vengeance 64GB/4x16GB @ 3200
OS Drive: GIGABYTE GP-ASM2NE6500GTTD
Asset Drive: Seagate IronWolf 4TB
Operating System: Windows 10 Home 20H2
Nvidia Drivers Version: 528.02
Daz Studio Version: 4.21.0.5
Benchmark Results
2023-01-14 10:06:27.329 [INFO] :: Finished Rendering
2023-01-14 10:06:27.363 [INFO] :: Total Rendering Time: 2 minutes 30.53 seconds
2023-01-14 10:06:30.381 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2023-01-14 10:06:30.381 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (NVIDIA GeForce RTX 4070 Ti): 1800 iterations, 1.293s init, 147.115s render
Iteration Rate: 12.235 iterations per second
Loading Time: 3.415 seconds
-----------------------------------------------------------
Daz Studio Version: 4.21.1.26
Benchmark Results
2023-01-14 09:52:49.598 [INFO] :: Finished Rendering
2023-01-14 09:52:49.637 [INFO] :: Total Rendering Time: 2 minutes 10.23 seconds
2023-01-14 09:53:29.322 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2023-01-14 09:53:29.323 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (NVIDIA GeForce RTX 4070 Ti): 1800 iterations, 1.087s init, 126.981s render
Iteration Rate: 14.175 iterations per second
Loading Time: 3.249 seconds
Well, the thing is even though Daz notes say that build 4.21.1.26 uses NVIDIA Iray (363600.1657) but in reality, it uses NVIDIA Iray (363600.1268)
Two different users with 2+ minute times with the 4070ti. So that pretty much confirms it being slower than the 3090, which is totally backwards from basically all other applications. At least it has 12gb, and is much faster than a 3060.
I would be curious to test the 4070ti with other scenes against the 3090.
System/Motherboard: SuperMicro X12
CPU: 2x Xeon Gold 6348
GPU: A6000 GPU (Ampere Generation)
System Memory: 512 GB DDR4 ECC @ 3200 MHz
OS Drive: SK hynix Platinum P41 2TB PCIe NVMe Gen4
Asset Drive: 256 GB RAM DRIVE
Operating System: Win 11 Pro
Nvidia Drivers Version: 527.27 Production Branch
I did some benchmarks comparing Daz 4.15, 4.21.13 and the current Beta, 4.21.26. To get a clearer picture, I calculated each Device’s Iteration Per Second speeds for comparisons in all (3) versions of Daz. I did a few passes with each version. On each pass, only one CPU was used (28 Cores/56 Threads). Rather than fill another page with benchmark scores, here is the summary result of about a dozen passes:
I know this this discussion has shifted a bit towards ADA, and we are diving deep into the beta versions, but the Ampere issues seem mostly unaddressed:
As others have noted, perhaps we are not seeing the most current version of IRAY at work here in Beta 4.21.26?
For some reason, my memory clock does not hit the same Mhz as gaming, however when I overclock the memory, then it hits rated target speed.
RTX 4090 Stock
2023-01-16 09:38:24.092 [INFO] :: Total Rendering Time: 1 minutes 18.29 seconds
2023-01-16 09:42:05.457 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (NVIDIA GeForce RTX 4090): 1800 iterations, 0.914s init, 75.560s render
Iteration Rate: (1800 / 75.560s) 23.82 iterations per second
Loading Time: ((0 * 3600 + 1* 60 + 18.29) - 75.560s) 2.730 seconds
RTX 4090 +1750 in Afterburner
2023-01-16 09:45:23.486 [INFO] :: Total Rendering Time: 1 minutes 9.95 seconds
2023-01-16 09:48:40.086 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (NVIDIA GeForce RTX 4090): 1800 iterations, 1.069s init, 67.153s render
Iteration Rate: (1800 / 67.153s) 26.8 iterations per second
Loading Time: ((0 * 3600 + 1* 60 + 9.95) - 67.153s) 2.797 seconds
I've noticed that with newer versions of Daz, render times have increased. So I decided to test my system again with this benchmark. I think my previous best results with two RTX 3090s at stock speed was 38.53 iterations per second. My best results with a single RTX 3090 was 22.28 iterations per second. It appears that render times have increased and iterations per second have dropped.
System Configuration
System/Motherboard: MSI MEG Ace X570
CPU: AMD Ryzen 9 5950X
GPU: Two EVGA RTX 3090 Kingpins at stock settings with XOC VBIOS
System Memory: 32GB of G.Skill DDR4 Trident Z neo at 3800
OS Drive: Sabrent Rocket 4 NMVe 1TB
Asset Drive: XPG SX8100 4TB
Power Supply: BeQuite Dark Power Pro 12 1500 Watt
Operating System: Windows 10 Pro 2H22 Build 19045.2486
Nvidia Drivers Version: 528.02
Daz Studio Version: 4.21.05
2023-01-21 00:17:39.250 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend progr: Received update to 01800 iterations after 58.052s.
2023-01-21 00:17:39.254 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend progr: Maximum number of samples reached.
2023-01-21 00:17:39.756 [INFO] :: Finished Rendering
2023-01-21 00:17:39.796 [INFO] :: Total Rendering Time: 1 minutes 0.38 seconds
2023-01-21 00:17:42.093 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2023-01-21 00:17:42.093 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (NVIDIA GeForce RTX 3090): 898 iterations, 1.079s init, 57.142s render
2023-01-21 00:17:42.093 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 1 (NVIDIA GeForce RTX 3090): 902 iterations, 1.033s init, 57.121s render
Benchmark Results
Iteration Rate: (1800 / 57.142) = 31.50 iterations per second
Loading Time: ((60.38) - 57.142) = 3.238 seconds