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 kept versions back to 4.12 because of previous issues, so I might try it. I think there are some products that require 4.21, aren't there?
Very nice.
There may be a few things that require a new feature in newer versions. The main one being any VBD based product, which added volumetrics. Genesis 9 will work in older versions in spite of claims otherwise.
Some light products might also need tweaking, since lighting is slightly different in 4.20+ as well. But this also works in reverse, there are many products made before 4.20 which in fact work better in those versions. The original Ghost Lights products will work in those versions as intended, but are totally broken after 4.20. The Euro Apartment set is a great example, it looks hideous in 4.20+, but like it is intended in 4.16.
So it depends on what items you have. Unfortuneately the Daz store page is little help, as it always lists the newest version as being what any product is made for, even if that new version broke it (like Euro Apartments or any classic ghost lights.) But in general, most products will work in 4.16. The exceptions are only going to be ones that require brand new features like VBD. I can't really think of something besides VBD that was introduced in 4.20.
There may also be certain plugins that may work only is specific versions of Daz. Again, this just depends on what plugins you use, if any. Hopefully you backed some of them up, too.
You can always use the Daz Beta if you want the current program. I recommend this to everyone. Keep the main Daz Studio at 4.16.0.3 (or slightly older) and keep the Beta up to date. They install to different locations and do not overlap. You will need to move plugins to the proper folder for the beta, because again, they are totally separate. This way if you really need to use a new product or feature that requires 4.21, you can do so with the beta, and you can render with speed in the primary version. I use the beta as the up to date version because the beta updates more often, it gets new features before the general release.
The 4090 will work with several versions going back, I cannot remember exactly which, but it certainly works in 4.16.0.3. This version is also the most up to date version of Daz Studio that still offers the older, faster Iray. So more things will work in this version without issue. 4.12 is too old, there are a bunch of new features that 4.12 does not have, like PBR Skin shaders. I am not sure if 4.12 even has RTX, some of the earlier ones might not, and I don't think the 4090 will work in it. I would use either 4.15 or 4.16.0.3.
System/Motherboard: SuperMicro X12
CPU: 2x Xeon Gold 6348
GPU: A6000 + RTX6000 ADA 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: 528.24
PSU: Corsair AX1600i
Some 1x CPU-only comparisons running the full 1800 iterations across four different versions of Daz.
Just using the one Xeon CPU (28 cores/56 threads), nothing else:
4.21.29 Beta
2023-02-05 20:35:00.783 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2023-02-05 20:35:00.783 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CPU: 1800 iterations, 1.831s init, 722.511s render
2023-02-05 20:35:01.562 [INFO] :: Finished Rendering
2023-02-05 20:35:01.638 [INFO] :: Total Rendering Time: 12 minutes 7.15 seconds
4.21.26 Beta
2023-02-05 17:57:30.450 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2023-02-05 17:57:30.450 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CPU: 1800 iterations, 1.813s init, 843.158s render
2023-02-05 17:57:31.196 [INFO] :: Finished Rendering
2023-02-05 17:57:31.287 [INFO] :: Total Rendering Time: 14 minutes 7.87 seconds
4.21.05 Pro (Current Release)
2023-02-05 19:47:40.841 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2023-02-05 19:47:40.842 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CPU: 1800 iterations, 1.926s init, 889.265s render
2023-02-05 19:47:41.655 [INFO] :: Finished Rendering
2023-02-05 19:47:41.735 [INFO] :: Total Rendering Time: 14 minutes 53.29 seconds
4.15.02 Pro
2023-02-05 18:40:34.290 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2023-02-05 18:40:34.290 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CPU: 1800 iterations, 1.638s init, 917.397s render
2023-02-05 18:40:35.196 Finished Rendering
2023-02-05 18:40:35.277 Total Rendering Time: 15 minutes 22.15 seconds
Summary of Results:
The bottom chart makes comparisons versus 4.15.02 and we can also say that vs. current Pro release (4.21.05) - that 4.21.29 is 23% faster on this CPU. Vesion. 4.21.26 has done better in prior tests, but you get the idea here. Moving onto the multi-ADA configuration soon.
Great - thanks for the help! Any idea of whether we're going to be seeing any improvement in the next release? I've gone back through this thread (to when the 4090 was released last October) and I didn't see anything concrete. I've run the beta tonight as well and didn't see any difference, so I'm not if that's indicative of what's to come or not.
You'd have to ask the Iray dev team that. This issue has been going since 4.20 released, well before the 4000 series launched. The Dev team claimed to fix a performance regression with 4.21, and it seams they did move the needle a tiny bit. But from my testing it is clear they did not address the full performance regression. The fact that they even admitted there was a regression is well, proof that regression has happened. But they may not even view it is a regression from their point of view.
You really should make the "Iterations per dollar" be the "cost to run the card" in power, not the MSRP, which is an irrelevant, one-time fee. If you were throwing the card away after the render, then MSRP is relevant as a cost per iteration.
A card that costs $5000 but runs at 1/100th the wattage per iteration of a card that costs only $500 MSRP, is the better deal. Yes, it is more expensive and the MSRP cost per iteration is higher, but the RUNNING cost is 1/100th. You need less power, smaller power-supplies, less CPU/RAM/MOBO hardware, etc... As opposed to needing 40 motherboards and computers to run the $500 cards to some form of equality. Which yields nearly 400x more power costs to run "per iteration".
Since rendering is a time-intensive and power-intensive operation, it makes more logical sense to use "iterations per watt", or "Watts per iteration", which we can compare with our "wattage-hour rates". Also the fact that MSRP is moot in almost every case of cards, from new to used. You would have to update the values based on "current prices", (real prices that the cards sell for), which also differs per country.
Yes, wattage price differs too, but wattage-values are the same everywhere, as are iterations per hour, which would be the "base-line".
Wattage for a card is not hard to check. You measure the wattage, at the wall, of the computer, while it is just idling with Daz3D running while a simple scene is loaded and "done rendering". (Talking about the IRAY preview, which runs until it is done doing the preview-render.) Then you actually hit render, and watch the wattage level. It normally levels-out, after a few seconds of rendering. (Unfortunately, you need a scene that takes more than a few seconds to render, which is nearly equal to the demo. Bumping the quality to level 4 or 10 would be a non-impactful solution for those tests. It isn't doing anything "different", just doing it longer. As well as setting "Resolve" to 100%, so it is not "cutting short" when it "thinks it's done-enough".) The "consumption" of the render, is the total wattage, MINUS the idle-wattage used. If you want to get technical, you could ALSO include that idle-wattage, but it is unfair in multi-card setups, like mine, where I have 4x cards running, while another would only have 1x and may have a wasteful PSU setup, like a 1200W power supply, feeding a 300w system, which operates at about 80% "inefficiency", while mine is matched to operate at 95% "efficiency", being correctly matched to the consumption. Yes, power-supplies, even gold rated, are horribly inefficient when not correctly loaded. They only get 95% efficiency if you are within about 80% of the PSU's load limits. Below that, they quickly become horribly inefficient, and near losses of 80%. (Not an issue being 80% inefficient if you are idling at 2 watts per hour, but it is an issue if you are operating at 300-500 watts, but consuming near 1000 watts to get that 300-500 watts of power to render, or play a game.
The Max wattage of the cards is MOOT, because IRAY doesn't use but a fraction of the GPU commands, unlike a game, which uses EVERY trick in the book, to get speed, for a game. (Stuff that doesn't require "photorealism perfection", which is IRAY's focus of commands, which is real limited in the GPUs.)
Anyways, at the end of the day, since "Power cost" is the only constant "fuel" and MSRP of the device, quickly becomes a fraction of the cost... It is a more realistic portrayal of "Cost per iteration". Mostly since you are paying THAT cost, for EVERY iteration, unlike an MSRP, which was already paid and the more you iterate, the less it costs, "per iteration", because you don't keep paying the MSRP each time you iterate/render again.
Looking at the power usage is always important. Aside from the costs for the device is the power needed to cool, which adds a great deal of complexity to the question you bring up. It is said that nearly 50% of a data center’s costs are due to cooling needs. In the middle of winter, this is less a factor, but in the summer, we could be talking about a great deal of money. I have always felt that companies like AMD & Nvidia should be more focused on elements of efficiency vs performance. A great post in this thread relevant to this topic was the MSI RTX 3090 Ventus 3X analysis by Rakete:
He determined an "optimal" power input (undervolted) of about 70% for that specific GPU while rendering this scene. This brings to light the concept that for each GPU, there is an optimal balance between rendering speed and input power, though this could be influenced heavily by factors like: a.) The IRAY Version, 2.) The scene complexity and 3.) The Daz settings and resolution applied to the render.
When you said, "If you were throwing the card away after the render, then MSRP is relevant as a cost per iteration." I think you misunderstand the metric in terms of cost to the consumer. People spend hundreds or thousands of dollars on a video card. Some are concerned about performance per dollar and others about overall performance with minimal consideration of cost.
With many PC components, including graphics cards, the difference is barely noticeable in electricity costs unless you are doing rendering 24 hours a day, 7 days a week, 365 days a year. Power consumption can be a factor. However, if you spend 4 hours a day, every day of the year, rendering with a particular card/gpu at 100 watts more power consumption than another card, that's $10-$30 a year of electricity costs in the U.S. and $50-$70 a year in electricity costs in some of the higher electricity priced areas of Europe
You said above, "Wattage for a card is not hard to check. You measure the wattage, at the wall, of the computer, while it is just idling with Daz3D running while a simple scene is loaded and "done rendering." However that method is incorrect. Depending on the model and manufacturer of graphics card, it can be anywhere from 15 watts to over 100 watts at idle. Also idle power consumption can vary depending on what is being displayed on the screen and background tasks. Graphics cards don't consume zero watts at idle. The "you are idling at 2 watts per hour" opinion is incorrect for most newer graphics cards used in 3D rendering. You will typically see 15-30 watts power consumption at idle on the low end with 100+ watts power consumption at idle on the high end. Using your method, You would be measuring the difference in power consumption at idle and during rendering, but not total power consumption of the card. In addition to that, power consumption of other components can increase during rendering. To accurately measure the power consumption of a graphics card, you need third party tools that measure both the PCIE slot power consumption and the power consumption through the power cables.
I also disagree with the "MSRP of the device, quickly becomes a fraction of the cost" opinion. Even with a $300 graphics card would take several years of usage in a high electricity costs area for electricity costs to be more than the cost of the graphics card usage. In many parts of the world, a $300 graphics card could take a 5-10+ years of usage before electricity costs end up costing more than the price of the card if you did rendering for 4 to 8 hours a day. For cards costing $500 to $1,000+ it could take several decades for electricity costs to make a difference over a lower power consumption card in some parts of the world unless you were renderign 24/7/365.
I understand the idea behind measuring performance per watt. However, its not as big of a factor as many people think it is outside of a few places with extremely high electricity costs or outside of environments where the cards are rendering 24/7/365. Also, accurate measurement of a graphics card's power consumption is a lot more difficult than system idle power measurement and system under load power measurement.
A fresh post from the Iray Dev Team finally addresses a couple things we have been debating for a while:
So there you have it. Confirmation of what some of have been saying. They finally admit this new version is slower, though to say ~5% is not my observation (come on guys, I have literally seen a 60% reduction before). If it was just 5% I would never have complained! So I have to say I am not too excited if 2023.0.0 is only 5% faster than what we have now. Hopefully they are low balling the number here, and at least it is something they are working on. I have to say I am surprised how much they are changing this product. Iray has never undergone such drastic change before, not even when they swapped OptiX Prime for full OptiX to use RTX. I sure hope these changes pay off.
They also finally talk about the 4000 series, and admit that YES the 4090 is not being fully utilized. I am not sure what to make of this. After all, other render engines "just worked" as Jenson Huang likes to say, and demonstrated amazing performance gains right off the bat. While it is great that Iray has some back compatibility now, I just have to wonder what Iray is doing to limit its full potential on Lovelace while other render engines do not seem to have that issue.
Still, I suppose this is kind of good news to 4000 owners. Your cards should get faster whenever this finally drops. But notice there was no time frame given, unlike the other issue.
Also, it shows to me that Iray is indeed under active development. I know that may sound like an odd thing to say, but we have to admit it went dry for a while. All the things they added were kind of small in scope. But rewriting the intersections...that is pretty serious. I don't know if some understand (and I admit I am far from the ultimate expert), but that is basically a fundamental building block of path tracing. That is total overhaul. Think of it like this, up until recently, the changes to Iray were like pimping out your car, but not really changing what was under the hood. The move to Optix might be like an engine. This is like changing the transmission. The engine was largely the same as before, just a newer model that could run on RTX fuel. But the transmission can really change the performance of the car. They need to get the gear rations right so Iray can hit its top speed. I hope they get there.
Get a 4070TI, from a 3080, worth it? I mean, for daz iray render.
From a 3080, I don't see much reason to get a 4070ti. You might not even see much speed boost if any, and only 2GB more VRAM assuming you have a 10GB 3080. It only makes sense if you use BOTH cards at the same time. If you combine a 3080 plus a 4070ti, then you can render significantly faster. But replacing a 3080 with a 4070ti doesn't make sense to me.
System/Motherboard: SuperMicro X12
CPU: 2x Xeon Gold 6348
GPU: A6000 + 2x RTX6000 ADA Generation +1100 mhz
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: 528.24
PSU: Corsair AX1600i
Daz Version: 4.21.26 Beta
2nd ADA GPU arrived:
2023-02-14 20:54:00.075 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2023-02-14 20:54:00.075 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (NVIDIA RTX 6000 Ada Generation): 647 iterations, 1.558s init, 27.439s render
2023-02-14 20:54:00.075 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 2 (NVIDIA RTX A6000): 418 iterations, 1.623s init, 27.666s render
2023-02-14 20:54:00.076 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 1 (NVIDIA RTX 6000 Ada Generation): 675 iterations, 1.460s init, 28.071s render
2023-02-14 20:54:00.076 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CPU: 60 iterations, 0.734s init, 27.741s render
2023-02-14 20:54:01.109 [INFO] :: Finished Rendering
2023-02-14 20:54:01.224 [INFO] :: Total Rendering Time: 33.53 seconds
Loading Time: 33.53-28.071 = 5.459
Rendering Performance: 64.12
Onto the topic of power, I tend to run 350+ watts at the wall while idle or browsing and just over 1400 watts during renders without the CPU. With CPU & 3 GPUs it reads over 1800 at the wall, so CPU is not worth enabling at this scale. Cutting out the CPU and running at factory clocks puts this system at just over 60 iterations per second with this configuration.
With regards to the Ampere to Lovelace performance concerns here, Puget ran some tests with the pro ADA GPUs recently using other applications. Look at the OctaneBench-Ray leap here. Blender did even better, with 117% increase in performance between the A6000 and 6000 ADA generation.
Considering IRAY is an Nvidia tech, it would be great to see ADA performing at least as well as these other programs. Puget's results got me thinking though, especially in terms of the Topaz AI Denoiser. Each program is going to capitalize on the hardware differently. IRAY's Guided Sampling features, for example, have been left on the sidelines of this conversation. Running this benchmark scene with Guided Sampling on and comparing the Ampere A6000 vs ADA Generation 6000 GPU - the ADA had an absolutely massive lead. To get a better picture and create a wider sample group, I threw some kicks into RayDAnt's benchmark scene below. I ran 800 HD frames at 500 iterations each and found that on average the ADA cards were performing 85% better than Ampere was across the sample group.
Here is the 4K upscale from those frames: Daz3D Sample Group:
Just some food for thought while we wait for IRAY to get with the program.
I am guessing the sampler makes the GPU work a bit harder, and that lets it 'flex' its muscle more. It makes me think of a bottleneck in gaming (and other software, but games are an easy example). In many games, rendering at low resolutions is faster, but the CPU can hold back the GPU. This makes it harder to see differences in GPU hardware. At higher resolutions the GPU gets to stretch out as it becomes more fully GPU bound, and you can see the difference in hardware better. The bottleneck in this case being Iray itself. I have a feeling that using a variety of resource heavy features, like guided sampling and VBD, running at higher resolutions could also help show more of a difference between hardware with Iray. I am purely guessing, though.
Guided Sampling has been a big point of focus by the Iray team. Basically their goal is to make Guided Sampling what you use by default, regardless of scene. They have pretty much stated this in their blog which I post below.
The guided sampler is actually a little less precise than caustic sampling in previous versions. It may be faster, but it is literally cutting corners to achieve this. The idea is that it is still accurate enough. When I turned on the sampler in 4.16, I thought something was wrong, but then it hit me. The spheres in the scene are not subdivided that high, so the polygons are visible when magnified. So the caustics hit these objects and the jagged edges of the "sphere" are cast on the ground. In 4.21, the caustics effects are visibly smoother.
This is the full blog post from December 2021 back when Guided Sampling made its debut (yes, over 2 years ago now.)
This was all great news, but the problem came when the render engine became overall slower as a whole. The guided sampler can indeed be faster to achieve a certain noise level in a render, but this comparison was to the itself. It is very important to understand that these comparisons were all to the same version of Iray they were launching at the time. Previous versions of Iray were faster, and capable of overcoming this difference with their sheer speed.
Still, the take away here is that if you render complex scenes and are using 4.21, then check out the guided sampler. The mention of guided sampling scaling better with multiple GPUs is also curious to me, as I use multiple GPUs all the time, and frequently use both when I test things.
My TitanXP is rated at 250W and every render it uses almost 250W:
The piccie shows the max power draw for the board, everything else is the current value (RH of the bar graphs). The minimum power draw is 8.5W, the non-render draw (DAZ open but not Iray rendering) is around 11W. Unless I launch PhotoShop nothing else should be using the TitanXP; it isn't used for the display. (Chrome is nasty, I use Chrome but I believe I might haved stopped it using the GPU for adverts.)
The peaks on the board-power-draw bargraph are somewhere around 200W (so that's not the absolute max, that's something like the average of the peaks). I restarted Tech GPU-Z during a render to find the min/avg/max over a reasonable period of a render and it comes out to 134.5/185.4/197.8W That's pretty good use of the available power. Whether it is a good use of the resources is an entirely different question.
It would be interesting if a gamer out there with a TitanXP can post the same figures, particularly the min/avg/max, GPU-Z reports while running a game. Preferably both a cloud game and a local one (since I suspect the cloud games spend most of their time waiting for the internet).
Well a cloud game is just like streaming youtube, so only enough power to play a video. Playing games locally can vary wildly, even between different graphically demanding games. Assuming you are talking about Titan X Pascal, you can compare it to 1080ti numbers (I couldn't find a Titan XP that showed power draw during play). This video shows several games, and you can see the power use varies between them all. Some are right at 250 Watts, but one only hits 200, and Hitman can go over 260. The same video shows a 3080ti as comparison, so you can see the power it uses, which also varies.
https://youtu.be/JWBfildvm2k
But you can easily cut power by limiting frame and playing lower resolutions. You only hit the max with an uncapped frame rate and being fully GPU bound. If the CPU is bottlenecking, that can drop GPU power draw as it waits on the CPU.
Rendering with Iray is so different because the CPU and other components are not as much of a factor in most situations. The CPU basically directs traffic, but it isn't something that impacts rendering much.
Indeed. In a game. You always hit the max with DAZ because the render is fully GPU bound, as my figures demonstated.
Actually this is not necessarily true. Certainly Iray is 100% GPU bound, but there is one more thing. Video games can utilize the pcie bus more than Daz Iray. Games are constantly swapping data in and out of VRAM. Most modern games only have a so many seconds worth of data in VRAM at any time. That data is moving from RAM to VRAM over pcie. Iray does not do this. All of the data of your scene is loaded exactly one time, when you start a render. This why you need your scene to fit in VRAM, if it doesn't, you go to CPU purgatory mode. Video games don't do that. It hurts performance, but the GPU will still try to run the game and pull data from pcie. Some games perform much better with faster RAM than others, and that is because of all the data they are constantly chewing through. Iray doesn't benefit at all from RAM speed, unless perhaps with CPU.
This swapping of data over pcie takes energy, any time electrons move, energy is used. Pcie can handle up to 70 Watts. If you look at my chart a number of posts back, you will see how every GPU except one is running under its TDP (and that lone GPU is probably an anamoly of an overclocked model). My 3090 uses less than 300 Watts to render a given scene. I would expect your Titan to fall in line with that. Both of my 1080tis rendered Iray well under their TDPs. Did you overclock your Titan?
Hardcore gaming is actually much more brutal on pc hardware than Iray. Though Iray can run the card 100% for long periods of time, a hardcore gamer can easily exceed the power draw Iray uses, and with more frequent heating up and cooling down periods.
As a gamer myself, I haved observed this with every GPU I have owned since starting Daz 6 or 7 years ago.
Is there any amount of VRAM that helps Daz renders or helps Daz after the rendering? I've seen people elsewhere claiming that you get better denoising, psot processing, and faster saving/export if you have 4 times the system RAM compared to your VRAM. I'm not sure if that's some old Daz guidance that is no longer applicable.
I've never seen Daz use more than 24GB of RAM, even on complex multi characters/object renders on a system with 128GB of RAM.
I have several times seen DS take over 32GB's when rendering Iray with 64GB RAM, depends on the scene and how it's setup.
...And that was with my previous 8GB GPU
I'm seeing some interesting results in my benchmarks. When I run the test from here I do see a good change in time from the 4.21.05 agaisnt the Beta 4.21.1.41. But when I run one of my regular scenes I'm seeing worst times in the Beta.
With the default test scene I got from here these are my results
Beta: 4.21.1.41
023-03-02 19:17:52.713 [INFO] :: Total Rendering Time: 43.97 seconds
2023-03-02 19:18:03.406 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2023-03-02 19:18:03.407 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (NVIDIA GeForce RTX 4090): 1016 iterations, 1.357s init, 39.921s render
2023-03-02 19:18:03.407 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 1 (NVIDIA GeForce RTX 4080): 784 iterations, 1.324s init, 40.288s render
Current release 4.21.0.5
023-03-02 19:14:37.303 [INFO] :: Total Rendering Time: 51.62 seconds
2023-03-02 19:14:42.143 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2023-03-02 19:14:42.143 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (NVIDIA GeForce RTX 4090): 1025 iterations, 2.788s init, 47.010s render
2023-03-02 19:14:42.143 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 1 (NVIDIA GeForce RTX 4080): 775 iterations, 2.809s init, 47.162s render
2023-03-02 19:15:49.993 [INFO] :: Saving Layout: C:/Users/RenderMachine/AppData/Roaming/DAZ 3D/Studio4/layout.dsx
There's defintely a difference in time. But as soon as I load a custom scene with 2 Genesis3 characters, some emmiters, geoshells, etc. The scene is basically two women in mud.but I don't think I can put a sample :p I see slightly worst times in the Beta release
These are the results with my custom scene:
Beta 4.21.1.41
023-03-02 18:54:41.697 [INFO] :: Total Rendering Time: 3 minutes 33.31 seconds
2023-03-02 18:55:11.348 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2023-03-02 18:55:11.348 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (NVIDIA GeForce RTX 4090): 170 iterations, 14.806s init, 172.963s render
2023-03-02 18:55:11.348 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 1 (NVIDIA GeForce RTX 4080): 130 iterations, 16.021s init, 171.176s render
Current 4.21.0.5
023-03-02 18:40:14.093 [INFO] :: Total Rendering Time: 3 minutes 19.46 seconds
2023-03-02 18:40:14.527 [INFO] :: Loaded image: r.png
2023-03-02 18:40:48.888 [INFO] :: Saved image: C:\Users\RenderMachine\Documents\DAZ 3D\Studio\Render Library\Eva mud test 4080+4090.png
2023-03-02 18:40:49.757 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2023-03-02 18:40:49.757 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (NVIDIA GeForce RTX 4090): 172 iterations, 17.038s init, 165.178s render
2023-03-02 18:40:49.757 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 1 (NVIDIA GeForce RTX 4080): 128 iterations, 17.515s init, 164.484s render
Or at least by what I understand about calcualting the rendering times. Please consider that I've never run tests before so if I'm missing something forgive my ignorance and let me know :D
The bench is not absolute, just a general test of what it can be. It is hard to say what is different about your scene without being able to look at its makeup. It can be anything. Your characters probably have hair, the hair in the bench is very basic (and free). Your hair could be strand hair, or heavy on materials and shaders. You probably have more complex skin shaders. By emitters, are you talking about a mesh light? Mesh lights can be very expensive if they have many faces.
I would bet your scene is probably a lot more shader oriented than the benchmark. The benchmark had to be light on textures, and is more of a Cornell Box which is more about light bounces in geometry. So the new Iray may be benefitting the ray tracing cores but not so much the shaders. Also, the Iray dev team remarked that their new Iray fixes an issue that is used by our benchmark scene. Your scene may not be using this as much as the benchmark.
I don't suppose you have a DS before 4.15 or 4.16? That would be the more interesting test for your mud. I bet that 4.15/4.16 would easily beat both of those results. With only 300 iterations I don't know how much faster it may be, but I am betting it would be noticeable.
Thanks a lot for the concepts, Yes I have some mesh lights and a couple ghost lights( primitive planes) also something I forgot to say is that in my scene the resolution is very big 4000x4000 and I have post denoiser. I will try some changes on the scene to be able to show it and make new tests.
Unfortunately I updated long ago from the 4.15, it wasn't until recent that I learned about the longer render times in new releases so I have no way to see how much the difference should be in my regular scenes.
Thanks a lot for the concepts, Yes I have some mesh lights and a couple ghost lights( primitive planes) also something I forgot to say is that in my scene the resolution is very big 4000x4000 and I have post denoiser. I will try some changes on the scene to be able to show it and make new tests.
Unfortunately I updated long ago from the 4.15, it wasn't until recent that I learned about the longer render times in new releases so I have no way to see how much the difference should be in my regular scenes.
I'm a bit lost here.
I'm planning to invest in a rtx 4000: I'm hesitating between 4080 and 4090. 600€ difference. Not rendering that much but I'm modeling also.
What kind of time difference is there between 4080 and 4090? Is it 20% or 100%? The rest of my hardware is just under assembling ATM and ok to get the appropriate GPU (based on i9 13 900k/psu be quiet platinium 11 1200w).
For the difference between a 4080 and a 4090, the main difference IMO is the VRAM capacity. Is 16gb enough for you, or might you need the extra 8gb that a 4090 with 24gb offers? Because if you run out of VRAM, the speed of the GPU doesn't matter, it doesn't render at all in Iray.
That said, you can just look at the previous post for a small comparison of speed. In the benchmark, the 4090 did 1025 iterations vs the 4080's 775. In the other custom scene the 4090 did 170 vs the 4080's 130. So just do the math on that. That's just over 30% uplift over the 4080.
Then you just need to factor if that is worth the extra money to you. To me, personally the VRAM is more important than the raw speed, but that is just me. If you don't use that much VRAM, then the capacity is not as big of a deal. I have a 3090 and a 3060. There are times when the 3060 does not render in my scenes because I exceed its 12gb VRAM. But the 3060 was a (relatively) cheap card, so I accept that trade off.
Just a note for builders, keep in mind these cards are massive. Many 4080 models often uses the same cooler as the 4090 does, so the 4080 is just as huge. This is the physically largest mass market GPU ever produced, especially if you get one of the 3rd party AIB models. You need to make absolutely sure it will fit inside your case.
Thank you for the explanation. But to be clear, does it mean that the total render time is 30% faster? Because in 5 or 6 posts above I read this:
Current 4.21.0.5
023-03-02 18:40:14.093 [INFO] :: Total Rendering Time: 3 minutes 19.46 seconds
2023-03-02 18:40:14.527 [INFO] :: Loaded image: r.png
2023-03-02 18:40:48.888 [INFO] :: Saved image: C:\Users\RenderMachine\Documents\DAZ 3D\Studio\Render Library\Eva mud test 4080+4090.png
2023-03-02 18:40:49.757 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2023-03-02 18:40:49.757 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (NVIDIA GeForce RTX 4090): 172 iterations, 17.038s init, 165.178s render
2023-03-02 18:40:49.757 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 1 (NVIDIA GeForce RTX 4080): 128 iterations, 17.515s init, 164.484s render
I don't understand "x iterations". The render are the same for 4080 and 4090.
Few years ago, there were a benchmark scene and the results were simplier: total render time. I'm noob here because I've been out of this "game" for 3 yearss
In Daz Studio, an iteration refers to the process of rendering an image or animation multiple times, each time improving the quality of the image or animation. During each iteration, the software calculates the light, color, texture, and other visual elements of the scene, producing a higher quality image or animation each time until the desired level of quality is achieved. Each iteration involves a balance between the time required to calculate the visuals and the quality of the image or animation produced. Iterations can be adjusted and customized by the user to meet specific needs and preferences.
For the post above, we are talking about a 4080 vs 4090 doing the same scene (Eva mud test) requiring about the same total time to complete 300 total iterations. 300 Iterations is probably what we could consider a rough draft of the image and similar to a baseline for an animation frame, whereas 2000+ iterations would bring in much more detail and realism. Total Rendering time for touji's 300 iterations is listed at the top of this one @ 3 min and 19.46 second (199.46 seconds) to create the image, which helps us separate time lost for loading @ 199.46 (Total Rendering Time) - 165.178 (the greatest device time) = 34.282 seconds for loading time. The rest of the post helps us see how the GPUs performed in that particular system while working at the same time for about the same amount of total time.
Sometimes these are easier to visualize. From the sample data provided by touji_ayanami, it might look like this:
Doing the math, you have a 28.95 speed advantage with the 4090 in this particular scene (which we would really need to see, for scientific purposes). As others have mentioned, the VRAM component may also be something to consider. The best card here is clearly the 4090 in terms of overall performance but the 4080 may be the better value if you are willing to forgo some VRAM and speed.
Rendering is about time. The old benchmark was simple...too simple. It did not take into account the loading time, which could really skew the numbers in such a short benchmark. Plus they even allowed people to remove a couple of spheres if they wanted to, which invalidates a large chunk of the numbers. You need consistency for a proper benchmark.
The iteration rate can vary wildly between different scenes. You might make a scene that plows through iterations, and then another scene where a single iteration might take several seconds. So you cannot compare two different scenes this way.
Also, different scenes will behave differently, and the performance gap between two GPUs might increase or decrease. We can only use our scene as a general guideline. There are no websites or youtubers out there that are benchmarking Iray anymore. Certainly not across numerous generations and products. So this thread is pretty much all we got to go on.
You are spot on that loading time can skew the overall results. More recent versions of Daz seem faster in terms of loading, at least with certain configurations. Here are 3 scores from today, swapping out Daz disk images with 4.16.03 Pro, the current release 4.21.1.05 and 4.21.46 beta. Notice that 4.16 Pro was able to tip just over 107 iterations per second but both the current release and beta versions were faster in terms of total time to complete the render. The loading times here represents a range of 20-30% of the total time for the scene.
System/Motherboard: SuperMicro X12
CPU: 2x Xeon Gold 6348
GPU: 4x RTX6000 ADA 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: 528.24
Daz 4.16.03 Pro:
2023-03-06 20:53:53.876 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2023-03-06 20:53:53.876 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 2 (NVIDIA RTX 6000 Ada Generation): 468 iterations, 4.449s init, 16.817s render
2023-03-06 20:53:53.877 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 3 (NVIDIA RTX 6000 Ada Generation): 429 iterations, 4.431s init, 16.056s render
2023-03-06 20:53:53.877 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 1 (NVIDIA RTX 6000 Ada Generation): 436 iterations, 4.375s init, 15.828s render
2023-03-06 20:53:53.877 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (NVIDIA RTX 6000 Ada Generation): 467 iterations, 4.226s init, 16.738s render
2023-03-06 20:53:54.250 Finished Rendering
2023-03-06 20:53:54.320 Total Rendering Time: 24.40 seconds
Rendering Performance: 107.034 iterations per second
Loading Time: 7.583
DAZ Studio 4.21.1.05 (Current Release Version)
2023-03-06 20:59:55.347 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2023-03-06 20:59:55.347 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 2 (NVIDIA RTX 6000 Ada Generation): 469 iterations, 1.250s init, 18.633s render
2023-03-06 20:59:55.347 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 3 (NVIDIA RTX 6000 Ada Generation): 462 iterations, 1.170s init, 18.449s render
2023-03-06 20:59:55.347 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 1 (NVIDIA RTX 6000 Ada Generation): 462 iterations, 1.142s init, 18.478s render
2023-03-06 20:59:55.347 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (NVIDIA RTX 6000 Ada Generation): 407 iterations, 1.157s init, 18.522s render
2023-03-06 20:59:56.233 [INFO] :: Finished Rendering
2023-03-06 20:59:56.278 [INFO] :: Total Rendering Time: 23.29 seconds
Rendering Performance: 96.602 iterations per second
Loading Time: 4.657
Daz 4.21.46 Beta
2023-03-06 21:07:28.008 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : Device statistics:
2023-03-06 21:07:28.008 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 2 (NVIDIA RTX 6000 Ada Generation): 467 iterations, 1.302s init, 18.555s render
2023-03-06 21:07:28.009 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 3 (NVIDIA RTX 6000 Ada Generation): 461 iterations, 1.226s init, 18.462s render
2023-03-06 21:07:28.009 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 1 (NVIDIA RTX 6000 Ada Generation): 465 iterations, 1.232s init, 18.521s render
2023-03-06 21:07:28.009 Iray [INFO] - IRAY:RENDER :: 1.0 IRAY rend info : CUDA device 0 (NVIDIA RTX 6000 Ada Generation): 407 iterations, 1.289s init, 18.506s render
2023-03-06 21:07:28.863 [INFO] :: Finished Rendering
2023-03-06 21:07:28.898 [INFO] :: Total Rendering Time: 23.58 seconds
Rendering Performance: 97.009 iterations per second
Loading Time: 5.025
I wonder if a W3495X type setup would reduce these load times.