Fastest Render Engine for Daz Studio
This discussion has been closed.
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
© 2025 Daz Productions Inc. All Rights Reserved.
Comments
So, to speed up rendering in Daz Studio one need:
for 3Delight - get better CPU or maybe many of them, if you have the other licence for 3Delight engine
for iray - get multiple Nvidia GPUs and the computer that can handle them, the best you can afford.
If you are only going to render from within DS, not via RIB exporting and then the standalone, then (unless something has changed) you should even have an unlimited core licence.
Sharing some results I did with SickleYields iray starter scene - https://www.daz3d.com/forums/discussion/53771/iray-starter-scene-post-your-benchmarks/p1
Rendered with my Core i7 4770K at stock clocks, with the updated AWE Surface and AWE AreaPT for the lights. Turned off all photometric lights so everything is done fully with 3delight path tracer. Used the default 128 irradiance samples.
With 3delight, my 4770K renders the same scene at the same res in roughly the same time as iray on SickleYields machine with four GPUs. By my estimation, that's roughly equivalent to a single GTX1080/RTX2070. Doesn't take into account denoising of course.
... ⇧ This ⇧
I already dealt with this nonsense but I can again. Intentionally gimping a test to disfavor a technology, by denying that technology it's designed operating environment, shows bias against it and is utterly useless as a basis for evaluation.
This would be exactly the equivalent of saying that since an iRay testbench doesn't need much in the way of CPU the 3delight bench shouldn't either. Instead both benches would get exactly the same high end CPU's and as much RAM as was reasonable.
This is most definitely not true; making sure one render engine can always perform at it's optimal is not how you compare different renderers. If it is too big for the scene or uses settings that slow it down more than most, then that gives more accurate and varied results which are required for true comparrison. It dropping to CPU only is a valid part of any testing, however so too is when a scene fits on the card.
The modern GPU has largely evolved to become its own computer. And most of the limitations of using GPUs to render are now gone, at least for studios. You can now buy a single GPU that contains 48GB of VRAM. You can buy a small box that uses several of these GPUs to stack up to 512GB of VRAM, which totally eliminates the old argument that GPUs lack the memory for these tasks. Now many major studios are in fact moving to GPU rendering for their projects.
Somebody mentioned Disney Pixar, well Pixar is now using GPUs to render. They have had a relationship with Nvidia for a few years now, and this year they added RTX ray tracing to their RenderMan engine. This is technology that will be seen in their future movies.
So the VFX industry is moving quickly to GPU, and this should come as no surprise. GPU rendering IS faster and it is cheaper. I know that may sound crazy to you as a consumer, but $500k for a rendering box is a literal steal for a VFX studio, and it also takes up a fraction of the space and power. For the VFX industry there is actually not one single negative to using GPU rendering today.
It will take time, but eventually some of this will filter down to us consumers. The new Titan RTX has 24GB of VRAM in a single card, and while it costs $2500 it still presents a leap for consumers. Consider that 2 of these Titans with Nvlink will offer more than 40GB at a cost comparable to Xeon powered Intel machines, while still rendering far, far faster. In time lower end cards will start offering more VRAM as well.
GPUs are not just rendering, they are doing other tasks as well. You may recall Nvidia providing machines to NOA a year or so ago. GPUs are helping to predict our weather forecasts faster than ever before. Technology is starting to move towards GPU, and it is doing so for a reason. While I am excited that AMD has reignited the CPU market, the fact remains that CPU design basically stagnated for a full decade while GPUs continued to evolve. The fastest render engines will be the ones that use GPU.
Oh, and if you want the fastest rendering engine...video games. Real time rendering. The Unreal Engine is adding ray tracing and will be able to have the entire scene ray traced if you so choose, or have certain surfaces ray traced. Your choice. Some animations have already been produced with Unreal.
pwToon via 3DL is still the fastest & for most graphic comic storying telling sufficient.
For One to One 'closest realism with the real world' iRay is still faster but also better given the necessary requirement that the two comparitive renders have to match each other pixel for pixel as pixel-perfect and newer nVidia GPUs are improving on that all the time with regards to iRay.
Hmm, I think mustakettu's point is to compare only with the same hardware, which is a valid point in terms of 'pure academic' comparison. However, I don't necessarily agree with her about not running iray with GPUs when comparing it to 3delight or other renderers. That's ultimately a user's choice and likely to vary depending on budget, priorities and circumstances.
Comparing 3delight using dsDefaultMaterial/AoA/UberSurface with UE2/IBLMaster/AoA ambient to iray interactive/photoreal/Luxrender/Octane is just plain silly though. That's basically comparing direct lighting plus AO without any form of importance sampling to Metropolis light transport, coupled with features like multiple important sampling, Russian roulette etc.
I'd love to see a more up to date comparison between iray with CPU and/or GPU with AWE Surface/3delight. Helps me to focus on what I should improve next with my shader.
Kinda off topic. But I don't think that's completely true.
Threadripper 1950X is faster than a 1080Ti, consumes less power for the duration of the test and cheaper too (based on current prices). From - https://twitter.com/themikepan/status/1022669040815403008
More explanation here - http://blog.thepixelary.com/post/160210224042/cpu-vs-gpu-rendering-in-blender-cycles
Sorry, no. Lower end cards will never offer more VRAM than what's needed for their intended markets (right now, realtime 3d at 1080p or VR resolutions).
The fastest renderer will be the one that resolves variance with the least amount of samples.
...however it then becomes and "apples and oranges" situation.as 3DL does not have GPU rendering capability while Iray does.
Even with matched CPUs and memory, 3DL would still be the clear winner in the render speed category using both engines to get the best results with each. I've run tests of both on the same 8 thread CPU with the same 12 GB of memory for my comparisons as at the time I had a GPU card with insufficient VRAM (that also is no longer supported for Iray rendering). So, that is a "fair" test.as Iray can produce the same degree of render quality whether in GPU or CPU render mode and thus is not "gimped", just very slow.
..only two RTX Quadros can be linked together to pool VRAM for rendering so you are looking at 96 GB (which is still a lot). Only Tesla V100s can stack VRAM of beyond two cards (up to 8 for a total of 256 GB) and that requires an 8 way NVLink motherboard as well which are priced and geared towards datacentres and supercomputer development.
It will be a while before this technology filters down to our level.and budget range so not holding my breath.
...
...once I'm up and running again I'll be pitting 3DL with aweSurface against Iray in CPU mode in another set of comparison tests. Found a backup file of the bus stop scene I made, so I won't have to recreate it from scratch again, just need to convert marterials for the Iray test.
You are misinformed. Any of the 20xx cards with an Nvlink connector can pool VRAM. Also older Quadros with the connector can. It is not limited to the RTX Quadros.
You also continue to insist on biased testing conditions in order to achieve your desired result. That isn't how testing is done.
I think this argument over the "right" testing conditions needs to end. Anyone wishing to know which engine is faster needs to include sufficient conditions to decide whether GPUs are relevant or not, the answer will be system- and circumstance-specific.
...as to linking RTX cards.for (GForce, Titan, or Quadro) the link only supports stacking between a single pair of cards, unlike the former SLI which could link up to four cards for improved frame rate. The only way to link more cards together for stacking memory is to have a motherboard which has NVLink slots along with cards that have an NVLink interface with the board.
The only Quadros other than the new RTX ones that can take advantage of memory stacking under NVLink are the Volta cards (which require two bridges between a pair of cards) which are ridiculously expensive, even for the Quadro line. Pascal and older Quadros are still limited to the older SLI link which does not support memory stacking.
For supporting memory stacking with two 20xx cards, as was stated some time ago, it must be implemented by the individual software dvelopers, this is what Otoy and others have been been working on.
Of you mean that applications have to be configured to send some command to the driver to pool VRAM sure but apps like vRay already did that and were able to make it happen as soon as they got 20xx cards. Since it is a driver level thing I'm sure it's simply a matter of enabling it and rolling it out in a new patch in any render engine that wants to do so.
Sorry, it's not as simple as turning a switch in the driver. https://www.chaosgroup.com/blog/profiling-the-nvidia-rtx-cards
"For NVLink to work on Windows, GeForce RTX cards must be put in SLI mode from the NVIDIA control panel (this is not required for Quadro RTX cards, nor is it needed on Linux, and it’s not recommended for older GPUs). If the SLI mode is disabled, NVLink will not be active. This means that the motherboard must support SLI, otherwise you will not be able to use NVLink with GeForce cards. Also note that in an SLI group, only monitors connected to the primary GPU will work. Additionally, if two GeForce GPUs are linked in SLI mode, at least one of them must have a monitor attached (or a dummy plug) so that Windows can recognize them (this is not required for Quadro RTX cards nor is it necessary on Linux)."
"The NVLink speed is also different between the RTX 2080 and the RTX 2080 Ti cards, so we expect different performance hits from using NVLink."
Comparing 3DL to Iray is pretty pointless unless you can ensure the exact same features and shader complexity, which is impossible. I have to assume Iray as a primarily GPU renderer is not optimized for CPU in the first place, so even if everything else was exactly the same, you'd still have the elephant in the room that Iray is optimized to run on GPU.
Hell, comparing render speeds in renderers that are optimized for the same hardware, say Redshift vs Octane, is pretty hard when you cannot have the same features in both. You can make a subjective statement that you were able to render a scene faster in Redshift than Octane and state that you got both renders to match as much as possible, but you still have no hope of fully matching it. One will render refraction a little differently. One may have different settings for samples and whatnot, some of which you won't even find in the other renderer. Any comparisons will always turn into one side proclaiming happily that something rendered faster in their renderer and the other noticing that it doesn't look as good or throwing in valid or invalid reasons why the test wasn't fair.
OpenGL still in the lead
That's still driver stuff. Nothing to do with the application. Certainly no render engine has ever cared about sli. As you helpfully point out it is simply flipping a driver switch, except for some issue with Windows not recognizing that two 20xx cards connected by an Nvlink should have sli enabled by default.
Blender is not exactly the most popular render engine for Hollywood, though it has been used in a few films. Meanwhile, Disney Pixar uses their own RenderMan, and Pixar should know what works best for their own render engine. And they are increasingly shifting all of their work to GPU, and so are many others. Vray was used to render the environments in Black Panther, which had over 2500 special effect shots. That's right, Wakanda was rendered with Vray!
Many major VFX vendors were at Nvidia's booth for its SIGGRAPH RTX.
https://www.reddit.com/r/nvidia/comments/964k10/during_siggraph_all_major_vfx_render_engines_are/
I'm not sure what you are talking about saying low end will not get more VRAM, like ever. History says you are totally wrong. The GPU segment has moved constantly for some time. Lets rewind a couple of generations. In 2012 the GTX 600 series launched. The top end GPU, the king of its day, was the GTX 680, and the 680 came with two options, 2GB and 4GB. The 670 also came with these two options, but the 2GB model was much more common. It actually quite difficult to find a 670 with 4GB for several years. The lesser cards in the 600 series had even less VRAM, the lower tiers had only 1GB, and some only had 512MB, just half a gig. 2012 was not that long ago. Today the 1050ti, which is a low end GPU, is faster than the 680 was, and it even has the same memory spec. The 1050ti released in late 2016, less than 5 years after the 680. The top of the line GPU in 2012 is now a low tier card, AND even its exact VRAM spec is available to low end. Less than 5 years to trickle down.
There are two things that drive spec, competition and console. Both of these are going to rise in the next couple years. AMD is already competing with the 1060 by offering 8GB in it RX 580. What better way to differentiate your cards from Nvidia than offering more memory? Next you have new consoles coming from MS and Sony, and both will probably increase memory over PS4 and Xbox One. New consoles often spell an uptick in spec for PC games as well. Again if you go back to 2012, the aging Xbox 360 and PS3 had a paltry 512MB of memory. The PS4 and Xbox One increased this amount to about 4GB VRAM depending on a few factors. And what do you know, that sounds about like what many PC games target today, and many GPUs are in that spec. So you have 2 very compelling reasons to see more VRAM in the future. I predict, in 5 years time 8GB will become more the norm in x50 and x60. You will start seeing 12GB in the x70 and x80, and the x80ti will have more. The Titan is up to 24GB now, and that is a massive gap. And again, competition could compel Nvidia to add more VRAM.
The engine that renders the most samples is going to be GPU, if the engine is properly optimized in any capacity for GPU.
Stacking memory is not an issue for the studios. They can use the Nvswitch in the DGX-2, which allows more GPUs to stack and is how the DGX-2 achieves 512GB of VRAM. Obviously this is expensive at $400K, but that is nothing for a big studio. And obviously Quadros are expensive, nobody is debating that. But things do eventually trickle down. Most here did not believe me when I said that gaming cards would get ray tracing and tensor cores...and look what happened. Some users here thought that idea was nonsense, but here we are. And most people thought that Turing would cap at 12GB, but again, the Titan RTX features 24GB, twice the VRAM of the last Titan. Give it time. In about 5 years you will see more VRAM in consumer cards. But today, right now, you can Nvlink two Titan RTX and get almost 48GB VRAM and the fastest GPUs in existence. Yes, that costs $5000, but this is a price many people in content creation will happily accept.
Here's some fun stuff, a 12 minute fully CG short was made by a few people as a passion project. What is notable is how they did this, it predates the RTX thing. They came up with a number of solutions on hardware that is now vastly outdated. The final short is rendered all with Vray, and they also used a real time capture mode, which is really cool to see in the making of videos. This also goes to show how gaming engines can be used for this purpose. Though the real time feature in this video may come faster, you get the idea. Just like how Daz Studio likes to advertise how its "character tolerances" were useful for certain movies, this ability to see your film in real time is invaluable to the industry. It is a real game changer. It would be interesting to see how it works with RTX.
https://www.chaosgroup.com/blog/four-years-210-all-cg-shots-the-film-that-changed-vfx-workflows
My day job is as IT support at a data center. We do financial stuff that doesn't get benefit from the sort of compute power graphic cards supply but a lot of my industry is moving towards doing most of their heavy lifting using GPU's. Whenever you see talk of deep learning or AI that's almost always being done on GPU's. If you look into Nvidia's most earnings projections they predict that a big chunk of their sales over the next few years will be in autonomous cars.
So what bearing does that have on whether consumer cards will get more VRAM? Simple, the demand will be there. Beyond the fact that gamers are going to want high refresh 4k which will require a larger video buffer, startups in all kinds of fields who can't easily get into the Quadro ecosystem will want consumer cards with enough VRAM to prove their ideas to investors. The big hangup at present, and imo the reason the 20xx cards don't have more VRAM than the 10xx, is the global shortage of RAM due to production capacity issues which may be in part due to price fixing, again, by the 3 companies that make RAM.
Off topic, but here goes.
No. You miss the bold out statement by Vladimir. "This means that the motherboard must support SLI, otherwise you will not be able to use NVLink with GeForce cards." Just because you have two physical x16 PCI-E slots doesn't automatically mean you can use SLI. No SLI, no NVLink, no shared memory pool. Even if the actual cards used are SLI/NVLink capable.
As far as I know, Renderman is still a CPU only renderer, Hyperion too. GPU is only used for denoising. They have a GPU accelerated lighting tool, but that's an in-house app. Renderman with GPU rendering support have only been recently announced, as is Arnold's GPU support. Vray is the only one with GPU rendering support (ever since Vray RT).
Read my statement carefully. I wrote, "Lower end cards will never offer more VRAM than what's needed for their intended markets." The intended markets for low end cards are not HPC compute, machine learning or even offline CG rendering (where it makes sense to have lots of RAM to be used with large datasets). They are for games and games hardly uses more than 4 GB for 1080p right now. Will that change? Most likely when 4K displays is the norm, but low end will always be on the far end of the stick - make them as cheap as you can get away with.
I'd say you're missing the whole point of rendering. Read up Matt Pharr's presentation - Real-Time rendering’s next frontier: Adopting lessons from offline ray tracing to real-time ray tracing for practical pipeline - http://advances.realtimerendering.com/s2018/index.htm
The most important reason you need more samples is to cut down variance (noise). As he outlined, there are algorithms that minimizes variance using the same amount of samples (moving from uniform, to cosine sampling, light/BRDF sampling, combining light/BRDF sampling via MIS, making extra use of BVH info etc). So, a renderer employing those variance reducing techniques don't need as much samples as a pure brute force renderer. Just because a hardware is generating more samples doesn't automatically mean it will converge faster. Look back in the early days of Luxrender, where the CPU has bidir and the GPU don't. Even with less samples per sec, the CPU based bidirectional path tracer mode converges faster than the non bidir GPU one.
You think motherboard support for sli is an issue? LoL. OK. Try to find a board not meant exclusively as a budget low end offering that doesn't. You won't. AMD requires it for their Bx50 and Xx70 AM4 boards and on the Intel side it may not be flat required but no manufacturer would put out a board aimed at the enthusiast market without that capability. So as long as you don't try to spend less than $50 for you MoBo it will have sli supportsupport barring of course you getting used server or workstation stuff neverr meant to support consumer parts in the first place.
No it's not. It's more than just a driver switch. You're the one misinformed. Read about Project Lavina for example.
Nvidia already stated that RAM pooling for consumer cards has to be done through the application and that developpers have to make it happen
There is also framework choice and lots of programmation to be done with that. Chaosgroup had to program Vray to use DXR (that explains the need of SLI). Octane has to be completely migrated to use Vulkan instead of it's current framework and for Iray its seems that Ram Pooling is only automatic for Quadro Cards with Optix framework. Consumer Cards may not be allowed to pool memory and Iray may not have the RAM pooling capability as it relies on Optix Prime Framework instead of Optix
In fact you can. The question is not about knowing which hardware they are optimized for but really to know if they are efficient or eventually how they are implemented and in which situation they may be slow
Example : 3delight can be used as a unidirectionnal multibounce pathtracer provided you have the right shaders. Iray is using Metropolis Light tranport pathtracing algorithm
So roughly, both are pathtracers (like the majority of renderers today) but with their own implementations. They could be used to render the same scenes and thus be compared
The argument of the shader is not relevant : it is possible to write equivalent shaders for both engine (and I didn't say equivalent shaders are provided inside DS). The measurement is the final look of the render as well as the rendertime
That's what was done here https://www.blenderguru.com/articles/render-engine-comparison-cycles-vs-giants for example
Now about 3delight vs Iray, I'd say that from a technical standpoint, Iray should be quicker than 3DL even in CPU mode, especially in indoor scenes. However I did a test few years back and Iray isn't that optimized really (may not be the case today with denoiser in the beta version). Then outside of the technical implementation of the renderer, there is also the technical knowledge of the one using them. Some renderer can offer some tricks to speed up rendertimes but you have to know them to optimize your scene. If comparing apple to orange isn't a problem (remember the question is : fastest render ), should I also say that 3delight without GI and just with shadowmaps or limited raytracing is very quick ?
Outside of the render engine and the user capability there is also an other factor : the engine integration into the DCC and the features available. If the renderer A is 2x quicker than renderer B but you need 4x the time to tweak materials or scene before final render, I'm not sure people would use engine A. I think that is exactely the complain of people using the Octane plugin, and also from people using Blendbot to render inside Blender Cycles.
For me the strength of Iray is it's integration inside DS. From what I read, other third party renderers, are not as feature rich and tightly integrated and that is an important point because even if there are faster renderers outside, the path to get a DS scene rendered with one of them can be long and tiedous
Are you saying that Nvidia will sell a lot of consumer GPU that will be integrated in autonomous cars, or that people doing AI development are going to so massively buy consumer GPU and integrate them in Datacenter that they will be able to influence Nvidia? Nvidia has segmented it's offering. Basically, there are cards for gaming and cards for professionnal use. The push of AI will not have any impact on Gaming market. I have seen no evidence that Nvidia encouraged the use of gaming card instead of their professionnal counterpart. Rather the opposite as they have modified their EULA to forbid the use of consumer GPU in Datacenter. What you say has no impact on consumer GPU. Nvidia has separate products for AI and deep Learning
I also fail to see why 4K gaming would shift Nvidia or AMD to make more VRAM on gaming GPU. The hardware is already there and 8GB is enough for 4K Gaming. AMD will continue to concentrate on entry-mid level market but still with max 8GB cards and Nvidia will have no incentive to change it's offering.
Back to the original question, without quality or context, I think the answer was given. Fastest render engine is OpenGL. And because there is no speficication of realistic render in the question, I'd say second fastest is hands down 3delight. Outside of Iray for GPU renders, there are plugins for Luxrender and Octane, and from the three, Octane should be the fastest
You completely misread what I wrote. Nowhere did I wrote or imply that Nvidia's consumer cards were at all involved in the autonomous market.
You also clearly don't know what a datacenter is.
Yes, consumer cards will get used by startups and tinkerers, just as they are being used now, for AI research. Just because the full toolset is unavailable doesn't mean CUDA is.
Yes, more VRAM will be needed for 4k gaming at higher refresh rates. You can claim 8Gb is enough but experience shows that to not be true. Beyond the need for more frame buffer as refresh rate increases 4k also requires non trivial textures. Anyone who uses Daz should know full well how inadequate 8Gb will be for that job. Also 4k is not the end of the resolution road. Those 5k displays are out there and theoretically so are 8k.
Hmm. Way off topic but OK. I'll play ball.
These are 'still' above $50 on Newegg
https://www.asrock.com/mb/AMD/Fatal1ty AB350 Gaming K4/index.asp, https://www.asrock.com/mb/AMD/AB350 Pro4/index.asp, https://www.asrock.com/mb/AMD/AB350M Pro4DASH/index.asp
All three are AMD B350 boards with dual physical PCI-E x16, supports Crossfire but not SLI.
https://www.asrock.com/mb/AMD/B450 Pro4/index.asp
Even B450 don't necessarily support SLI.
X470 boards with the same criteria (Crossfire support but not SLI).
https://www.msi.com/Motherboard/X470-GAMING-PRO
https://www.msi.com/Motherboard/X470-GAMING-PLUS - Hey, even Buildzoid kinda likes it.
Edit. It's very unlikely someone is going to buy 'low end' graphic cards to play at 4K, especially at higher refresh rates.
All X470 motherboards support sli. AMD requires it as part of their chipset standard. That you found one that doesn't state it on its webpage is irrelevant.
Those are not B350 boards! Those are AB350 boards. That means they're SFF boards using a SFF chipset. Of course they don't support sli. I do find surprising that anyone get a more than $50 for those.
That's what people said a few years ago about 1080p. This insistence that nothing will ever change flies in the face of decades of experience.
Hmm.
https://forum-en.msi.com/index.php?topic=305142.0
" For SLI to be supported the board model needs to be licensed by Nvidia (of course for a fee). With the slots running at x8/x8 Sli any X470 board could theoretically run SLI but to cut down costs (and make the board cheaper) Gaming Plus does not have an SLI license. It can therefor just be used for crossfire if multi gpu is what you want to do. "
It's not just those boards above. Even ASUS TUF X470 Gaming don't support SLI.
No SLI certification, no SLI support. It's as simple as that.
https://www.computerbase.de/2017-01/am4-x370-b350-a320-chipsatz-vergleich-update/
Varianten mit Unterstützung für SLI gibt es aber nicht. - Bing Translate : However, there are no variants with support for SLI.
Lol. I'm done with this. Check with AMD. They flat out require it in the spec. If some board vendor doesn't pay for the sticker is again irrelevant.