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
I never said the denoiser is faster, I said you get better results at lower convergences. And since you may have an image that looks better at a lower convergence, you may choose to stop the render sooner. So in that way the denoiser is cutting render times, and this can be quite drastic.
About the 100% performance claim of the 2080ti vs the 1080ti:
Your render time was 1 minute 10 seconds. Others are getting similar times, some even faster.
Here you see a single 1080ti render over 2 minutes 6 seconds, and TWO 1080ti's together render 1 minute 4 seconds. That is just a few seconds faster than your 2080ti by itself. There are many 1080ti marks here, and that 2 minute time is pretty common. My own 1080ti renders the scene in 2 minutes. With TWO 1080tis I can render the scene in exactly 59 seconds. Again, this very close to your own and other 2080ti marks. We are literally talking about 5-10 seconds difference at most, and this may even be within the margin of error. This is really close to a full 100% scaling here, no matter if you take the single 1080ti or the dual 1080ti vs the 2080ti. Certainly more than your claimed 75%.
But it is actually MORE than that. Considering that the 1080ti times were achieved with OptiX ON, while the 2080ti can achieve these times with Optix OFF, the difference becomes even GREATER. If you factor Optix OFF times, the performance increase jumps well beyond 100%. With a single 1080ti taking 3 minute and 34 seconds to render. In the order of fairness, it would only be logical to compare the cards on equal ground. So I was actually being very conservative saying it was 100%.
Optix Prime is not Optix. They have similarities, but ultimately are not the same, and that should be obvious by now. Prime is a simplified Optix, stripped down for speed. And as you can see, that speed difference is pretty big with a 1080ti shaving over a minute off the render time of this small scene. If you used a large scene, that time savings could turn into hours saved. Prime operates in a different way, and way the memory is treated is totally different. Prime also must be totally recompiled every time a new GPU is launched in order to support it. The full Optix does not need to be updated every time a new arch is released (other than supporting the new features like RT and Tensor). As I said before we all got very lucky that Turing just happens to be similar enough to Volta to work with Daz Iray. Otherwise you would not even be able to start Daz, just like what happened when Pascal launched. It took several months for the beta at that time to even support Pascal, so for a while any Daz user who bought Pascal on day one was unable to use Daz at all unless they had kept their old GPUs. This was a really ugly time, as the situation was out of Daz's control. But this time was different. The 4.11 beta was out a long time before Turing, there is no way that they had implemented Turing into this Iray SDK and Turing did not even have a name at that time. You will not find any mention of this future GPU in the release notes for Iray 2017, the only GPU mentioned was Volta. Of course there is only one Volta for consumers, the Titan V.
Another huge difference between Optix and Prime is that Prime has a CPU fallback mode. The full Optix does NOT have CPU, is a pure GPU renderer. And on this fact alone, that proves that Daz Iray does not have the full Optix, even when you turn the Optix Prime off. Otherwise people who have no GPU at all would not be able to render anything with Optix off.
Your "proof" link does not mention Prime at all. I know Optix is getting updated for RTX. Prime is a different story and Nvidia's own confirm this.
"My question is whether OptiX Prime counts under the umbrella of OptiX for this purpose."
No. We have no plans to support RTX acceleration with OptiX Prime.
From https://devtalk.nvidia.com/default/topic/1039223/-optix-optix-prime-compatibility-with-cpu-and-rtx/?offset=14
This thread also demonstrates some of the fundamental differences between Optix and Optix Prime, and why Prime cannot be updated.
So there are a lot of questions to ask. Will Iray continue to offer Prime or will Prime be replaced entirely? This is a big question, because the render times are going to be greatly effected for anyone not using a RTX card, as demonstrated by the 1080ti Optix ON vs OFF.
And finally, Tensor and CUDA are nothing more than instruction sets. Tensor ultimately is a 4 by 4 by 4 matrix to perform calculations. To say that Tensor can "only" do one thing is naive. Tensor can do whatever you want it to if you write your code for it. You can even write full apps to use Tensor cores, and nothing to do with AI or denoising. There are programmable using C++ language that CUDA uses. That they happen to be very helpful for deep learning is a bonus and Nvidia is pushing for that, but there is no stopping a software from running entirely on Tensor instead of CUDA, or one that runs as a hybrid on both and splits tasks up.
Once I get done with benchmarking the Titan RTX (should have results to post by tomorrow) and my GTX 1050 based laptop (more as a reference for judging Turing behavior vs. Pascal than really wanting to know just how much worse it performs :( ) I'm gonna start deep-diving this thread to see if I can put something together.
Fair warning - I can see just from a quick skim that this thread is filled with partial results and compounding factors (eg. at least three independent benchmarking scenes referenced with people often not identifying which one their results were for; missing key system stats like system memory bandwidth) so I really can't say how useful the effort will be. However there should still be something useful there, and I'm thinking that with some clever Google docs integration I should be able to put a database together that both gives everyone access to viewing results AND the ability to contribute (in a much more normalized fashion) more test results to it in the future. Maybe through the use of a Google form.
I just did the old test render and my video cards didn't seem to add up correctly...
I have a i9 7900x, gigabyte x299 ud4 pro, 32gb 2400 ram, 970 Pro 512GB, a 980 ti and a 2070~
1x GTX 980 Ti
OptiX Prime Acceleration On
2 minutes and 40 seconds
1x RTX 2070
OptiX Prime Acceleration On
4 minutes 6 seconds
1x RTX 2070 & 1x GTX 980 Ti
OptiX Prime Acceleration On
1 minutes 6 seconds
The times make no sense, how can it be 4 minutes on its own but then make it 1 minute when its both of them together?
anyone has 1050 Ti OC graphics card? how long it takes to render a simple to a moderate scene with iray?
So.... I did the tests in the Beta version.... and what I don't get is why my 980 Ti now performs shitty in Beta when it performs great in live? But the 2070 performs well in Beta and shitty in live?
=============================================
1x GTX 980 Ti
OptiX Prime Acceleration On
4 minutes 4.81 seconds (2 minutes and 40 seconds in current D|S Version)
1x RTX 2070
OptiX Prime Acceleration On
2 minutes 29.1 seconds (4 minutes 6 seconds in current D|S Version)
1x RTX 2070 & 1x GTX 980 Ti
OptiX Prime Acceleration On
1 minutes 37.91 seconds (1 minutes 6 seconds in current D|S Version)
====================================
So what I see here between live and beta, is using both cards actually performs worse in the beta version.
I don't have denoiser on. And yes, windows and nvidia drivers are up to date. I am just gonna stick with live version of daz for now since it seems to be performing the best.
Interesting. My Titan RTX wouldn't even render AT ALL on the non-beta despite being properly recognized in the preferences menu hardware info section.
Are your results with the beta based on multiple passes? Because according to about two days of sporadic testing I've been doing with it on my 10XX series based laptop some renders can occasionally take significantly longer to complete than others of the same scene/render settings despite no indicators of anything different going on in the log file. And this is even with completely closing down DS between each render.
Am I the only one noticing the memory values here?
cuDNN memory 6.3 MB, total 139.0 MB
vs
cuDNN memory 2.4 MB, total 135.1 MB
Is Daz's version of cuDNN that bloated? Nearly 3x larger than the two versions you were using.
Is anyone checking the final rendered output of the "benchmarks", to see if it is actually rendering the same exact output images?
We honestly need a more intensive scene to render, with a longer render-time, as these cards get faster. Loading-time should not be that much of a factor and most things in the "real world", will be rendered for more than a few seconds, or even minutes. With the "uber shader" being used for many new models, shouldn't this be a relative common addition to the scene? As opposed to cut-back shaders that few would ever use.
Fwiw the most intensive benchmarking scene found in these threads to date is this one from Daz’s own DAZ_Rawb:
https://daz3d.com/forums/discussion/comment/3969916/#Comment_3969916
ETA: Also, don't forget that upping the iteration limit from the default 5000 will make any benchmarking scene take as long as you want. Considering it's from a beta release, I'd expect that has more to do with it being purposefully not-yet-optimized code than unintentional bloat.It's because of the OptiX Acceleration. As far as know, for the newer 2xxx cards, you have to turn it off to get better results. Iray is also different in 4.11 compared to 4.10 and renders slower for some reason that I haven't really dug into yet. I also thought you couldn't even use the 2xxx cards on the live version.
Can you do a test with OptiX turned off in both versions with all your cards?
Hey so, did you test out your Titan RTX? How does it perform vs the 1080 Ti and the 2080 Ti?
Still working on it. I've been having issues with getting the beta to install correctly (presumably because of having had the non-beta fully installed on the same os previously.) Once I get that ironed I'll post what I get.
Sylvie1998, I have a funny feeling that it takes just a wee bit more than swapping out the DLL file to switch Prime to full Optix. Just a wee bit more. If Prime is so similar to standard Optix, why bother branching the software to begin with? And again, if they are so similar, why can't they just flip a switch and plug standard Optix into Iray? We are going on 4 or 5 months now with no update.
Tensor cores can absolutely be a huge benefit for HPC applications even with a loss of precision. There is a 12 page paper investigating this topic and they determined this to be the case. Tensor is not just for AI, and indeed it is just a simple matter of coding for it. https://arxiv.org/abs/1803.04014Certain DAZ plugins like Measure Metrics really don't like it when you have multiple paths to Daz Studio installations floating around on your system. And although MM shouldn't matter for this sort of testing, benchmarking software when you know there's something funky going on with it idue to a previous installation is generally a no-go.
With that said, I've decided to just go ahead with it for now (since life is short and performance seems to be great regardless) so here's what I have for now:
Test Notes:
SickleYield's Benchmark (Iray Scene For Time Benchmarks)
Titan RTX + OptiX Prime ON: 1 minutes 5.59 seconds
Titan RTX + OptiX Prime OFF: 1 minutes 24.77 seconds
Outrider42's Benchmark (Iray Test 2018 B):
Titan RTX + OptiX Prime ON: 5 minutes 18.56 seconds *** SLOWER!
Titan RTX + OptiX Prime OFF: 5 minutes 8.17 seconds
DAZ_Rawb's Benchmark (Iray Render Test 2 - see https://direct.daz3d.com/forums/discussion/comment/3969916/#Comment_3969916):
Titan RTX + OptiX Prime ON: 4 minutes 30 seconds *** SLOWER!
Titan RTX + OptiX Prime OFF: 4 minutes 21.29 seconds
Aala's Benchmark (Iray Cornell Box 1k Iteration Limit - see https://direct.daz3d.com/forums/discussion/comment/4282266/#Comment_4282266):
Titan RTX + OptiX Prime ON: 1 minutes 33.36 seconds *** SLOWER!
Titan RTX + OptiX Prime OFF: 1 minutes 30.69 seconds
Conclusion(s):
The Titan RTX (which has the same stock cooler as the FE 2080ti) performs better than the FE 2080ti, but is about on par with the 2080ti when using better cooling (most every 2080ti benchmarked so far in this thread has an upgraded cooler design over the FE.)
More interestingly imo, enabling OptiX Prime acceleration sometimes decreases the Titan RTX's overall rendering performance (depending on scene content) rather than always increasing it as was seen on afaik all past chip generations - e.g. Pascal. Given that this only seems to be the case for the more taxing benchmarking scenes, I'm guessing this is a side-effect of the Turing architecture (specifically its cuda core/scheduling unit redesigns, since those are its only fully backwards compatible new hardware features) already being hardware optimized for raytracing purposes. Meaning that the only thing gained by enabling OptiX Prime on it with ray-tracing heavy workloads is slightly increased rendering times (since you've got software spending extra time optimizing code before sending it to hardware that can already do those same optimizations internally.)
So overall about what I expected tbh. The reason to get this over the 2080ti is because of the 24GB frame buffer - not so much increased performance. Although I'm definitely interested to see what this baby can do under water. The pcb is physically identical to the 2080ti, so putting together a watercooling solution shouldn't be too much trouble. So as soon as my kidneys grow back...
ETA: Added results for Aala's benchmark.
We really need to investigate how Iray has changed between 4.10 and 4.11. You're getting much better performance vs 10xx generation cards on 4.11, but those cards render those scenes in 4.10 only about 10-15% slower. And the images themselves also differ. Skimming through the log files on the other forum thread about DAZ Studio Beta, I couldn't pinpoint what exactly has changed to cause this.
Any ideas?
If anyone happens to use an egpu enclosure over thunderbolt (Win 10) with an RTX 2080ti or RTX Titan I would love to know the benchmarks and if it behaves well with 4.11. This is what I am planning to buy next month, not sure which card yet, for my laptop. I saw the prior benchmark done with an RTX 2080 using an egpu and believe that there will be minimal impact on performance, based on the reports from that user, using one but would still be interested in any numbers before making my purchase next month. My laptop has a P5000 (tcc mode) that I will be pairing with whichever I get but I do not see an issue with the less vram capacity of a 2080ti, if I choose to get it over the Titan RTX, as most of my scenes do not exceed 9-10gb.
Most likely just the fact that it's beta software. Forgive me those with a better handle on software development lingo, but ideally the only key distinguishing factor between beta and final release software is efficiency. The reason for this is because a complex piece of software like Daz Studio relies on lots of general purpose programming libraries/APIs to do what it does. And since these sorts of libraries almost always include far more features/functionality than what any single piece of software could use (since they're "general purpose") their use always leads to a direct negative impact on runtime performance (since the OS has more code to search through each time the piece of software asks a library/API to do something.)
Once a piece of software's code is finalized, a programmer can safely eliminate this type of performance hit by telling the software's compiler to arbitraily strip out all unused code (aka software runtime optimization). However, since beta software's code is still actively being worked on, there are no guarantees as to which bits of code can safely be omitted. So no runtime optimization.
Hence why the package size for the current official release of Daz Studio is around 303MB, whereas the current beta's is around 519MB - despite both versions sporting almost the same functionality. And also why they exhibit such different performance numbers when rendering the same scenes on identical hardware.
Hi! On January 3, I posted a test using my 750Ti. Reading some comments in this forum helped me realize I shouldn't be so eager to build a monster machine just yet. So I added a new card to this current one. I thank you and my wallet thanks you.
So I bought a GTX 1060 6GB for $250 and I'm here to post the results. I'd be here sooner, but the new card messed up my dual-booted Linux - but that was a crisis for a different forum.
My machine:
i5-2500K, 8G RAM
ASRock Extreme4 with PCI-3 2.0 x 16 slots.
No overclocking! (This is my only child. Must protect)
Here are the tests:
Optix on for all 3 tests. Everything was the default setup of the scene file.
Test 1 from my post last January 3:
750Ti rendered and ran the monitor
--> Total rendering time: 11:23
Test 2 yesterday:
750Ti did not render. Only ran the monitor
1060 rendered.
--> Total rendering time: 4:47
Test 3:
Both 750Ti and 1060 rendered.
750Ti ran the monitor.
--> Total rendering time: 3:28
For test 3, I downloaded GPU-Z to monitor GPU temps. It didn't detect my 1060. I don't know why.
While rendering, the 750Ti jumped up to 52deg. It normally hangs around 30deg. The fan stayed at 40%.
If my old 750Ti can handle the heat, the 1:19 time savings per scene would make it worth using both GPUs. Doing larger scenes or animation might raise temps higher, so I'll keep a watch.
The CPU temps barely rose above the average 35-40deg. So I think this ol' machine can handle it. And for $250, what an improvement!
A little speculation from me. The 2080 Ti has not only more and design updated Cuda Cores but also has new types of cores as part of the Cuda Core Clusters that are not on the 10XX series cards. The 2080 ti has extra cores the 1080 ti does not posess. The 2080 Ti have over 700 additional Cuda Cores: 4352 on 2080ti vs 3584 on 1080 Ti. The 2080 Ti also has 544 Tensor Cores, and 68 RT cores that are not on a 1080 Ti and are part of the same clusters as the Cuda Cores. My guess is maybe Daz Studeo Beta is using the the additional Raytrace features in the Tensor and RT cores to further boost the render process.
The problem is that Iray isn't being developed by DAZ3D, but Nvidia (or whoever they contract to work on Iray). Iray is bsaically a plug in that we get out of the box and all DAZ3D does is integrate it with DAZ Studio. As soon as you click 'render', DAZ3D has no real control over how the scene is rendered. That's why you have two log changes on the other part of the forum, one for DAZ Studio itself, and one for Iray. But each version of Iray has some new feature, some bug fixes and stuff like that, and it has changed quite a bit since its first introduction. Remember the SSS bug? Iray's bad handling of that forced skin shaders to be more on the blue side until they fixed it. And now, as it happens, Iray has changed again between 4.10 and 4.11 Beta. I'm just not sure how.
Nah, CUDA cores are the same. Specifically, they output the same values, they absolutely have to, otherwise all our apps we use would crash. They only differ internally and how fast they process information, but ultimately one plus one has to equal two in the end, no matter how you got to the result. Iray on the other hand changes constantly, and those are the changes that are affecting render time and render output. I tested out DAZ_Rawb's Benchmark with my own 1080 Ti in both 4.10 and 4.11, here are the results:
DAZ 4.10: 1977 iterations, 18.402s init, 335.040s render
DAZ4.11 Beta: 1870 iterations, 3.471s init, 429.821s render
DAZ 4.11 Beta Optix Prime OFF: 1880 iterations, 6.318s init, 475.317s render
And notice the difference in the final images too: