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
My guess is that it will be a few revisions of CUDA and Iray and the other software associated with Turing before all the pieces of the puzzle are in place, and Turing is optimized to work with all the fancy new architecture. I assume that's one reason why no Iray results have been released by NVIDIA so far. If they had them, and they were blowing away all others, you can bet they'd let everyone know about them. But when you think of all the different components that have to be re-written and re-deisgned to take advantage of all that, as well as work well with older generations, it boggles my mind. And certainly not surprising that it might take a while for it all to come together.
I can't imagine that they actually planned to have the new RTX cards cost more than the older generation, but not give proportionally more performance, and more. But I guess we'll see. I'm not holding my breath. Maybe 1Q next year or something.
Rendered in 4 minutes 40 seconds on 2x 1080 (non-Ti).
Since no one has tested just one 1080Ti, I took the liberty for a baseline comparison before any raytracing stuff gets enabled in Iray... came in at 6m 40s (OptiX on). So yea, neck and neck really.. it's going to down to what gains we can get out of RTX as to whether the performance justifies the price tag.
We have a couple of 2 x 1080 and 2 x 1080Ti results, here's mine with 4 x 1080Ti's.. 1m 51s.
Edit.. I just realised those numbers are a standard 2080.. I thought it was a 2080Ti.. so there is some hope yet that they pack more punch than a 1080Ti still.
This is pretty much in line with what the reviews are saying. 2080 is equal to the 1080 Ti in terms of speed (1080 Ti has a memory advantage ofc), and the 2080 Ti is around 30% better.
Just tested the scene on my Gigabyte Aorus 1080 Ti.
OptiX on: 5m 40s
OptiX off: 6m 22s
I'm probably stating the bleeding obvious, but to get the full benifit out of these new cards you'll probably need a combination of the latest CPU, loads of onboard RAM and the right motherboard to hit the sweet spot to get everything humming along happily. Plonking a 2080ti into a 4 year old rig may not actually bring nirvana.
Thanks a lot posting the test results. I have even older computer, so unless they will make some magic improvement with iray on 2080ti,
I can wait and save money for the new computer first.
Common practice in the past was that the old graphic cards just disappeared from the market.
And for some perspective on rendering speeds, I rendered Rawb's test scene (Thanks for sharing :->) on my GT 730 4G card ( minimum spec GPU that DAZ recommended for Iray, and the best I could afford at the time ). AMD Phenom II X4-945 3.0Ghz 16G RAM; perhaps similar to PCs many Iray users have.
As you can see from the log entry, it stopped at the default 2 hour time limit before completion.
2018-09-21 04:37:41.041 Iray INFO - module:category(IRAY:RENDER): 1.0 IRAY rend info : Received update to 01591 iterations after 7203.311s.
2018-09-21 04:37:41.041 Iray INFO - module:category(IRAY:RENDER): 1.0 IRAY rend info : Maximum render time exceeded.
I let it finish to completion which totalled - at 2 hours 22 minutes 27 seconds.
Comparing that with the latest RTX 2080 giving a 6 minutes & 3 seconds performance shows considerable improvement in GPU technology over the years.
I remember waiting many, many hours for very simple scenes of chrome balls and bricks to render in the early days of ray tracing. Rendering is so much faster now.
However, a 6 minute render speed is not REAL TIME ray tracing, which is what Nvidia has been promising. It would have to be in the millisecond range; a hundred to a thousand times faster than current GPUs. And have available upgraded software to drive the new RT core hardware. Which is clearly not going to be available for some time.
Maybe next year, Now it is just a lot of marketing hype.
TLDR version:
"2080 Ti OctaneBench5 : 360-370 with RT cores on, 280 with RT cores turned off!"
- - -
Ansel Sag posted some first unofficial Octanebench results on Twitter that were shared by developers on facebook and the Otoy forum.
Those results are to be treated as "unofficial" because the benchmark software has not yet been updated to fully support all Turing card features.
However those results can be taken as a first indication of the relative performance difference between the cards before any optimization.
Sources:
https://render.otoy.com/forum/viewtopic.php?f=33&t=68810#p346322
https://www.facebook.com/groups/OctaneRender/permalink/1105771256266618/
https://twitter.com/anshelsag/status/1042485755937841153?s=21
Read the comments on facebook and twitter for further background information...
- - -
What do the scores actually mean?
"The score is calculated from the measured speed (Ms/s or mega samples per second), relative to the speed we measured for a GTX 980. The ratio is weighted by the approximate usage of the various kernels and then added up. If your score is under 100, your GPU is slower than the GTX 980 we used as reference, and if it's more your GPU is faster."
Source:
https://render.otoy.com/octanebench/results.php
- - -
When can we expect official benchmark results?
"Nothing official yet, the current OctaneBench at the moment of writing these lines does not have Turing support so any numbers we'd be giving you now would not be accurate as they'd be generated using a different version of Octane.
We are planning a v4 benchmark rather sooner than later which will let you compare."
Source:
https://render.otoy.com/forum/viewtopic.php?f=33&t=68810#p346304
- - -
This go around is a little different. The floor fell out rather suddenly on the mining craze, and there's a lot of excess inventory of the 1000 series cards. A few in the hardware review community speculated the initial price of the 2xxx cards is high to help sell off the inventory of the 1xxx cards. On top of that, there's a lot of used 1xxx cards that were being used for mining being dumped in the used market.
I think they are right, and with luck we'll see the pricing drop. Hopefully not that far out. The reception from gamers is that the RTX tech isn't worth the hit in framerates, and not enough titles support it. Also many point out the performance per dollar isn't any better.
edited and removed original post
Some people asked for preliminary results and that was the reason I tried to provide further information how to interprete the shared graph.
In hindsight it seems more reasonable to just wait for the official benchmark results that may be available when the software is updated.
Looks like the Titan V now has RTX support.
http://us.download.nvidia.com/Windows/411.63/411.63-win10-win8-win7-desktop-release-notes.pdf
I guess that those 'RT cores' were just Tensors in disguise. Marketing...
Omen
Edited to just leave the most important core information:
A card with an OB 200 renders 2x faster than a card with OB 100 because it calculates double the amount of MS/s for the same test scene during the same render time.
- - -
For those interested in understanding Octanebench please compare the values for each test scenes for a GTX 980 and a GTX 1080 Ti.
https://render.otoy.com/octanebench/summary_detail_item.php?v=3.06.2&systemID=1x+GTX+1080+Ti
https://render.otoy.com/octanebench/summary_detail_item.php?v=3.06.2&systemID=1x+GTX+980
For each test scene with different kernels the MS/s of the GTX 1080 Ti are rounded up two times the MS/s of a GTX 980.
Therefore in a simplified form you could say that the GTX 1080 Ti renders twice as fast than a GTX 980.
- - -
Rendering = the task of computing samples
Render speed = calculated samples per second. MS/s
Render time = total time taken to calculate samples
Mega is a unit prefix in metric systems of units denoting a factor of one million
1 MS (Megasample) = 1'000'000 samples
- - -
Number of samples / samples per second = Rendertime in seconds
- - -
That says everything you need to know about the Ray casting + shading relative power between cards specifically to Octane render engine. It's not a bad way of measuring. Since there are mainly two time consuming area in rendering aka ray casting and shading. If you put some simple shaders in the scene, you have your 1.76x speed up or more in render time
From the numbers I see here, I'm pretty sure the 2080 render times provided don't use Turing new features at all.
There is already a good amount of brute power added, and without any Turing feature.
Since Nvidia introduced the AI denoising feature, there is a new third metric that is also important and which performance is not measured with a just global render time
There lack of a measurement with/without denoiser and with different convergence ration. And possibly if you want to properly measure the influence of Tensor Cores, a measurement with these deactivated should be done, but I doubt that there's an easy way to get that.
For me there is room for speed up that should come when DAZ releases a build with a newer Iray
- RT Core doing the ray casting => speed up > 2 in the ray casting
- Some cuda cores freed from doing the ray casting = more cores to do shading => speed up depending on the scene shaders >2 ?
- Speed up of the denoiser with the Tensor Cores. => speed up 2-3x ( see https://developer.nvidia.com/optix-denoiser )
From the decomposition above, it can also be stated, that even if the future 2050 and 2060 don't have RT cores, they may get a speed boost from the Tensor Cores in the denoising area
I think the new RTX cards are very promising so all is left to do is...just buy it
edited to remove quote of removed post
edited to spare you from reading a long wall of text with some number examples.
Discussing the results of the posted chart risks to derail this thread.
Please, if you have questions how the OctaneBench works ask directly on the Otoy forum.
I think this is a sensible expectation based on what we know with a caveat: The Optix denoiser needs pre-training before it can be effective in Daz Studio. This means that although you could very well see a 3x speedup in denoising, it could be nullified by poor inferencing due to lack of proper training.... and the onus for the deep learning will be on Daz3D for this, not Nvidia. Each deep learning task is application specific, with the results passed to the video card as a driver update for use by the tensor cores in realtime. From what I have heard, Optix denoising is terrible right now in Daz Studio. This is likely due to Daz hasn't finished the training yet.
....this is one of my concerns, as well along with possibly needing DirectX12 (which is only supported by W10). Unless I win a lotto and can afford a brand new system with the Enterprise Edition of W10 (so I can permanently disable all the rubbish I don't need and deal with updating on my terms), ain't happening.
...this is why with my old rig, I'm pretty content with my Titan-X.
...I remember watching comparison results between between rendering at normal speed and using the "speed boost" with LuxRender and the faster renders tended to be lower in quality.
Same reason I never picked up Render Throttle (either version).
..I believe they are just talking about the driver itself. As I understand Tensor and RTX cores are different from each other just as both are different than CUDA cores. RTX cores are designed to specifically handle Ray tracing tasks. Tensor cores come into play for higher level tasks such as deep learning and AI functions.
None of the older generation cards listed as being "compatible" have Tensor cores.
...one small point. What I find a bit bothersome about benchmarks is not seeing the actual results on the screen. One card may render a scene quicker than another but how good is the quality of the end product?
Quality is the same with all cards. Different cards don't render images differently. Only the speet at which those tasks are acomplished differs.
It depends on the benchmark software you are using.
Some of them actually do feature a viewport that shows you how the test scenes look while being calculated.
compare this video example:
This benchmark software can be downloaded for free here:
https://render.otoy.com/octanebench/
extract the .zip
click on the _run_benchmark.bat
- - -
A benefit of community based benchmark software is to get a large number of results by different users on different systems. The more users upload their results the more extreme results will be averaged out in a normal distribution.
compare:
https://en.wikipedia.org/wiki/Normal_distribution
- - -
It will take some more time and patience until such results are available for the new Turing cards.
- - -
The other part of the concern has allready been answered by Aala.
- - -