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
Sorry.....
Techie OCD kicking in.
Yes jpeg compression sacrifices quality for size, but it does not loose 3D info becuase there is no such info stored in traditional image formats.
On the other hand, you do loose the "alpha channel" (the part of the image file that determines transparancy levels of any given pixel)
The way jpeg compression works would be best described as a series of "Eh, good enough" peices. (The larger the file size the closer it will look to the original image.)
PNG files on the other hand are full range RGBA (Red, Green, Blue, Alpha) images that always include 100% of the data required to recreate the original image on your screen.
PNG files are not always uncompressed. The compression used by PNG files is lossless. This means that even when compressed, 100% of the original image will be recreated on your screen (pixel for pixel prefect copy). Daz Studio outputs compressed PNG files
A few weeks ago I did a test using my dual Xeon CPUs to see how well Iray scales across multiple CPU threads.
I used the benchamrk here as a base, but set it to render up to 100 itterations (Samples)
Here are the results:
IRAY Starter Essetials Benchamrk Scene
Dell Precision T7500
2x Intel Xeon X5680 3.33GHz (Total 12 cores 24 threads)
24GB RAM (6x4GB DDR3 1333MHz 6 Channel)
Windows 10 Pro 64Bit
CPU Rendering Only, Optix Prime Disabled
Max Samples set to 100
No Preload 24 threads
51.90 Sec
The following are all preloaded (first render window still open)
--24 Threads
31.29 Sec (7.50 Sec per itteration / thread)
--20 Threads
36.80 Sec (7.36 Sec per itteration / thread)
--16 Threads
44.90 Sec (7.19 Sec per itteration / thread)
--12 Threads
59.10 Sec (7.09 Sec per itteration / thread)
--8 Threads
90.81 Sec (7.26 Sec per itteration / thread)
--4 Threads
190.45 Sec (7.16 Sec per itteration / thread)
--2 Threads
461.61 Sec (9.23 Sec per itteration / thread)
The render time reduction per core is extremely linear (not counting the 2 thread render).
The following are render times using only primary CPU cores.
Secondary Hyperthreading cores are disabled.
Enabled cores are distributed across both CPUs.
--12 Threads
40.43 Sec (4.85 Sec per itteration / thread)
--8 Threads
61.99 Sec (4.96 Sec per itteration / thread)
--6 Threads
86.63 Sec (5.20 Sec per itteration / thread)
--4 Threads
136.83 Sec (5.47 Sec per itteration / thread)
As you can see here, the per thread itteration time is quite a bit lower, but not as fast as running full Hyperthreading.
_____________________________________________________________
I propose that anyone wanting to benchmark a CPU rendering only system use this 100 Sample test.
To calculate your itteration per thread speed :
CPU threads / 100 = a
Render time in seconds = b
a x b = time each CPU thread takes to complete one itteration
Yeah, I just tried the benchmark scene with my Ryzen 7 1700, stock settings, with 8 CPUs/ 16 threads, and got almost the exact same 100 iteration time as you got for your tests....45 seconds. Though I'm not quite sure why the 100 iterations test is important. Personally I find the full scene render to be more useful, and kind of assures that everyone is using the same default settings that come with the scene.
In any case, the total render time on my system with the GTX 1070 was about 3 minutes, and with CPU only it's only at 93% after 15 minutes. And it locks up my system with all threads pegged at 100%. Which begs the question...why?
FWIW, the final render time for 16 threads CPU only using the benchmark scene was over 24 minutes and 4,700 iterations. Compared to 3 minutes if I have my GTX 1070 do it.
Yeah, I think your right. I think was actually TIF format that keeps 3D information which is why TIF's are commonly used for Normal Maps.
You have to consider this benchmark is 400 by 520 pixels in size, it is absolutely tiny. It was created over 3 years ago for Daz 4.8, not 4.9. The Genesis 2 figure predates Iray, as do all of the items in the scene. Even though they are converted to Iray, they are not truly optimized for Iray. That is because this bench was made to ensure everyone had the items needed, and that everyone could run it. Back then Daz had no Iray textures for these models that were free and included with the program. Even the base Genesis 3 is AoA by default, not Iray, because Iray only released shortly after the debut of G3. It is only now with Genesis 8 that Iray textures are included with the base model by default. What I am saying, is this test may be past its useful measure, as every new GPU is fast on it. The difference between a 1060 and a 1070, to a 1080ti is just seconds, that leaves a lot of room for error. Even just a few seconds would translate to several percentage points. Maybe it is time to create a new test, with a G8 model that was created from the ground up for Iray, including the outfit and hair. And we need a much larger image, so we can really test the true performance of modern graphics cards. I propose QHD, 2560 by 1440. Benchmarks are supposed to stress cards and push them, the current test no longer does that. The scene could be given different resolutions, which would allow those with lesser memory still test. That would also give us a good idea of how much resolution effects the render times, which is something I'd really like to see. We could go by general TV resolutions since they are numbers everybody knows.
Small: 1280 by 720 Medium: 1920 by 1080 Large: 2560 by 1440 and maybe extra large: 3840 by 2160
I just noticed that Daz doesn't yet list any G8 stuff as being included with the software. "View full list of items included in Daz Studio." Hmm...might want to update that sometime soon Daz??? G8 has been out how long now, lol? Perhaps we could use freebies from some place to dress the G8.
Every benchmark gets updated over time, they don't stagnate. That's because running new hardware on the old benchmarks all show great performance and do not properly demonstrate the performance gap at the high end. It is even possible that the software itself is bottlenecking this benchmark for high performance cards, or some other hardware spec is causing a bottleneck. For all we know, there could be a software speed limit for Iray, limiting how fast it can go rendering an image that is only 500 pixels. I believe a more demanding benchmark test would show a larger gap between the higher end machines and cards, such as going from a 1070 to 1080ti.
That is a good idea. One thing that needs to be kept in mind is that it does need to render fast enough that mid range GPUs don't take hours to render it, unless the idea is for it to benchmark GTX 1050ti and up.
I'm setting up a content folder on my computer that only has Genesis 8 Starter Essentials and see what I can come up with.
That all sounds good and reasonable. However, swordkensia verified in another thread that the results seem to agree with his results for a much larger scene. I think his took like 30 minutes or something. But I think he got the same % improvement from a GTX 1070 to a GTX 1080ti (?). Here's the link:
https://www.daz3d.com/forums/discussion/comment/2860026/#Comment_2860026
So without actual data proving that this benchmark doesn't actually reflect larger or newer or whatever scenes it's just, well, a myth.
So yes, I agree it would be awesome for somebody to propose an updated benchmark that everyone could finally agree on, and I'd be more than happy to participate. But we tend to get stuck with myths rather than finding facts.
Also, keep in mind the more detail you stress in tests the less it will apply to the entire audience. For example you mentioned G8. A lot of people don't use G8. And a lot of people don't render in higher resolutions. And so on.
I think if we're trying to attain some absolute accuracy in determining how fast our GPU's will be in all instances we'll be sorely disappointed, for the reasons you mentioned. There are tons of variables.
So IMO if we're happy with "oh, a GTX 1080ti will give something like a 40% improvement over a GTX 1070, plus or minus 5-10%", then I think we're being a lot more realistic. If we try to achieve accuracy nirvana I think we'll be very disappointed. And at the end of the day, does +/- 5 or 10% really matter to most of us?
And if you think that a newer and updated benchmark is going to make the GTX 1080ti give an 80% improvement over a GTX 1070, rather than 40%, you'll be very disappointed.
It's not about weather or not people use G8. The important part is that the surface shaders on G8 are much more advanced and designed for Iray vs the G2 surfaces.
Right now I'm cooking up something that uses G8F Starter Essentials, and Westpark (The one that comes with Daz Studio) with Iray surfaces. Doing a test render right now and it looks like my GTX 1080 ti will hit 5000 itteratioins at 1080p in about 35min.
Here's a screen capture of the in progress render. I am very much forward thinking on this scene. If the render speed jump from Pascal to Volta is similar to Maxwell to Pascal, then a GTX 1160 should complete this scene @1080p in about 40min and the GTX 1180 should do it in 10 - 15min.
Well, yes it does. If using G8 (for whatever reason) results in significant improvement in render time, but 80% of users don't benefit from that because they dont' use G8, then how does that help them? Personally, I don't use G8. All my stuff was built with G3, and I never felt the need to change it out to G8. And I'm sure many others are in the same boat. And that's just one example of the many possible scene variations that could affect render times.
But at the end of the day, does the difference really matter, or do we just believe that the G8 improvements make "some" difference without knowing exactly how much and whether it really matters?
Ok here we go :-)
Everything on stock clock, Scene is preloaded (second renderwindows is paused)
OptiX always enabled
Benching on:
-Threadripper 1950X (16 cores / 32 threads) all cores active
-4 x 1080ti Founders (Blower cards)
-32Gb DDR4 2666
CPU ONLY @ 3.4ghz/99%-100% Load:
2017-10-15 14:14:16.088 Iray VERBOSE - module:category(IRAY:RENDER): 1.0 IRAY rend progr: 94.86% of image converged
2017-10-15 14:14:16.093 Iray INFO - module:category(IRAY:RENDER): 1.0 IRAY rend info : Received update to 04748 iterations after 541.823s.
2017-10-15 14:14:21.659 Iray VERBOSE - module:category(IRAY:RENDER): 1.0 IRAY rend progr: 95.06% of image converged
2017-10-15 14:14:21.664 Iray INFO - module:category(IRAY:RENDER): 1.0 IRAY rend info : Received update to 04797 iterations after 547.393s.
2017-10-15 14:14:21.677 Iray INFO - module:category(IRAY:RENDER): 1.0 IRAY rend info : Convergence threshold reached.
2017-10-15 14:14:27.185 Saved image: XXX
2017-10-15 14:14:27.207 Finished Rendering
2017-10-15 14:14:27.282 Total Rendering Time: 9 minutes 13.99 seconds
4 x1080ti:
2017-10-15 14:17:42.861 Iray INFO - module:category(IRAY:RENDER): 1.0 IRAY rend info : Received update to 04725 iterations after 29.165s.
2017-10-15 14:17:43.613 Iray VERBOSE - module:category(IRAY:RENDER): 1.0 IRAY rend progr: 94.91% of image converged
2017-10-15 14:17:43.613 Iray INFO - module:category(IRAY:RENDER): 1.0 IRAY rend info : Received update to 04767 iterations after 29.918s.
2017-10-15 14:17:43.697 Iray VERBOSE - module:category(IRAY:RENDER): 1.0 IRAY rend progr: 95.02% of image converged
2017-10-15 14:17:43.701 Iray INFO - module:category(IRAY:RENDER): 1.0 IRAY rend info : Received update to 04811 iterations after 30.006s.
2017-10-15 14:17:43.703 Iray INFO - module:category(IRAY:RENDER): 1.0 IRAY rend info : Convergence threshold reached.
2017-10-15 14:17:44.796 Saved image: XXX
2017-10-15 14:17:44.818 Finished Rendering
2017-10-15 14:17:44.872 Total Rendering Time: 32.15 seconds
CPU+ 4x 180TI
2017-10-15 14:20:18.160 Iray INFO - module:category(IRAY:RENDER): 1.0 IRAY rend info : Received update to 04715 iterations after 29.823s.
2017-10-15 14:20:18.228 Iray VERBOSE - module:category(IRAY:RENDER): 1.0 IRAY rend progr: 94.94% of image converged
2017-10-15 14:20:18.233 Iray INFO - module:category(IRAY:RENDER): 1.0 IRAY rend info : Received update to 04757 iterations after 29.897s.
2017-10-15 14:20:18.389 Iray VERBOSE - module:category(IRAY:RENDER): 1.0 IRAY rend progr: 95.07% of image converged
2017-10-15 14:20:18.389 Iray INFO - module:category(IRAY:RENDER): 1.0 IRAY rend info : Received update to 04798 iterations after 30.053s.
2017-10-15 14:20:18.398 Iray INFO - module:category(IRAY:RENDER): 1.0 IRAY rend info : Convergence threshold reached.
2017-10-15 14:20:19.442 Saved image: XXX
2017-10-15 14:20:19.469 Finished Rendering
2017-10-15 14:20:19.518 Total Rendering Time: 32.15 seconds
Somehow ... Rendering process with CPU and GPU didnt benefit at all ... strange :-|
Hopfefully this info serves someone :-D
Thanks. And not strange at all, IMO.
Your 32 threads renders in a little over 9 minutes. Your 4 GPU's render in 32 seconds.
In the time it takes for the GPU's to render the entire scene, the CPU's have only completed about 5% of the scene. So they can only give 5% assistance to the GPU's. And 5% of 32 seconds is, well, 1.6 seconds. And 1.6 seconds is somewhat irrelevant, and probably not even measurable, and certainly within the margin of error.
This seems to be just one more example of how CPU rendering for most of us is somewhat irrelevant if you have a decent GPU.
What I find more interesting is that with 4 GTX 1080ti's you're only getting a 32 second render time for this benchmark. Others have reported 30 seconds with 3 x GTX 1080ti's, and 1.3 minutes with 2 x GTX 1080ti's, and 2 minutes with just one.
It seems like there's some decreasing returns here. Adding a second or third gives about 35% improvement each time, but a fourth gives no additional improvement? I would expect you'd get something like 15-20 seconds with four of them.
After the benchmark i switched to 3xGPU without CPU. Result was a little over 40 secs. Thiss would suggest that the cards scale quiet linear :-|
I suppose we're back to the principle of each additional GPU gives you something less than cutting your render time in half. Something more like a 40% improvement with each card, not 50%. Give or take 5-10%.
You may be looking at a situation where as you add more GPUs, more load is going onto the setup's ability to cool everything. You cards may be running at a lower clock speed with all 4 pegged at 100% usage.
Good point. Just one more reason why it's best to take benchmarks with a grain of salt and not try to get down to the last % point or 0.01 second.
Cooling is fine. Cards boost to 1850-1900 ...
Dont know why this is like it is ...
All i can imagine is that it is a driver issue. hat to switch back lately to a older driver since the newset one gave me a hard time (cards not showing up in DAZ).
I will repeat the bench in some time ... with newer drivers ... and we will see :-D
And don't forget chipset drivers too. I had a problem with chipset drivers with my Ryzen 7 and the GPU was dumping to CPU unexpectedly all because the chipset drivers. They didn't have the Ryzen-specific power plan or something.
Uhh ... that could be ..
I am still using the first driver version that came out. Didnt update at all. Will do this later. Thanks ... didnt think of since that is one of those things one can forget about later. I mean ... after years i used my last system i didnt do chipset updates anymore since there was no "new" stuff out for a long time.
-
I'm not sure what you mean here - a linear improvement would not halve the time for each additional card, it would mean two cards took half the time of one, three cards took a third the time of one, four cards took a quarter the time of one (and so on given really fancy hardware). Halving each time would imply that additional cards added more than oen card's worth of processing power.
Yeah, I think you're right. My brain isn't real sharp on Sunday mornings.
It's not about the charcter that's depicted in the benchmark render. It's about the types of shaders that are used in the render. It's just like with video gaming benchmark suites. Sure you could keep using the 2003 version and see how many points your modern video card scores, but all of the surface and lighting effects are extremely simple compared to what is in a modern videogame, and not an acurate depiction of modern performance needs.
That's what's happening with Genesis 2 using the base materials and seven year old surface materials.
Also, on the old benchmark scene it is outdoors. This eliminates most of the multi-bounce light calculations. A modern benchmark scene needs to stress both the GPU and the VRAM's ability to feed it data. With the recent inclusion of GDDR5X and HBM memory onto video cards, it could cause for example a Quadro GP100 with it's HBM2 VRAM to outperform the GTX 1080 ti in some memory intensive scenes even though it has less Cuda cores.
My test scene results so far.
GTX 1080 ti 11GB
about 7 1/2 min at 480p
about 33 min at 1080p
2160p render cooking right now 10 min so far 370 / 5000 iterations. Should theoreticaly hit 5000 in about 2.5 hours or so? (This resolution will probably be for high end Volta cards)
GTX 1060 6GB
about 13 min at 480p
about 70 min at 1080p
GTX 1080 ti 11GB + GTX 1060 6GB
about 4 1/2 min at 480p
about 23 min at 1080p
At any rate, I'm going to post the current version of my benchmark scene in my Art Studio thread if anyone wants to give it a try, and see if I messed up at keeping it self contained using free Starter Essential components.
https://www.daz3d.com/forums/discussion/171636/jamesjab-s-art-studio-full-of-renders-and-incomplete-projects#latest
Would love to do your bench ... however ... cant find it :-|
It's in my thread now
So you got about a 53% improvement from the 1060 to the 1080ti at 1080p (33 min vs 70 min).
And the results from the original benchmark in this thread show 4.5 minutes vs. 2 minutes, which is about a 55% improvement.
So it looks like the original benchmark agrees with your benchmark, at least between the 1060 and the 1080ti.
Tonight and tomorrow I'll run the 4k version on both cards while I'm asleep/at work and see if a gap starts forming once the resolution bumps to 4K
Thanks for doing all of this. It really helps to confirm or deny what's real and what's myth. Hopefully we can finally get an answer that everyone agrees with.
Do all the 4 PCIe slots have 16 lanes? I imagine that matters too.
Why would that matter? I assumed that once the CPU has transferred the scene to the GPU and it starts rendering, the PCI lanes are kind of irrelevant?