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
Is this standard practice people are using for the benchmarks? I didn't do this when I ran mine. I know it's 30 seconds to as much as a minute faster if I do but I didn't think most people were doing that. Ideally everyone would be following the same process so we can get good apples to apples comparisons for those trying to make buying decisions, whatever process that is.
Don't know about everyone else but that's how I ran mine
Wow, I didn't see the very first post stating to change your viewport to iray before running. So my benchmarks are without changing the viewport first. Like you said Apples to Oranges comparison.
It does? Hot digity, I'm gonna have to try that tonight! Thanks!
Switching the viewport to use Iray Interactive mode really should NOT be done. This pre-loads and pre-calculates a lot of stuff, and knowing how that load time is affected (by both the card itself, as well as how many PCI lanes it is running on, and the CPU as well) it really isn't the greatest comparison.
By having the numbers posted based on a fresh start up of DS, load the scene, modify the render settings (optix, optimization), and hit render, then the numbers are very consistent. If you switch to Iray Interactive viewport mode, then suddenly the performance could be wildly different, depending on drivers, etc. And for CPU-only testing, having the viewport in Iray mode actually could cause problems.
The whole point is to get the numbers in the log file anyway, which are startup time, and rendering time. The total time is typically what gets posted, but really it should only be the rendering time we are concerned with, as startup time has a whole host of other dependencies that could affect it greatly.
I too had not noticed that blurb in the OP to switch like that to Iray Viewport beforehand. And I stick to my view that it isn't the correct way to do it.
To add to the discussion of "preview window in Iray mode or not", the above was without preview mode on. With it on, all other settings the same, I got this:
GTX1060 6GB with Iray preview on prior to render: Total Rendering Time: 4 minutes 33.74 seconds
Not an earth shattering difference, really important if you were doing animationsas 25 seconds per frame would add up really fast. But it seems for a static render it's a relatively small change, at least with my hardware.
Following up, was too busy to run this before this morning:
CPU + GTX 1080 Ti: 2 minutes 11.74 seconds
CPU + GTX 1080 Ti + GTX 660: 1 minutes 53.41 seconds
Obvioulsy the benefit of adding in the 660 is pretty low at this point - I really only have it in there as a dedicated PhysX card for gaming, and I don't really need that anymore with the power of a 1080 Ti to play with. I would put the 780 back in as a second card instead, but the backplate on it gets REALLY hot, and my initial testing showed that it was actually causing the 1080 Ti above it to throttle. Maybe someday if I get a new CPU and motherboard with an extra slots' worth of spacing between them, but for now this is what I get.
Hm, on my i7 4771 (stock) and GTX 970 (MSI 100ME Edition) I had 4 minutes 21 seconds, with GPU only and Optix on (optimized for memory), Windows 7 x64, so this model of Ryzen can have a bit of advantage over Haswell cores ^_^
On Zotac 1080 AMP! the time was 2 minutes 51 seconds (the rest of hardware was the same).
Testing on my slightly wonky old system:
AMD Phenom II X6 1090T Processor, 3.20Ghz (6 cores)
16GB RAM, 12800 DDR3
Zotac GF GTX970 AMP! Extreme 4GB
Mobo: M4A89GTD PRO/USB3 AMD890GX AM3
7200RPM HDD
Clean startup of Studio each time, regular openGL viewport each time, benchmark scene, unmodified:
CPU + GPU + OptiX Prime:
hits 90% converged at ~206s @3238 iterations
hits convergence threshold 293s @4833 iterations
CPU + GPU:
hits 90% converged at ~324s @03229 iterations
hits convergence threshold 460s @4812 iterations
So even with an AMD CPU, I'm definitely benefiting from OptiX. I was wondering if that was of most benefit to Intel chipsets but apparently not.
GPU + OptiX Prime:
hits 90% converged at ~194s @3213 iterations
hits convergence threshold 279s @4840 iterations
Takeaway: Wonky dusty AMD CPU is bottlenecking me. I will keep that unchecked going forth!
Thought I'd share my new Ryzen PC build results. I'm keeping this as my base as I tweak and update it like RAM speeds, bios updates and such...which will be a plenty being so new.
My build all at default settings (Win 10 Home at Balanced Power setting)
Ryzen 1700 65w plain jane
Asus Prime X370 mobo
16gb Corsair @ 2133
1TB Plextor M.2 x4 SSD boot and DAZ content
MSI GTX 970 4GB OC model
Gigabyte GTX 1070 8GB OC model
I ran these runs in this manner for each bench test. Opened DAZ, loaded SickleYield scene, ran first load run, another second run with SickleYield scene iRay loaded already on each hardware config.
Ryzen CPU only
1st Total Rendering Time: 25 minutes 16.23 seconds
2nd Total Rendering Time: 24 minutes 48.97 seconds
1st Total Rendering Time: 21 minutes 31.16 seconds Optix on
2nd Total Rendering Time: 21 minutes 12.57 seconds Optix on
GTX 970 only
1st Total Rendering Time: 7 minutes 29.23 seconds
2nd Total Rendering Time: 7 minutes 18.42 seconds
1st Total Rendering Time: 4 minutes 32.13 seconds Optix on
2nd Total Rendering Time: 4 minutes 16.96 seconds Optix on
GTX 1070 only
1st Total Rendering Time: 5 minutes 11.0 seconds
2nd Total Rendering Time: 4 minutes 58.57 seconds
1st Total Rendering Time: 3 minutes 17.36 seconds Optix on
2nd Total Rendering Time: 3 minutes 3.75 seconds Optix on
GTX 970 and 1070 together
1st Total Rendering Time: 3 minutes 17.27 seconds
2nd Total Rendering Time: 3 minutes 4.36 seconds
1st Total Rendering Time: 2 minutes 5.25 seconds Optix on
2nd Total Rendering Time: 1 minutes 52.15 seconds Optix on
CPU+970+1070 together
1st Total Rendering Time: 3 minutes 5.33 seconds
2nd Total Rendering Time: 2 minutes 52.58 seconds
1st Total Rendering Time: 2 minutes 3.17 seconds Optix on
2nd Total Rendering Time: 1 minutes 49.94 seconds Optix on
Looks like a nice boost with Optix on! :)
https://www.daz3d.com/forums/discussion/162446/gtx-1080-ti-benchmark
My Machine consists of the following specs:
GPU : 2x GTX 1080 Ti's Both from Inno3D (NO SLI) (one at 1950 Mhz second at 1800Mhz)
CPU : Ryzen 7 1700X @3,6Ghz boost
RAM : Corsair Vengeance3000 64Gb @2111Mhz
MOBO : Asrock X370 Taichi
Storage: Samsung 1 Tb 850 EVO SSD
Test1:
Sickelyield's benchmark:
https://www.daz3d.com/forums/discussion/53771/iray-starter-scene-post-your-benchmarks
using 2 Ti's and optix acceleration. NO CPU
Completed in 1 min 20s
+- 5000 iterations
you can see more benchmarks in the first link.
Hi,
no Pascal Titan X in the list yet, so I thought I post the results of testing the CUDA-monster I bought yesterday.
Dell Precision T3500 with Corsair 850 Watts 24 GB DDR3-ECC-RAM
CPU:Xeon X5675 @3,07GHz 6 Cores/12 Threads 95 Watt TDP
GPU0:Nvidia Titan X @1898MHz (air) 12GB 3584 CUDA-cores (Pascal)
GPU1:Gigabyte GTX 980 G1 GAMING @1494MHz (air) 4GB 2048 CUDA-cores (Maxwell)
Optix ON:
GPU0+GPU1
Total Rendering Time: 1 minutes 35.53 seconds
GPU0
Total Rendering Time: 2 minutes 11.94 seconds
GPU1
Total Rendering Time: 3 minutes 38.28 seconds
Optix OFF:
GPU0+GPU1
Total Rendering Time: 2 minutes 25.18 seconds
GPU0
Total Rendering Time: 3 minutes 32.73 seconds
GPU1
Total Rendering Time: 5 minutes 56.32 seconds
No CPU testing, would be bottlenecking the GPUs anyway.
Display connected to GPU0
Each test with freshly started Daz Studio 4.9.3.116
No Iray viewport opened.
Loaded the benchmark (Thank you, Sickleyield!)
Hit RENDER
With Viewport set to IRAY (textures and geometries already loaded to the GPUs)
the first Render (GPU0+GPU1+Optix ON)
took Total Rendering Time: 1 minutes 16.88 seconds
i5 3470 3.2Ghz
16GB DDR3 Ram
Gigabyte GTX 1070 8GB Mini ITX OC
Crucial 256GB SSD and WD 7200RPM 1TB HDD
CPU + GPU with OptiX on 3 min 18 Secs
GTX 1070 Only 5 min 2 Secs
GTX 1070 Only with OptiX 3 min 2 Secs
The Titan X is a beast of a card. But there must be some strange driver issue, intricacies of Daz3D with the hardware or its just this particular scene. My two overclocked 1070's came in slightly faster (.03 seconds) than this configuration of a Titan X + 980. I didn't try first setting Daz to iray viewport and always restart the program before testing. Any thoughts by anyone because perhaps balancing two hardware cards is better than mixing and matching cuda's?
For all, in the interest of more accurate benchmarking: Do a "holding" render by starting a few iterations, then cancel. Don't close the rendering window. This keeps the geometry and textures in memory. This part has almost nothing to do with the GPU, and is basically entirely processed by the CPU.
Then, for each subsequent test, use your own stopwatch to time between Iteration 1 and the final iteration. Don't include the initial process of setting up for the render, which likewise heavily involves the CPU.
Optix on/off may not be much of telling parameter as it accelerates ray tracing to triangle geometry, whereas most of the Daz human characters have quad geometry. Depending on your scene it may have a big impact, or it may have none. It would be interesting to do benchmarks with only a base G3 character ... no hair, clothing, set pieces, etc.
Junk, your observation of the two 1070 SC's is in keeping with others. The Titan and 980 are mismatched in architecture, and it's been suggested Iray will "clock down" to the slowest card in the system. This may be what's causing the disparity in the results, or it may be the testing procesure.
Junk,
of course two Pascal cards (your 1070s) can be faster than my Pascal/Maxwell setup.
Your 1800x cpu is much faster than my Xeon, its 15374 vs. 8556 Passmark score.
8 cores/16 threads will load the scene much faster via pcie to the two gpus, who are identical, at least concerning bus width and bandwith.
As Tobor stated, the Titan and 980 are mismatched in architecture, it´s all about different cuda compute capability (5.2 vs 6.1), different host-to-device
and device-to-host speeds.
This will slow down the whole system a bit. But Iray does not "clock down" anything.
The logfile states that in any render under 4GB where I can use both gpus, the Titan X computes
two thirds of the iterations, and the 980 one third. So 11433 GFlops versus 4612 stock makes 16045, plus the OC bonus makes about 17700 GFlops.
For me that reduced the rendering time for a 1 sec/30 frames anim from half an hour to 10 minutes.
And the Titan has to drive my monitor, too.
Yes it is a beast, and I was lucky to get a used one for 850 EUR from a gamer who needed the bucks for 2x1080ti in sli.
If I can acquire a second one, I will. Or I´ll change the 980 to another Pascal card, a 1080ti maybe, the specs are almost the same as the Titan X.
I had problems when I mixed Titan X(Maxwell) and 1080(Pascal) cards. Could only get system to run stable with max of three cards. System fell over completely if I swapping in a 1080Ti beside Titan X(Maxwell). A recent BIOS update and using four identical cards appears to now allow system to run reasonably stable, but has proved not functional as the four 250w double width cards get very hot when rendering for more than a few minutes. I think MEC4D had the right idea starting with a commercial grade chassis and identical video hardware.
I didn't even think about the CPU affecting overall performance let alone PCI generation 1 vs 2 vs 3 loading data to the cards as a factor. So, I believe once laoded to the cards it would shine through. This makes sense and for very short renders of less than 2 minutes it's hard to guage.
And when it comes to utilizing passmark scores I have the Ryzen overclocked to where it is getting a score of 19800ish with 19859 as a high on a semi-consistent level. I just love squeezing every bit of performance out of what I have. Also factor in that if the Titan X was OC'd it would just trounce most everything.
Upgraded my System with New CPU, Mobo and Ram.
AMD Ryzen 5 1600 Overclocked to 3.7Ghz (6 Cores and 12 Threads)
16GB DDR4 Ram Running at 2933Mhz
Gigabyte GTX 1070 8GB Mini ITX OC
Asus PRIME B350M-A Motherboard
Crucial 256GB SSD and WD 7200RPM 1TB HDD
CPU + GPU with OptiX on 2 Min 42 Seconds
Main rig
i7-6700K
48 GB RAM
2 Titan X (Pascal)
Opti-X enabled, CPU disabled. SLI is enabled. The kid was playing a game and I had too much stuff running to turn it off.
1 min 11s
Render rig
AMD FX-8320
32GB RAM
2 Titan X (Maxwell)
Opti-X enabled, CPU disabled. SLI off.
1 min 54 s.
Interesting how big the discrepancy is. I wonder if the system is slower with the older AMD processor. On bigger scenes the difference is around 30-40% faster with the Pascal.
Ryzen 1800X @3.65Ghz stock, GTX Maxwell Titan X, 64GB RAM @ 2666Mhz
GPU Only (OptiX on): 3 minutes 3.27 seconds
CPU Only: 17 minutes 51.60 seconds
CPU+GPU: 2 minutes 47.4 seconds
Nividia 1060 GXT
GPU only 4:46.
Is good enought or not?
For what? As long as you can render what you want to render in an aceptable time it's good enough, if not you need to look at optimising your scenes or a hardware upgrade.
-Desktop- Dell Precision T7500
Dual Intel Xeon X5570 @ 2.93 GHZ (2x4 cores 16 threads total) 24GB RAM
Nvidia Geforce GTX 1060 6GB RAM
3 minutes 48 Seconds - CPU/GPU (OptiX on) (CPU holds @ 3.13 GHZ boost throughout entire render)
4 minutes 17 Seconds - GPU only (OptiX on)
Included picture is a screencap as the render finished, showing GPU stats and render time. The break in the middle of the graph is between renders (GPU+CPU on the left and GPU only on the right)
-Notebook- Dell Precision M6700
Core i7-3840QM @ 2.80 GHZ (4 cores 8 threads) 16GB RAM
Nvidia Quadro K5000M 4GB RAM
12 minutes 38 seconds - CPU/GPU (OptiX on) (CPU holds @ 3.31 GHZ boost throughout entire render)
15 minutes 20 Seconds - GPU only (OptiX on)
From my proudly self built 2 day old rig:
I7-7700K
Asus Strix Z270E
Asus Strix 1080 ti (overclocked out of box)
16 g RAM
--------------------------------------------------------
2 min 0.56 secs
GPU + Optix Memory
1 min 53.98 secs
GPU + Optix Speed
3 min 19.88 secs
GPU - Optix Memory
2 min 16.49 secs
GPU - Optix Speed
With these different settings, is the final render quality the same? To my eyes it is.
Just got a new PC with 3x 1080 TI's! Looking forward to giving it a shot.
Quick question: Is there an appreciable difference in render time depending on HOW MUCH of a GPU's memory is left?
Or does it only matter that the scene fits in the GPU's memory?
Nope. Scene just has to fit, the rest is cudas.
holy smokes that's a lot of CUDA CORES! Don't blink so you see it render! :)
Great thread.
Ryzen 1700 build. (1080 ti + 1070) with Optix on:
Based on the previous reports, the best value for money dual-gpu machine would be a 1070x2 setup (if you can get each for < 350 usd) - faster renders vs a single 1080ti. (I'm biased since I started with just the single 1070)
For me, the additional 300$+ for a dual-1080ti setup (10 sec improvement at most) didn't make sense.
I know! But you know what? It still takes me an hour+ to get to 99.9% convergiance on those pesky interior scenes. Argh!