dForcing Hair - 30X0 vs older cards, CPU, etc

I'm curious how much faster some hair items complete dForcing on the new GPUs. Actual dForce hair is usually pretty quick generally, but some of the hair with dForce items can take a while.
Vanilla Settings:
- Collision: Better
- Memorized Bones: Off
- Default Pose for hair (as it loads) and character
Here's what I got on a 2080ti as OpenCL device
Edited because I am dum
Bang Bob Hair with dForce - 4 minutes 35.72 seconds 1 minute 44.27 seconds
dForce Virgin Hair - 29.9 seconds 12.3 seconds
Biscuits Jam Hair with dForce - 1 minute 26.4 seconds 1 minute 18.25 seconds
Classic Long Curly Hair with dForce - 8 minutes 13.18 seconds 2 minutes 45.18 seconds
Classic Long Hair with dForce - 7 minutes 27.61 seconds 1 minute 53.57 secionds
If anyone wants to advance the cause of Science!, please try one or two.
Edited: IDK what I had going on in the background while running the first test. May have had Folding@Home running and forgotten about it.
Comments
Bang Bob Hair with dForce - 1 minute 5.22 seconds
dForce Virgin Hair - N/A
Biscuits Jam Hair with dForce - 0 minute 59.21 seconds
Classic Long Curly Hair with dForce - 1 minutes 51.85 seconds
Classic Long Hair with dForce - 1 minutes 21.54 seconds
Wow. What card?
Edit: Well, turns out a bit less wow after running my own test properly, but still an improvement
Sorry, 3090.
Thanks. That's a much bigger speed improvement on the heaviest hair products than renders get.
You should probably standardized the figure, pose etc. As those effect how long a sim takes.
I took that to mean load default G8F, load hair, hit simulate.
Yup.
What about simulation settings supplied with the hair? Before/after loading it with 'dForce Virgin Hair' it made a difference.
...hmm I have a Maxwell 750Ti (three generations older) with only 640 cores and 4 GB connected to the displays. So if if it takes just over 8 min for Classic Long Curly Hair, on a 2018 Ti with almost 7 times the core count [plus 544 Tensor compute cores on top of that], sounds like I could start the sim, head to the kitchen, wash the dishes, sweep the floor, take out the rubbish and recycle, put the kettle on for afternoon tea, and maybe it would finish just as I sat back down at the desk, cup of hot tea in hand.
This thread is merely intended to compare older apples with newer apples.
I think by default the simulation time is guided by "Stabilization Time: 1.00". Unlike iRay it probably does not run until it reaches a "quality threshold".
Maybe on a lesser card it runs as long as on a 3090, but the results are very different?
Your (Sevrin) numbers look way off though. Classic Long hair took me 2:22 on a 2070 Super, compared to your 7:27.
Note, the very first run took 10s longer for something something setup (compiling kernels?)
Update for advanced science! (Archaeology, the 30xx people might say)
0:04 PhilW's Longdrape Hair (not an error - this hair is a bit simpler than other long hairs but blazing fast)
0:27 Chevybabe's Sea Salt Hair
0:43 Linday's Classic Long Curly Hair
0:58 Mireille's dForce Long Hair
2:22 Linday's Classic Long hair (looks slow but it's the only one which starts the hair from 'electrocuted' position which is very useful in any 'not standing' position)
Maybe not as much apple as yours since I used 4.12.0.85 Beta (on Win7), but I also wanted to try. My results on an old GTX 770 (4GB) are as follows:
Bang Bob Hair with dForce - 18 minutes 24.63 seconds
dForce Virgin Hair - 48.31 seconds (without 'global simulation settings') or 1 minute 6.11 seconds with
Biscuits Jam Hair with dForce - N.A. (don't have it yet)
Classic Long Curly Hair with dForce - 31 minutes 14.67 seconds
Classic Long Hair with dForce - 23 minutes 31.16 seconds
This is meant to be a comparison of hardware. Are you using default simulation settings?
...yeah, those are about the same numbers I'd expect to get. Hmm, maybe could even throw in running the sweeper over the carpet in the main room .
Yes, all default. Fresh restart, hit the Simulation tab > [Defaults], checked the settings.
I listed the other hairs just as a baseline for others who, like me, don't have most of your dForce hairs.
The result of the one item we both tested (Linday's Classic Long Hair) is really strange. Your 2080ti should be faster than my 2070s, but it takes 3x longer.
Might be on my end, of course, but well the simulation works and the result looks good so it can't be that wrong?
Setup: latest DAZ beta 4.12.2.60, latest nVidia Studio driver, regular up-to-date Windows 10, Ryzen 3700x, 32GB ram
update: dForce uses OpenCL 1 AFAIK, so I'd expect results roughly in line with this benchmark:
https://browser.geekbench.com/opencl-benchmarks
or more pronounced differences from 2070s to 2080ti to 3090 in this one:
http://www.luxmark.info/top_results/LuxBall HDR/OpenCL/GPU/1
Hmm. My system's OpenCL score was 136119, which is within range of the average for 2080tis, so it's not that.
Update: Ran the tests again with much improved results and updated the OP.