Morning MD Team
I've upgraded to MD10 which is great so far. Except for layers and cloth intersection.
I've put a belt around a dress and set it's layer to 1 (dress is 0) and no matter what I have tried the belt will not settle outside the dress.
It's not unique to this garment, I've experimented with the garments included with wd and get the same issue.
Is this a known issue?
Thanks John
Window 10, gtx3070, i7 10070k, 16gb
Checklist:
Simulate in CPU (best) mode. Ensure you have switched your layer back to layer '0' after the initial layer simulation, this is a technical requirement you will note in the manual.
In addition check all patterns are sewn with the correct face facing outwards, and finally that all sewing and sewing line angles are not using values that cause collision issues. Set all pattern collision additional pattern thickness such that the garments fit and surface offset has room to simulate.
ah GPU usage is mot that functional yet then, that's a shame.
Thanks for the help.
Hmm...
No it's as functional as it will ever get, BUT if you use GPU simulation out of context when it's also stated in the manual that CPU is for a high quality finished simulation and that garments layering needs to be returned to layer 0 then you are at fault not the software, not the manual, it's an expectations thing.
Your expectations didn't meet the finer points on the use of MD. Plain and simple communication interpretation human error. Cloth collision is unlike other forms of computation, it will always be sequential for a high quality detailed simulation, unless all of maths suddenly changes (which it never will as the foundation of our very world exists on certain fundamental scientific facts).
You somehow imagined none of this applies to you. Which is an expectation issue at the grass roots of what you know verse what is reality. Maths cannot predict where a collision will be before it has occurred, (unlike rendering) you can only chop out fidelity and resolution by estimating it may be there rather than over yonder, a statistical guess. Which means a low fidelity simulation by GPU parrallel administrative processing across many cores splits tasks according to statistical guesses, rather than actual sequential methodical outcome solving. Hence it will always be less accurate as you cannot parallel task (split) a random single collision event that is inherently sequential in nature. Otherwise you could predict all outcomes for all random events before they happen, which is nonsense on planet earth at this point in history. Maybe in 3031.
In the meantime you need to tailor your expectations relative to the state of the art technology that MD is - best in class standard internationally and you might have just 'panned' it outright as it didn't meet your personal view of maths ... or process function. So I hope this has recalibrated your expectations.
If not that's okay as it's personal choice , but where are you going to go on planet earth to get a better simulation? My suggestion is downsize your expectations, use the features for simulation in 'context' and reframe how you think about cloth simulation for now .... at least until another Newton or Einstein shows up, until then it's as good as you will get and outpaces all the other cloth simulations this side of Mars. But if you know of something better - please do tell.
It wasn't a criticism but an observation. It certainly wasn't my intention to offend anyone.
I've been using MD for years, and recommend it frequently and will continue to do so.
No one is offended - it is, what it is ... and why I am pointing this out in a frank manner and maybe a new expectations lens, so you can perhaps re-calibrate how to look at the features, the technology, and the future potential so that you can put into context where GPU and CPU simulation modes fit into the collision based workflow.
It's not a known 'issue' > as that is actually how it functions (for layers) and is likely to, for the foreseeable future (aka - ref your 1st post) and by re-calibrating your expectations (now that you know about the dial it back to zero bit of info in the manual you may have missed you can now explore that feature in detail).
As far as the mathematical 'observed' difference between GPU and CPU performance (a quick tip-toe through the MD user manual) should make you happier you are indeed using state of the art technology in cloth sim, and that you can choose your simulation performance mode for your workflow stages. Every thing else might just be expectational semantics, frustration on using the wrong approach, or a communication blip on not seeing that info in the manual. Now pointed out.
And... until another Einstein or Newton arrives to change maths and the world as we know it, don't expect GPU cloth simulation to outperform the quality of a CPU cloth simulation. Yes one mode is faster than the other in fps development of a garments sewing and arrangement layout in realtime cloth positioning, (a trade off in speeding arrangement workflow verse quality collision frame-steps) but for the qualitative beauty pass finished garment simulation, layering performance, quality constraints based drape and physical preset sim ... CPU will be the go to mode for finishing off a detailed garment for now. That is how state of the art technology best performs > contextually aligned relative to what is possible today on hardware ... and applied efficiently in the workflow in the most appropriate sim mode.
Yes we all want faster, better, lower cost, higher value - and through aligning features with appropriate workflow use, ... you can get there at a push, semantically.
I have lots of tips on how to speed your workflow, ways to improve simulation quality - so do ask, most of it is simple staged process mixed in with what to expect and why to adapt an approach when hitting a bottleneck. As cloth collision can get proportionally slow as the garment complexity increases, the need to adapt how you approach a garment build can make all the difference. In a nutshell the best performance increase you can get with MD is actually not hardware, it's process and aligning how you work relative to the features. That is one thing I have learnt over 8+ years and many garments that I can certainly pass on.
Please sign in to leave a comment.