GPU Accelerated OpenSubdiv is Coming!
Sergey implemented GPU accelerated Opensubdiv this weekend and increased the speed from 2.5fps on his machine to 30fps! It looks like this feature will become available in Blender 2.72.
Petit homme vert
Sergey implemented GPU accelerated Opensubdiv this weekend and increased the speed from 2.5fps on his machine to 30fps! It looks like this feature will become available in Blender 2.72.
Holy mother of…
Will this work on nvidia and amd GPUs?
What does this mean? What’s the benefit of this addition?
Hey Bart, I think you should add a link in the post to the video Ton shared on Twitter :-)
CloudFlare is all screwed up today, I can’t get the cache to clear. Until then, here’s the video!
https://www.youtube.com/watch?v=g27QyXcBW5Y
what is the name of this program ?
Blender?
yet another GPU accelerated feature that will only work on Nvidia GPU’s, well looking on the good side of life I suppose its a good thing because I currently have a Nvidia GPU but as everything has two sides, there is also the fact that AMD are becoming more and more the distant past and not the current future which means that the company as more time passes by will find it harder to catch up, but it is also annoying for me because I cannot afford a new Nvidia Gpu 700 or 800 series graphics card to replace my Current Nvidia 560Ti but AMD’s graphics cards are comparable to Nvidia’s graphics cards but are less expensive making them ideal since I can afford them how ever they do not bloody work, stupid company which is also another annoying thing because I realy want that company to succeed even though I am an Nvidia fan because that is the only brand I can afford.
Opensubdiv should work with AMD GPU`s, officially it supports: OpenCL, CUDA, GLSL, DirectX11, OpenGL and Open GL ES. So I expect it should work fine in Blender.
Very excited for this, fxguide had an awesome article about opensubdiv a year ago. And to see this in Blender is mind blowing, can`t wait to use it.
Thanks Sergey for working on it!
Thanks for the tip on the fxguide article. For the people who are interested: http://www.fxguide.com/featured/pixars-opensubdiv-v2-a-detailed-look/
thanks that’s nice to know, oh and I have to have a lok at what this is as well because I haven’t actually researched it yet lol, but it is look way col on the video.
Oh and one more thing, I would really like to know how many things you could actually GPU accelerate before the GPU didn’t give any performance improvement over traditional CPU implementations, but I guess it doesn’t make a difference because how many different GPU accelerated features are you gong to use at once any ways, but it would be nice to know, I guess it would also depend on the GPU you have and how much ram it is at its disposal.
So cool! Can’t wait to see how it’s implemented in the blender codebase. Newer AMD GPUs should work well, older ones have been tricky to ensure work. There may be platform specific ifdefs to choose OpenCL or CUDA to get good behavior by default. Rock on Sergey!
Fantastic!
Now you can model with even higher poly counts ;) . I want a tut on CG cookie about that when it comes up.
You can count on it!
Cool. Now wondering if it will work well with the solidify modifier. The solidify creasing is junk now.
Oh, yeah!!
Thank you Sergey and all you mad scientist devs!!
You rock!
This is great!
Will this include creasing, or I’ll be very basic at first?
Good question
Just awesome !!! Great works guys !!!
Wait, does this mean the CPU-only version is already in trunk and I missed it somehow?
Nope not yet. I guess they will come together.
I have a few questions about this OpenSubdiv Integration. are you going to target implementation of some more advanced features of the library such as a hierarchy edit system, Ptex support, some of Pixar’s creasing algorithms as well as adaptive distance biased falloff? Is it planed to implement features such as Hardware Accelerated Tessellation Shaders for viewport subdivision once blender migrates to OpenGL 4.0 to speed up viewport and possibly render subdivision? Will GPU Subdivison and most of the enhancements of subdiv work if the Subsurf Modifier is placed underneath another modifier or will it just fallback to CPU subdivision and still be as slow as always, and will the implementation of the algorithms extend to MultiRes? Lastly I heard that Sony Pictures tried integrating OpenSubdiv into thier Arnold render engine (which is very similar to Cycles in the way it functions as a Path Tracer, and is where OpenShadingLanguage originated from) and they had disappointing results when using subdivision through OpenSubdiv and found that when using the CPU their original algorithms were faster than OpenSubdiv on the CPU. They also found that the adaptive Tessellation had some major downfalls when used in Arnold’s path tracer render engine, will those shortcomings also have the same negative effects if implemented in Cycles?