(L) [2012/11/27] [tby Dade] [SmallLuxGPU v3.0 Preview Video] Wayback!A SmallLuxGPU 3.0 Preview video is available at [LINK http://youtu.be/jk-N4f9ze4k]
[IMG #1 Image]
SmallLuxGPU ([LINK http://www.luxrender.net/wiki/SLG]) is a LuxRays ([LINK http://www.luxrender.net/wiki/LuxRays]) demo and a LuxRender ([LINK http://www.luxrender.net/]) "proof of concept".
SmallLuxGPU and LuxRays are released under GPL license and the sources are available at [LINK http://src.luxrender.net/luxrays/]. Precompiled binaries are available on LuxRender forums (check the section dedicated to GPU acceleration).
The first part of the video shows a comparison between several features:
- Native Code (CPU-only) Vs OpenCL (CPU+GPUs, [LINK http://www.khronos.org/opencl/])
- Native Code (CPU-only) Vs OpenCL (CPU-only)
- Random sampler Vs Metropolis sampler ([LINK http://en.wikipedia.org/wiki/Metropolis-Hastings_algorithm])
- Path tracing + Light tracing => Bidirectional path tracing
- Bidirectional path tracing Vs Bidirectional path tracing with Vertex Merging ([LINK http://iliyan.com/publications/VertexMerging])
The last part shows an interactive session with Blender and SLG3 running in "Live Mode" and using the new Bidirectional path tracing with Vertex Merging rendering engine.
[IMG #1]:Not scraped:
https://web.archive.org/web/20161005111440im_/http://img141.imageshack.us/img141/4966/slg3small.jpg
(L) [2012/11/28] [tby Dade] [SmallLuxGPU v3.0 Preview Video] Wayback!>> dbz wrote:Great video. I noticed SLG3 is more focused on cpu rendering rather than gpu rendering. I guess it is hard to implement bidrectional pathtracing and vertex merging on the gpu?
Not exactly, it just easier to first write a C++ version and than port the code to OpenCL C. GPU debugging can give me a lot of headaches (and I'm still waiting for the release of OpenCL 2.0 with OpenCL C++ support [SMILEY :cry:] ). I'm planning to write a OpenCL BiDir (both classic and may be the streaming version). The support for Vertex Merging on GPUs is a bit more complex and uncharted territory, may be it can be done with stochastic hash grid as suggested for SPPM [SMILEY :idea:]
(L) [2012/11/28] [tby dbz] [SmallLuxGPU v3.0 Preview Video] Wayback!>> Dade wrote:dbz wrote:Great video. I noticed SLG3 is more focused on cpu rendering rather than gpu rendering. I guess it is hard to implement bidrectional pathtracing and vertex merging on the gpu?
Not exactly, it just easier to first write a C++ version and than port the code to OpenCL C. GPU debugging can give me a lot of headaches (and I'm still waiting for the release of OpenCL 2.0 with OpenCL C++ support ). I'm planning to write a OpenCL BiDir (both classic and may be the streaming version). The support for Vertex Merging on GPUs is a bit more complex and uncharted territory, may be it can be done with stochastic hash grid as suggested for SPPM
Ok, that makes a lot of sense. It is good to hear that more advanced renderers than path tracing will be supported for the gpu as well. Thanks.
