LuxrenderGPU

[Dade]

I'm afraid I don't agree with that the optimal CPU/GPU ratio would be 3 : 7.

3:7 but you have to factor I was talking of a single core of a CPU and assuming a ratio between CPU and GPU performance of 1:1 (that isn't usually true). I mean if we factor the average 4-cores CPU and average GPU performances, it looks like we need 1xCPU to drive 3 GPUs with our current architecture (your tests seems confirm this number too). I think it fits well the average Luxrender users.

Could you let me know the result of a profiling session of a single GPU thread, too?

This the data of the thread running on the CPU:

Code: Select all

Flat profile:

Each sample counts as 0.01 seconds.
  %   cumulative   self              self     total           
time   seconds   seconds    calls   s/call   s/call  name   
55.29     34.41    34.41 171963320     0.00     0.00  Path::AdvancePath(Scene*, Sampler*, RayBuffer const*, SampleBuffer*)
22.27     48.26    13.86     4020     0.00     0.00  PathIntegrator::FillRayBuffer(RayBuffer*)
  9.05     53.89     5.63 572215008     0.00     0.00  RandomSampler::GetLazyValue(Sample*)
  4.21     56.51     2.62     4015     0.00     0.01  PathIntegrator::AdvancePaths(RayBuffer const*)
  4.07     59.05     2.54 57712772     0.00     0.00  Path::Init(Scene*, Sampler*)
  3.37     61.15     2.10 57712871     0.00     0.00  RandomSampler::GetNextSample(Sample*)
  0.79     61.64     0.49       45     0.01     0.01  std::deque<RayBuffer*, std::allocator<RayBuffer*> >::_M_reallocate_map(unsigned long, bool)
  0.22     61.78     0.14        1     0.14     0.23  QBVHAccel::BuildTree(unsigned int, unsigned int, unsigned int*, BBox*, Point*, BBox const&, BBox const&, int, int, int)
  0.10     61.84     0.06  5291168     0.00     0.00  Union(BBox const&, Point const&)
  0.05     61.87     0.03 11198604     0.00     0.00  Union(BBox const&, BBox const&)


Just to give an idea of the times: 9% spent in "RandomSampler::GetLazyValue(Sample*)" is just a call to the Tausworthe random number generator.

[KyungSoo]

3:7 but you have to factor I was talking of a single core of a CPU and assuming a ratio between CPU and GPU performance of 1:1 (that isn't usually true). I mean if we factor the average 4-cores CPU and average GPU performances, it looks like we need 1xCPU to drive 3 GPUs with our current architecture (your tests seems confirm this number too). I think it fits well the average Luxrender users.

I agree with that it fits well the average Luxrender users.
But, it cost too much to build and maintain large render farm.
Please, don't stop here the development.

For your reference I attached a picture of 8/16 GPU server.
If we exchane the Tesla with nVidia GTX-295 or ATI HD-5970,
the number of GPU would be 16 (under development).

Tesla-Server-Main-Promo.gif (24.83 KiB) Viewed 853 times

Regards,

[namekuseijin]

damn, I take one week for holydays and evolution never stops...

Guess this year I'll need to get my hands on some Fermi card... *drools*

[KyungSoo]

damn, I take one week for holydays and evolution never stops...

Guess this year I'll need to get my hands on some Fermi card... *drools*

Unfortunately, more then one Fermi cards will be of no use at one CPU machine,
since Fermi will have more then twice power of G200 series GPUs.

[vildanovak]

guys, sorry to ask again, anybody has builds of luxrenderGPU for windows?

[tomb]

guys, sorry to ask again, anybody has builds of luxrenderGPU for windows?

I don't think it's quite ready for this yet - there are still essential stuff hardcoded in the source, such as the number of threads to feed the GPU(s) (which needs to be tuned to your particular HW for it to perform well). I'm sure it will be featured alongside the regular branch dev-snapshots when it reaches alpha-stage (i.e. ready for more "public" testing). I'm not 100% sure what the current status is, but I imagine it will need several refactors (large or small rewrites) before it gets to alpha. Dade, feel free to correct me if I'm wrong here

[Dade]

guys, sorry to ask again, anybody has builds of luxrenderGPU for windows?

I don't think it's quite ready for this yet - there are still essential stuff hardcoded in the source, such as the number of threads to feed the GPU(s) (which needs to be tuned to your particular HW for it to perform well). I'm sure it will be featured alongside the regular branch dev-snapshots when it reaches alpha-stage (i.e. ready for more "public" testing). I'm not 100% sure what the current status is, but I imagine it will need several refactors (large or small rewrites) before it gets to alpha. Dade, feel free to correct me if I'm wrong here

Yup, 100% true.

[vildanovak]

Well you don't have to count me as an end-user, I am actually quite active with blender development...And of course, I can wait

[tomb]

Well you don't have to count me as an end-user, I am actually quite active with blender development...And of course, I can wait

Well, in that case compiling the lux-opencl branch on linux is relatively straightforward (it's CMake driven) I have no idea what it would entail on windows, sorry.

[vimax]

I'm also trying to compile the LuxrenderGPU repositories on windows and I can't seem to do it.
I successfully built the Luxrender 0.7 dev so I know all the paths are there, I also added the OpenCL library but I'm using the lux.sln from the main dev which I'm trying to clean-up. (for instance glass2 has to be removed as it's not in the LuxrenderGPU.
Right now this it's what I'm getting:

Code: Select all

Linking...
warning C4743: 'struct cimg_library::CImg<unsigned char> `static struct cimg_library::CImg<unsigned char> __cdecl cimg_library::CImg<unsigned char>::get_logo40x38(void)'::`2'::res' has different size in 'e:\Kit\LuxBuild\surse\Lux\lux\core\exrio.cpp' and 'e:\Kit\LuxBuild\surse\Lux\lux\renderer\wxglviewer.cpp': 1688 and 24 bytes
warning C4743: 'struct cimg_library::CImg<unsigned char> `static struct cimg_library::CImg<unsigned char> __cdecl cimg_library::CImg<unsigned char>::get_logo40x38(void)'::`2'::res' has different size in 'e:\Kit\LuxBuild\surse\Lux\lux\core\film.cpp' and 'e:\Kit\LuxBuild\surse\Lux\lux\renderer\wxglviewer.cpp': 1688 and 24 bytes
   Creating library Win32\Release/core.lib and object Win32\Release/core.exp
scene.obj : error LNK2001: unresolved external symbol "public: unsigned int __thiscall lux::RenderingEngine::GetThreadCount(void)" (?GetThreadCount@RenderingEngine@lux@@QAEIXZ)
scene.obj : error LNK2001: unresolved external symbol "public: unsigned int __thiscall lux::RenderingEngine::GetThreadsStatus(struct RenderingThreadInfo *,unsigned int)" (?GetThreadsStatus@RenderingEngine@lux@@QAEIPAURenderingThreadInfo@@I@Z)
scene.obj : error LNK2001: unresolved external symbol "public: void __thiscall lux::RenderingEngine::GetNumberOfSamples(double *,double *)" (?GetNumberOfSamples@RenderingEngine@lux@@QAEXPAN0@Z)
scene.obj : error LNK2001: unresolved external symbol "public: void __thiscall lux::RenderingEngine::Signal(enum ThreadSignals)" (?Signal@RenderingEngine@lux@@QAEXW4ThreadSignals@@@Z)
scene.obj : error LNK2001: unresolved external symbol "public: void __thiscall lux::RenderingEngine::Wait(void)" (?Wait@RenderingEngine@lux@@QAEXXZ)
scene.obj : error LNK2001: unresolved external symbol "public: void __thiscall lux::RenderingEngine::Stop(void)" (?Stop@RenderingEngine@lux@@QAEXXZ)
scene.obj : error LNK2001: unresolved external symbol "public: void __thiscall lux::RenderingEngine::Start(void)" (?Start@RenderingEngine@lux@@QAEXXZ)
scene.obj : error LNK2001: unresolved external symbol "public: __thiscall lux::RenderingEngine::~RenderingEngine(void)" (??1RenderingEngine@lux@@QAE@XZ)
scene.obj : error LNK2001: unresolved external symbol "public: __thiscall lux::RenderingEngine::RenderingEngine(class lux::Scene *)" (??0RenderingEngine@lux@@QAE@PAVScene@1@@Z)
Win32\Release/luxrender.exe : fatal error LNK1120: 9 unresolved externals


can someone please shed some light as to where I'm supposed to look?
Maybe I can't compile it with VC++ Express? Or maybe a developer pities us and provides us with a build Luxrender that uses 1 thread GPU and 2 threads CPU (should cover most cases) just to get a taste of what's coming? I for one promise I won't whine about errors, bug or crashes. I just want to try it myself.