Re: Failure to run
Posted: Wed Oct 30, 2019 1:49 pm
It is pointless to multithread DGSource() because there is only one decoder on the GPU. All it will do is consume extra memory and add CUDA context switching overhead.
Avisynth+ has the ability to specify multithreading per filter. DGSource() would be serialized while the others can be multithreaded. I imagine that you can do that in 2.6MT somehow.
Seems to me your whole setup is chaotic, which isn't too surprising since you uncritically bring things forward to new environments. In one of your posts you load a plugin from an Avisynth+ directory, but you tell me that you are using 2.6, which I suppose is really 2.6MT. I can't make any sense out of it, and when I ask for clarification you ignore me or just give further chaotic things for me to scratch my head over.
And no answer to my question about the GPU in the Xeon machine.
Avisynth+ has the ability to specify multithreading per filter. DGSource() would be serialized while the others can be multithreaded. I imagine that you can do that in 2.6MT somehow.
Seems to me your whole setup is chaotic, which isn't too surprising since you uncritically bring things forward to new environments. In one of your posts you load a plugin from an Avisynth+ directory, but you tell me that you are using 2.6, which I suppose is really 2.6MT. I can't make any sense out of it, and when I ask for clarification you ignore me or just give further chaotic things for me to scratch my head over.
There you go with more chaos. You think Avisynth controls CUDA parallelism? Hey, it's a video decoder, it uses the video engine, of which there is only one.I'm sure all multi-threading for DGSource is done down at the cuda level (not processor level), transparent to the calling application (DGSource).
And no answer to my question about the GPU in the Xeon machine.