CV Apart from Cycles I have not seen a renderer is doing well in both CPU and GPU. Maybe Karma can go into similar places. Did I get it right that Karma can dynamically switch to CPU if GPU runs out of Ram?
XPU runs on both CPU and GPU so if there is less RAM it still can run on CPU. But i don't think there is a out of memory solution for the GPU.
CV Did I get it right that Karma can dynamically switch to CPU if GPU runs out of Ram?
CPU and (multiple) GPU devices are active while rendering. If a GPU device fails for whatever reason (eg runs out of memory) it will gracefully/silently shut down, leaving the other devices to finish off the work and finish the frame. This happens automatically, so provides a failsafe way to work.
CV Did I get it right that Karma can dynamically switch to CPU if GPU runs out of Ram?
CPU and (multiple) GPU devices are active while rendering. If a GPU device fails for whatever reason (eg runs out of memory) it will gracefully/silently shut down, leaving the other devices to finish off the work and finish the frame. This happens automatically, so provides a failsafe way to work.
Man I wish dynamics worked that way too. Every time a solver runs out of GPU memory we’re screwed
Ralph Pinel How long before Karma will be a serious replacement for Redshift in Houdini?
well .... according to how things are progressing in karma XPU it will be the fastest renderer their is today BUT if you have AMD card go with redshift because karma XPU (with GPU) only support nvidia card (and don't ask me why in the year 2023 the developers still don't develop in advance for all devices, because hip (amd library) exists for along time now .)
habernir (and don't ask me why in the year 2023 the developers still don't develop in advance for all devices, because hip (amd library) exists for along time now .)
Every development takes time karma XPU first target is surly to be production ready. Porting is one task and the support is the important one that can take ressources. Customer demand is also important when just some customer request for it won't happen.
habernir (and don't ask me why in the year 2023 the developers still don't develop in advance for all devices, because hip (amd library) exists for along time now .)
Every development takes time karma XPU first target is surly to be production ready. Porting is one task and the support is the important one that can take ressources. Customer demand is also important when just some customer request for it won't happen.
Customer demand its misleading because when you develop just for one company and thats nvidia then you created customers that most of them will be with nvidia cards (you didn't created them, you force them) and then of course you will have a demand mainly for nvidia technology , and thats the problem thats most of the companies doesn't understand ,thats its like magic wheel thats turn its self around and around soo sidefx customer will buy only nvidia cards because houdini will used nvidia cards and sidefx only develop their product for nvidia because their customer have nvidia cards . now you understand?
and yes every development its taked time BUT converting/porting code to another its take more time than to develop in advance for all devices.
