Cycle X (and OpenCL)
Re: Cycle X (and OpenCL)
This is something they have to write because "We will lock in your body and soul with NVIDIA" sounds really bad. Like I said, the best case scenario is an half hacked solution like current Cycles OpenCL support (i.e. useless).
Re: Cycle X (and OpenCL)
Don’t get me wrong, I get you, and right now CUDA/Optix is the most performant solution out there, I would never purchase an AMD card, not just because of CUDA, their drivers problems are terrible.
If you ask me, and I think it can be a non popular solution, I would remove OpenCL for at least one or two versions and wait until the dust settle in the raytracing noise outside Optix settle and new API’s are clear.
But this could be a bad thing for people using AMD with LuxCore, like people in MacOS for example, that’s why I say it could be a non popular idea.
If you ask me, and I think it can be a non popular solution, I would remove OpenCL for at least one or two versions and wait until the dust settle in the raytracing noise outside Optix settle and new API’s are clear.
But this could be a bad thing for people using AMD with LuxCore, like people in MacOS for example, that’s why I say it could be a non popular idea.
- alpistinho
- Developer
- Posts: 198
- Joined: Thu Jul 05, 2018 11:38 pm
- Location: Rio de Janeiro
Re: Cycle X (and OpenCL)
There is the whole SYCL thing that tries to somewhat standardize things accross vendors. https://sycl.tech/
The Intel OneAPI is apparently one implementation of that and there is https://github.com/illuhad/hipSYCL that tries to build a multi-backend implementation for that, but I don't know how mature this is.
Even this hipSYCL uses the ROCm stack in AMD, so that would be linux-only and not even all AMD cards are supported. The situation is really dire on the AMD side.
The Intel OneAPI is apparently one implementation of that and there is https://github.com/illuhad/hipSYCL that tries to build a multi-backend implementation for that, but I don't know how mature this is.
Even this hipSYCL uses the ROCm stack in AMD, so that would be linux-only and not even all AMD cards are supported. The situation is really dire on the AMD side.
Re: Cycle X (and OpenCL)
I've been waiting months to get an rtx3080, but there is a shortage. Its tempting to go with AMD again, the horse power is sufficient for my needs, but it looks like CUDA is the future. Have to admit Cycles X is looking very attractive now.
Re: Cycle X (and OpenCL)
What alpistinho says is what I mean, right now there is too much noise in what will be the standard or how to acknowledge it, so the only "standard-like" thing is CUDA/Optix, so I would go that route and plan to support other routes in the mid-term future, this will clear the way for solid improvements and allow a better implementation for the future, I don't see dragging OpenCL as something good sadly.
Re: Cycle X (and OpenCL)
I mostly thinking to a freeze more than a death: keep it as it is now and don't touch it. At some point in the future, it will be removed.
Re: Cycle X (and OpenCL)
That's not my expertise here but I agree with this solution considering many people especially with macOS use AMD GPU and they are out of luck with Nvidia.
Omid Ghotbi (TAO)
Last update information
viewtopic.php?p=34390#p34390
My Artstation https://www.artstation.com/omidghotbi
My Instagram https://www.instagram.com/omid.ghotbi
Last update information
viewtopic.php?p=34390#p34390
My Artstation https://www.artstation.com/omidghotbi
My Instagram https://www.instagram.com/omid.ghotbi
Re: Cycle X (and OpenCL)
They are also out of luck with OpenCL, it's deprecated nd not "supported".
@Dade Freezing it won't have the same caveats you mentioned before for the evolution of LuxCore?
@Dade Freezing it won't have the same caveats you mentioned before for the evolution of LuxCore?
Re: Cycle X (and OpenCL)
Hypothetical, the new render engine(s) would not use or share a single line with old OpenCL C code. The idea is to use current C++ code for CPU (not the OpenCL C code).
So it wouldn't be also a totally rewrite of the code.
Re: Cycle X (and OpenCL)
Interesting, I'm eager to see the results