my quote is from their answer for some months ago. They said that they’re trying some other ways to utilize AMD’s but not openCL. looks like they’re totally gave up on that.
So i think that’s a miracle that openCL version of cycles even exist at all.
If not, they will lose most of the users. This is too easy thing. And non-monopol apologist open-source society create monopol with their own hands.
I couldn’t disagree more. It is literally a waste of resources!
1 Like
it’s already stated a lot of times here. It has nothing to do with nvidia monopoly
CUDA vestion was easier to develop
OpenCL is harder to develop
thats it
Brecht could went for the principle - and ignore cuda (like sidefx does for example). So it is possible that we’d still waited for the first version of cycles (as I wait for 6(!) years of GPU Mantra (and now Karma) renderer from sidefx).
Brecht’s choice was obvious - he used api that allowed him to make thing that works.
1 Like
Then you never be a programmer. We must always waste of resources, because all platforms needs different necesserities. And if you want gain users of this platforms, you must waste of resources.
Brecht could went for the principle - and ignore cuda (like sidefx does for example). So it is possible that we’d still waited for the first version of cycles (as I wait for 6(!) years of GPU Mantra (and now Karma) renderer from sidefx).
What does this have to do with the present? Today Cycles supporting OpenCL. But each version they are broke codes. If you add a new ability, you must not broke old abilities and stabilities. But Cycles everytime do this. If you can not write good codes for this ability for OpenCL, then not add for OpenCL. But Cycles add this and broke stability and old abilities.
What does this have to do with the present?
It does that even big commercial graphic software companies still have a lot of problems when they try to write full production renderer with opencl.
Even companies that officially “hate” nvidia like sidefx.
So I could extrapolate it onto blender/cycles development. It totally explains why opencl version is “worse” then cuda one. Without any pro-nvidia conspirasy.
If you can not write good codes for this ability for OpenCL, then not add for OpenCL.
so it is now a bad thing that opencl version of cycles even exists?
This is not excuse for every version have broken stabilities. If one code is stabil, then this code stabil. If you preserve this stability, then add new abilities, if not, not add. We want stability over new abilities. We are consent some abilities not work on OpenCL. When developers can write stabil codes, then will add this abilities.
It is too late for that 
It is common practise to evaluate each platform to figure out how it can be used and whether it is going to be worth it. You are also doing the same with functionality that requires continuous support due to broken drivers or other third party issues.
Sometime the best call is to simply make the decision, it is not worth it!
I have seen plenty of projects where they made the call that OpenCL is not worth it, even though they threw out Mac users because of it and everyone with AMD GPUs.
Sometime the best call is to simply make the decision, it is not worth it!
Sorry, but real world is not too easy like this.
it definitely isn’t. You’re demanding open source, free tool for a things that even big commercial developers couldn’t deliver. Primarily because opencl is in a bad shape for that.
2 Likes
No one said that such a call would be easy. Nevertheless it may be the right call if too many resources are wasted.
Big developers only thinks own budgets. But this is foundation, not only a open source program.
And many users lost. If users a resource, yes you lose most of users at the same time.
man, you’re not listening.
Problem is not in budget or a goodwill.
Problem is in openCL and amd drivers.
Every programmer, who tried to write production raytracer say that.
1 Like
Man, give me an example code design. Why OpenCL bad? I am programmer. I am not specialist of OpenCL, but I am understand codes and architectures.
For 2.80, Cycles works good with OpenCL. How do this Cycles? You only offer excuses, but don’t know why you said this. Only said I hear to another. Cycles will keep stability, this is enough. If they can not add new abilities for OpenCL, then not add, only add for CUDA.
Why me. I don’t have experience with opencl api.
But i believe to whatm for example, redshift developers say.
Try ask them
Why you defend, if you don’t know?
I know what I has been said by people who has my credence.
You said yourself that you’re not opencl spesialist. People who spent years trying to make it work say that there are a lot of problems.
Then they must defend, not you.