GPUOpen-LibrariesAndSDKs / RadeonRays_SDK

Radeon Rays is ray intersection acceleration library for hardware and software multiplatforms using CPU and GPU
MIT License
1.07k stars 192 forks source link

Is this project abandoned? #202

Open gbeatty opened 5 years ago

gbeatty commented 5 years ago

It's been 8 months since last commit. Open pull requests not merged. Broken Embree and Vulkan builds... Is this project abandoned?

knightcrawler25 commented 5 years ago

They do have an SDK for both radeon rays and prorender, which was released recently here: https://www.amd.com/en/technologies/radeon-prorender-developers

As for the source code we'll have to wait and see if they update this repo with the internal branch.

qhaas commented 5 years ago

Was hoping this project could be a cross-platform alternative to Optix Prime to prevent our users from being locked into nVidia hardware and us from having to maintain separate Optix / Embree (for CPU-only compatibility) codebases. Hopefully it isn't quite dead and can reach feature parity at some point in the future. Wonder if a significant license change is in store for future versions, based on how RadeonRays 3.0 is being distributed (per knightcrawler25 post)...

gbeatty commented 5 years ago

I'm in the same boat. Maintaining multiple code bases is a royal pain.

ivalylo commented 5 years ago

This is exactly why I passed, and left the GPU for "better days". We have NVIDIA which is mostly trying to create a monopoly, and AMD which for such big company, mostly has no clue what it's doing in its software department. Radeon Rays was good idea to break into the NVIDIA monopoly, but now it ended up even worse - closed source, non-commercial license. Thanks, but no thanks... Will stick with Embree for now, which is ages ahead as software quality, feature list, licensing, support. Hopefully someone will fix this mess, or the situation will resolve naturally by people forced to use Optix, since it's at least more mature, better quality and licensing, not to mention much bigger market share...

reduz commented 5 years ago

Agree on this, I think making this a commercial library does not seem like a good decision from AMD, specially when NVIDIA is releasing actual hardware support for raytracing that you can use directly without paying extra after you purchase the hardware, and it works in older versions of their hardware now too.

qhaas commented 5 years ago

Last commit is now over a year ago, guess the open-source effort really is abandoned. It showed so much promise... :(

ivalylo commented 5 years ago

Abandoned for sure... I even wrote an email asking questions about licensing of their new proprietary Radeon Rays SDK, and didn't receive an answer. Too small to get attention I guess... I also say what I think, and they do not like critique... They said they wanted "constructive critique", but it's hard to say what they really mean by that... This project turned out to be a huge disaster and they lost credibility. Who will use their software, when you don't know if it will be available tomorrow? Even if you got licensed, my idea for this library was that its open source, and I can freely modify it to achieve better integration. This was its strongest point... + it's vendor agnostic, but after RTX, it's irrelevant.

Otherwise NVIDIA now has Optix 7 which is based on CUDA, so it fixes its biggest design issue (NVIDIA actually listened to the developers). Also, it supports hardware RTX. Also, it's a mature product with 10 years on the market. Also, the tech you see in Radeon Rays is mostly based on NVIDIA papers... And I get it, they make hardware for games, expecting others to lead the market like MS and Sony. My conclusion is, that if you want to support Radeon hardware, the best way is though DirectX12. I would never consider AMDs own tech. Mantle for example was another abandoned project, which true, evolved into Vulkan, but it's another lost IP. Close to Metal also failed, they moved to OpenCL, but with Apple abandoning OpenCL, this API future is also questionable... AMD dropped the CPU OpenCL support and now develops yet another tech - ROCm, which is currently Linux only. So, it's a mess, while NVIDIA have CUDA for 12 years now, supporting all OSes.

gbeatty commented 5 years ago

I keep getting the feeling that AMD releases a new SW product, sees that few people are using it and makes the (possibly incorrect) assumption that there isn't any interest... so they abandon the project. At least in my mind lots of developers would gladly use their products if they had any confidence that they would be maintained for longer than 1 year. How can I possibly develop any kind of commercial product when it's pretty much a guarantee that next year I'm going to have to find a new solution.

gbeatty commented 5 years ago

I do applaud AMD for working with Khronos to transfer the Mantle IP for use in Vulkan instead of abandoning it. But... don't even get me started on their linux-only ROCm/HPC solution. Yup... totally platform agnostic... as long as your platform is some flavor of linux :confounded:

ivalylo commented 5 years ago

Yes, Vulkan turned out to be a success, although the reasons are probably complex - OpenGL became too rigid to be modernized, DX12 is Windows 10 only, NVIDIA really adopted it (unlike OpenCL)... But it's no longer an AMD IP, it's not really their own success. IMO Radeon Rays for Vulkan was really a potent product. NVIDIA presented their ray tracing extensions latter, only for high end graphics cards. People would have used Radeon Rays as a base, and then vendor specific extensions. This is the power of an open source solution - it could have been the core in many products. IMO RR was huge lost chance to get a lot of people on their side. How many people they expect to get on their new closed sources non-commercial platform? While NVIDIA has free extensions, built in the driver, hardware accelerated on new cards... Much, much fewer. Especially after abandoning this project, so you never know what they will do next... Good Vulkan open source implementation was their best chance, even if it didn't attract so many people right away. They let NVIDIA dominate that one too...

I frankly can't understand what AMD is doing. Many hate NVIDIA because how aggressive they are, but they are also quite predictable. They put a lot of money into IPs and they fiercely promote and defend them. This has bad sides for sure, but the good one is - people have a stable platform for development, which is what matters at the end. AMD is jumping from one idea, to the next one, which is bad for a huge company...

manhnt9 commented 5 years ago

RadeonRays is updated, but inside AMD only. They have released new versions a while ago as compiled libraries. The open-source version seems likely to be abandoned anyways.

gbeatty commented 4 years ago

Well, at least we finally have an answer to this. Radeon Rays 4.0 just released as mostly open source. OpenCL support dropped in this release. Hopefully, they will actually maintain this one...

ivalylo commented 4 years ago

There is nothing open source in this project. It's a binary distribution under MIT license. Another AMD innovation I guess... Still better then their previous evaluation license.

Agorath commented 4 years ago

There is nothing open source in this project. It's a binary distribution under MIT license. Another AMD innovation I guess... Still better then their previous evaluation license.

It looks like RadeonRays 4.0 will, for the most part, be open-source soon: https://github.com/GPUOpen-LibrariesAndSDKs/RadeonRays_SDK/issues/206#issuecomment-629225584