[Feature Request] Cuda Core Shaders compilation

We’ll like most people with dual core pc’s i too have felt the slow shader compilation,but if i’m not mistaking the nvidia cards can use there cuda cores to do calculations normally faster then the cpu.


“Compiling shaders” is one of the most irritating things about using UE4, it should be possible to make more use of GPU acceleration.

Well yes. But you guys missing the point. Shaders must be compiled before they can be used on GPU. GPU can’t self-compile shaders. At least I haven’t yet heard about such thing.

They actually could, but you’d have to code it from scratch in the CUDA supported way.

They wouldn’t add a feature like that that would only work on Nvidia graphics cards.

Well, of course the ywon’t, I was jsut stating that it’s possible.

My thoughts are that it would be a LOT of work for very little gain, and that work would bring with it a lot of additional bugs and issues. Shader Compilation works mostly off of the CPU right now, you could always upgrade that instead if it’s that much of an issue?

Honestly though, compile times for me have always been pretty quick (less than 20-30 seconds when updating/opening a new project with a lot of custom shaders), even with shaders as high as 350 instructions. If you’re working on a shader and it’s compiling while you build it, you can go under the ‘Usage’ drop-down and un-check everything temporarily. When you re-open the material it should compile in the preview window much faster. I do agree that overall iteration times are much longer than they were in UE3, and displaying the preview material during changes rather than the previous material makes comparisons very difficult.

My GTX 970 has a Maxwell GPU that is supposedly a decent processor in its own right, Im surprised that no one has yet written an HLSL compiler that runs on the GPU.

Im not sure, but maybe its because GPUs shine on parallel computations, while code compilation is more or less linear…(?)

I’m surprised that no one yet written any compiler that runs on the GPU.

The thing is, the GPU supports only very basic operations, and you’d have to code everything from absolute scratch. When you create a standard compiler you already have Libraries and things like that. This does’t exist for the gpu. there are librarys to compile with…just saying. Dont know how hard it would be to implement.

Bump for this, especially for GTX 1070 and up it should be possible.

Well on a large project it can take days to recompile shaders after a major version bump and although not to much an issue on smallish project on a major project, such as ours, shader compiles really slows things down a lot. AKA major time sucking vampire.

So yea Cuda and hardware processing is desirable even if limited to nVidia cards as the function of the game has no dependencies to the hardware as a required function of the game if used to make shaders compile faster.

Meaningful stats

This is not an Xphysics thing