Good opencl tutorials

>>62454273
Just learn CUDA.

>>62454351
Is Cuda accepted on other(not Nvidia) gpus

>>62454384
No.
You don't need non-NVIDIA GPUs though.

>>62454273
Relevant to my interest.

>>62454396
I use Nvidia but I like to strive for the most cross platform solution

>>62454436
It doesn't work like that for GPUs. And since AMD is throwing money at ProRender instead of ROCm, you gotta learn CUDA.

>>62454436
GPGPU is CUDA unfortunatelly, there is little effort to push Apple's OpenCL forward, yes OpenCL was created by Jobs.

>>62454451
Frankly OpenCL sucks and barely moved forward. AMD started Boltzmann exactly because of that.

>>62454451
That's too bad. I was hoping for a standard similar to opengl.

>>62454466
OGL is also shit.
Good shit that AMD donated Mantle to Khronos, so, unsurprisingly, Vulkan is not shit.

I don't fucking care about Nvidia, give me info on OpenCL and let me work.

>>62454477
I'm not saying that opengl is the best, it certainly isn't, but it is widely supported.

>>62454486
Learn CUDA you moron or wait until AMD starts throwing dosh at ROCm.

>>62454496
Begone shill, this is not an Nvidia thread.

I don't care if it's Nvidia or amd. I want cross platform support like opengl. If that doesn't exist then just say so.

>>62454501
Holy fucking fuck you are fucking retarded.
GPGPU == CUDA until AMD will finally start throwing dosh at ROCm.
Learn fucking CUDA.

>>62454273
GPU for computer programming begin almost equals just create kernel function and schedule, better use CUDA more easy,more jobs,better documentation,support,debugging,even use Visual Studio plus Nvidia tools, then could jump to OpenCL.

https://www.youtube.com/watch?v=QA483lIvL-4

>OpenCL

Not going for Vulkan instead.

https://news.ycombinator.com/item?id=14383296

>>62454548
I'll keep that in mind but I would like to see some opencl resources. Cuda is easier to find.

>>62454562
They are not merged yet.
>>62454571
No shit, CUDA is 10 to 100 times more popular than OCL.

>>62454587
That's why I want the hard to find opencl resources.

>>62454396
You can do opencl on CPUs too though... OpenCL is pretty good from a specification and feature context; only slightly less strict to allow more hardware to be used (so slightly more work for pinned memory for example). Everything else that's sucks are implementations and libraries. For AMD GPUs it's a long way to go; even some BLAS libraries are slow and outdated (magma, Vienna), really hoping they will improve on those soon, then target DL/NN.

>>62454571
https://streamhpc.com/knowledge/for-developers/books/

"Programming Massively Parallel Processors"
Still begin gem even if begin CUDA centric.

>>62454609
That's why AMD said fuck it and launched Boltzmann aka ROCm. And since AMD is partnering with chinks and amajew ROCm will be in decent shape soon.

>>62454571
You aren't going to find "resources" for OpenCL. There are some things here and there, but they already assume a high level of familiarity with OpenCL.

>>62454587
They will be. Everything from Vulkan will be directly transferable to OpenCL 2.x. Throwing in to learning OpenCL now is like trying to figure out how an internal combustion engine works while you are developing a turbine engine. It's all backwards.

Fuck, the "primitive shader" and "surface shader" Vega introduces is a literal fuck you to both Khronos and MS.

>>62454521
This, fuck off shitposters.

>>62454611
This looks decent. Any videos with actual c examples?

>>62454670
Wrong. "Prim shaders" are agnostic to actual geometry and vertex shaders written in openGL just like fixed functions are. AMD is even handling how the drivers will determine the path the shaders will require as a blackbox so no one ever has to worry about it.

It's been 15 fucking years and we're still using DX9 tier shit.
And let's not talk about abortion that was geometry shader or bloat introduced by DX11 (fucking TWO hardcoded shader stages and one FF unit for tesselation alone).
Vega is literally pajeet's rage (the dude made the first DX9 GPU 15 years ago) compressed into one uarch.

>>62454732
No, I mean APIs are such dinosaurs AMD had to merge shit at hardware level, expose it to driver and make said driver "dumb" to finally circumvent inefficient abortions that are modern APIs.

>>62454718
OpenCL hell, tried compile real example, good luck, makes everything works and had support functions, spend hours on shitty doc AMD, never ever do OpenCL in Mac OS worst support even.

>>62454273
i keep parsing "opencl" as "open pencil".