Newbie alert: Julia and GPU. Basic concepts?

[Eric Forgy]

Hi guys,

I have been working with Julia for a while now. An good friend from college has been doing scientific computing with GPUs since around 2002 (early days!), so it is shameful I don't know more about it. I recently saw a demo from Azure that made me realize I have some very basic misconceptions about what computing with GPU is all about. Especially when it comes to GPU and cloud.

I don't understand the internals of a computer very well, but I thought GPUs worked by connecting directly to the monitor for fast display. But how can a GPU on the cloud lead to great visualizations locally? References would be great, but if someone here could give a very simple explanation maybe as it pertains to Julia, that would be awesome. The Azure video is already pretty high level, but it does seem like a talk targeted to GPU experts so I don't fully grok the concepts yet.

If the fuzzy image starting to form in my mind is anywhere near accurate, I can see JuliaGPU playing a bigger role in my FinTech startup.

Thank you and sorry for the vague question.

Cheers!

Eric

[Valentin Churavy]

In essence GPU are processors that are optimized for certain kind of computations. This originally comes from the graphics world, where there is a need to operate over large matrices, doing the same operation at every point. This has now evolved into a realm called GPGPU (https://en.wikipedia.org/wiki/General-purpose_computing_on_graphics_processing_units) using the parallel architecture of GPUs for general compute purposes.

In Julia we have tools to use GPUs as a graphics platforms (see https://github.com/JuliaGL/ModernGL.jl and https://github.com/JuliaGL/GLVisualize.jl) and as a compute platform (see https://github.com/JuliaGPU/OpenCL.jl https://github.com/JuliaGPU/CUDArt.jl).

I hope this helps a bit and this video looks like it might be helpful (https://www.youtube.com/watch?v=oi_9fujflhI)

[Eric Forgy]

Thanks Valentin. Since you replied, I watched the following video:

• Nvidia GTC 2016 Presentation Day 1
  

It's long (more than 2 hours), but I was enthralled and watched the whole thing :)