GPU Support TODO List - Githubissues

dmlc / XGBoost.jl

XGBoost Julia Package

Other

288 stars 110 forks source link

GPU Support TODO List #134

Closed ExpandingMan closed 1 year ago

ExpandingMan commented 1 year ago

The following needs to be done for this wrapper to get GPU support. From the C API docs it looks like this would mean only nvidia as I think it necessarily relies on CUDA.

[ ] Create a new xgboost_jll. I really have no idea how to do this for GPU targets, so that'll take some figuring out. If anybody is good with BinaryBuilder and wants to help with that it would be much appreciated, as this is the step that would likely take me personally the vast majority of the effort.
[ ] Check that the array handles in CUDA.jl are the same as those expected by the libxgboost API. If they are not we'll have to figure out what the difference is.
[ ] Add CUDA.jl as a dependency. I think this is a complete non-issue as using CUDA works fine on machines without an nvidia GPU.

Roh-codeur commented 1 year ago

thanks for your work on this, I look forward to seeing this. will add a significant impact on my project, thanks!

Roh-codeur commented 1 year ago

@ExpandingMan : hey mate, thanks for your help with this. did you get a chance to make progress with this? this will be of tremendous help to my project

thanks Roh

ExpandingMan commented 1 year ago

I have not, but I also have not forgotten about it.

The part of this with the most question marks in my mind is how exactly to setup the xgboost_jll which I'm going to have to figure out to some extent before I even attempt to make a wrapper. So if anyone wants to help this along getting the xgboost_jll updated to include GPU support and automatically use that build when nvidia cards are present would be best way to do that.

Roh-codeur commented 1 year ago

ahh ok, sorry, not my area of expertise or even knowhow. thanks for your help in this, ta!

tylerjthomas9 commented 1 year ago

Full GPU support is now available in v2.2.0 (https://github.com/dmlc/XGBoost.jl/pull/142).

Roh-codeur commented 1 year ago

Full GPU support is now available in v2.2.0 (#142).

this is awesome! thanks a lot @tylerjthomas9 and everyone else who helped.

[Edit] this is awesome! thanks a lot to @ExpandingMan and @tylerjthomas9 and everyone else who helped.

tylerjthomas9 commented 1 year ago

this is awesome! thanks a lot @tylerjthomas9 and everyone else who helped.

@ExpandingMan is the primary person behind all the recent improvements to XGBoost.jl

ExpandingMan commented 1 year ago

Forgive the mutual admiration society, but by far the hardest part of doing this was updating the XGBoost_jll to properly handle CUDA dependencies which was done by @tylerjthomas9 . The wrapper part was, fortunately quite easy, thanks in part to the very easy to use CUDA.jl.

So yes, I think this wrapper with GPU support is a nice success story for the Julia wrapping and packaging ecosystem :smile: