using CUDA to alleviate high-dimensional problem in projection method

RJDennis / SolveDSGE.jl

A Julia package to solve, simulate, and analyze nonlinear DSGE models.

MIT License

81 stars 25 forks source link

using CUDA to alleviate high-dimensional problem in projection method #36

Open v1cte opened 3 years ago

v1cte commented 3 years ago

How about using CUDA.jl to speed up the calculation in smolyak method? (the speed gain is quite promising according to github.com/ikarib/smolyak

RJDennis commented 3 years ago

I have a multi-threading option for computing the weights, but I haven't tried to use CUDA or any GPU processing; this might be worth looking into in the future. The particular package you link to does not appear to have a license so I believe its contents will be covered by international copyright, which essentially prevents me from looking at it.

v1cte commented 3 years ago

The package I link is object-oriented, and as far as I know Julia do not support OOP. The reason I link it to here is just to show the possible benefits of using GPU in projection method since the author's code finished task in 6 minutes, but I think Karibzhanov's parallel approach may help you. And here is a note may also help you in the future.

RJDennis commented 3 years ago

Okay. Thanks for the references.