Memory and Performance issue of Taichi

Question

Memory and Performance issue of Taichi

ifsheldon opened this issue 3 years ago · 4 comments

With the current Taichi (v0.9.1 - 1.2.1), calling Tube N times will result in N^2 time complexity because when creating a field Taichi need to inject kernel information into a field which leads to memory movement which is O(M) where M is the number of existing fields. The complexity is simply 1+2+3+....+N = O(N^2). This is not our fault. And Taichi developers are fixing it, although it takes quite some time.

In the forward-only computation, this can be resolved by eagerly destroying fields and SNodeTree, which is included in stannum=0.6.2.

Answer 1 · 2022-06-30T12:57:02.000Z

related Taichi issue.

Answer 2 · 2022-09-14T07:51:15.000Z

The Taichi issue is closed. Looking forward to Taichi new release!

Answer 3 · 2022-09-20T14:56:10.000Z

The new release of Taichi 1.1.3 includes the PR for performance improvement, so this issue is resolved.

Answer 4 · 2022-09-23T15:03:02.000Z

Reopen this because Taichi still has the problem on other backends (Vulkan and Metal) while the problem was fixed on CPU and CUDA backends.

See this issue.