[BUG] PCA Dense CUML23.08 CUDA12

Question

[BUG] PCA Dense CUML23.08 CUDA12

Closed this issue 5 months ago · 6 comments

Please don't use cuml-23.08 with cuda 12.
PCA will fail for larger datasets

Answer 1 · 2023-12-12T15:13:51.000Z

@Intron7
Hello, I also caught the error as you mentioned in the rapidsai/cuml#5555.
So, did you fix this bug now?
And whether I need to downgrade CUDA 12 to 11.8 to work with cuml-23.08 (rapids-23.08)?

My ENV:
CUDA Version: 12.2
rapids-23.12
cuml-23.12.0a55

Answer 2 · 2023-12-12T16:18:00.000Z

@xflicsu i can't fix this bug. It's a bug that I reported to the rapids team a while ago. However so far it has not been fixed. The purple label rapids-bug indicate bugs that are on rapids side and are beyond my control to fix.

So for the time being I recommend to use the 11.8 version of rapids.

Answer 3 · 2023-12-13T00:45:09.000Z

@Intron7 Thanks for you quick response!

Answer 4 · 2024-07-04T09:27:33.000Z

@Intron7 Hi, I also encountered a similar problem, do you have a good solution now? I used CUDA 12.1 and rapids==24.08, 3ks!

Answer 5 · 2024-07-05T07:49:27.000Z

I really don't have a good solution. Other than to use CUDA-11.8. But from what I have heard it will be fixed with rapids-24.08

Answer 6 · 2024-07-05T11:47:27.000Z

I really don't have a good solution. Other than to use CUDA-11.8. But from what I have heard it will be fixed with rapids-24.08

OK，thanks for you quick reply, I tried rapids==24.08a and found that the problem still exists. But the official CUDA version is 12.0 and 12.2, I use 12.1 . I will continue to try, thank you again.