get_gpu_info_c not detecting any GPU (compatibility issue with CUDA 11?)

[jucendrero]

I've recently installed h2o4gpu and I've been able to successfully run the test code provided on the README.md file. However, I can't run any algorithm using the GPU because the library always selects the fallback sklearn class, even if I set backend='h2o4gpu' as a parameter.

I've noticed that the problem is that the get_gpu_info_c function does not detect my GPU (any call to this function returns a (0,)).

An important point to remark here is that I'm using CUDA 11 (which is the only version of CUDA available for my OS). I'm aware that the h2o4gpu installation requisities only specify previous versions of CUDA (8, 9, 9.2, 10); however, based on the answers to other issues ( #746 (comment) ), I'm assuming that all versions of h2o4gpu are forward compatible, so I used the cuda10 installer anyway.

I was wondering if this might be a compatibility issue with CUDA 11.

---

Minimal example

>>> from h2o4gpu.util.gpu import get_gpu_info_c
>>> get_gpu_info_c()
(0,)

Environment

• OS platform, distribution and version: Ubuntu 20.04
• Installed from (source or binary): pipenv
• Version: 0.4.0
• Python version (optional): 3.6.7
• CUDA/cuDNN version: CUDA 11.0 / cuDNN v8.0.2
• GPU model (optional): GeForce RTX 2060
• CPU model: AMD Ryzen 7 3800X
• RAM available: 32 GB.

[monk1337]

@jucendrero Did you find the solution? I am facing the same issue. If I am selecting the backend='h2o4gpu'

I am getting this error :
AttributeError: 'NoneType' object has no attribute 'make_ptr_float'

I checked with get_gpu_info_c it's not detecting GPU.

[sh1ng]

Tested on RTX 2080 and CUDA Version: 11.1.

Does anything specific in your environment? Last time I saw similar issue when multiple gpus are used and one of them without CUDA support.