[RFC] Drop support for CUDA 10

[shiyu1994]

Summary

CUDA 10 is very old. And with CUDA 10 we found some compilation problems that do not occur in CUDA 11. See (https://github.com/microsoft/LightGBM/pull/5605#issuecomment-1471250105) for example. Dropping support for CUDA 10 may reduce maintenance and CI test cost. Just want to hear your ideas about this @guolinke @jameslamb @StrikerRUS @jmoralez.

[jameslamb]

I'm +1 on moving LightGBM's minimum supported CUDA version to CUDA 11.x.

We have such a small team of maintainers here, and so few non-maintainer contributors around the project right now, I think the reduction in maintenance burden is necessary to ensure that at least the CDUA 11.x support is high-quality.

---

Some references to help with this decision...

The last CUDA 10.x release, v10.2, was in November 2019.

The first CUDA 11.x release, v11.0.1, was in June 2020. (release history).

And it seems to me that many other machine learning projects supporting GPU acceleration have already done that.

XGBoost has been requiring CUDA 11.x since at least June 2022: https://github.com/dmlc/xgboost/issues/8006#issuecomment-1159079213 (cc @trivialfis @hcho3 could tell us if it's even further back than that)

RAPIDS announced that they considered CUDA 10.2 "deprecated" as of February 2021 (https://docs.rapids.ai/notices/rsn0005/)... not sure if / when they formally removed support for it.

Pytorch seems to only be supporting be publishing precompiled binaries CUDA 11.x as far as I can tell (not sure if they support building from source against older CUDA)

• https://pytorch.org/get-started/previous-versions/
• https://anaconda.org/pytorch/pytorch-cuda/files

Tensorflow dropped support for CUDA 10 as of v2.4.0, December 2020 (support table, release history)

[hcho3]

@jameslamb We dropped CUDA 10.x support in February 2022: https://github.com/dmlc/xgboost/issues/7366

[jameslamb]

Perfect, thanks for that @hcho3 !

[jameslamb]

To add more evidence here... the default CUDA on Google Colab is v12.0.

Ran !nvidia-smi in a notebook with a T4 GPU attached tonight:

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 525.105.17   Driver Version: 525.105.17   CUDA Version: 12.0     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  Tesla T4            Off  | 00000000:00:04.0 Off |                    0 |
| N/A   39C    P8     9W /  70W |      0MiB / 15360MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+

[github-actions[bot]]

This issue has been automatically locked since there has not been any recent activity since it was closed. To start a new related discussion, open a new issue at https://github.com/microsoft/LightGBM/issues including a reference to this.