cusolver Error:

[GuoHan77]

Environment

Hardware Environment(Ascend/GPU/CPU):

Uncomment only one /device <> line, hit enter to put that in a new line, and remove leading whitespaces from that line:
gpu
/device ascend

/device gpu

/device cpu

Software Environment:

• MindSpore version (source or binary):2.2.10
• Python version (e.g., Python 3.7.5):3.9
• OS platform and distribution (e.g., Linux Ubuntu 16.04):Linux
• GCC/Compiler version (if compiled from source): 9.2.0

Describe the current behavior

Traceback (most recent call last):
File "", line 1, in
File "/opt/conda/envs/cfm/lib/python3.9/site-packages/mindspore/common/stub_tensor.py", line 49, in fun
return method(*arg, **kwargs)
File "/opt/conda/envs/cfm/lib/python3.9/site-packages/mindspore/common/tensor.py", line 493, in str
return str(self.asnumpy())
File "/opt/conda/envs/cfm/lib/python3.9/site-packages/mindspore/common/tensor.py", line 964, in asnumpy
return Tensor.asnumpy(self)
RuntimeError:

• cusolver Error:

---

Failed to create cusolver dn handle. | Error Number: 7

---

• C++ Call Stack: (For framework developers)

---

mindspore/ccsrc/plugin/device/gpu/hal/device/gpu_device_manager.cc:45 InitDevice

Describe the expected behavior

Steps to reproduce the issue

Related log / screenshot

Special notes for this issue

[zhouyifeng888]

Same as the issue:#333, this issue may be also caused by the following two reasons:

It could be that your operating system version is too old. The officially recommended version is Ubuntu 18.04. Versions above 18.04 have also been used by many developers without any issues. However, version 16.04 has not been officially tested, and there may be compatibility problems due to outdated dependency libraries.

I think the second reason is more likely: are you using a 40-series or newer GPU, but installed CUDA version 11.1 or lower? The 40-series GPUs do not support CUDA 11.1. Many developers, including myself, have encountered the same error when using CUDA 11.1 with 40-series GPUs: “Failed to create cusolver dn handle. | Error Number: 7”. However, everything works fine with CUDA 11.6. If you’re using a 30-series GPU, CUDA 11.1 should work without problems, but note that 30-series GPUs do not support CUDA 10.1 either.