Tensorflow GPU Could not load dynamic library 'cusolver64_10.dll'; dlerror: cusolver64_10.dll not found
Solution 1
Step 1 Move to C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2\bin
Step 2
Rename file cusolver64_11.dll To cusolver64_10.dll
cusolver64_10.dll
Solution 2
TL;DR For TensorFlow
ver >= 2.4.0 on Windows
, install exactly those versions of CUDA Toolkit
and cuDNN
highlighted below i.e. those listed in the official requirements.(v11.0 as opposed to v11.2)
On Windows
, the TensorFlow
^ install requirements at the time of writing are as stated here
-
NVIDIA® GPU drivers —CUDA® 11.0 requires 450.x or higher.
-
CUDA® Toolkit —TensorFlow supports CUDA® 11 (TensorFlow >= 2.4.0)
-
CUPTI ships with the CUDA® Toolkit.
-
cuDNN SDK 8.0.4.
-
(Optional) TensorRT 6.0 to improve latency and throughput for inference on some models.
The problem you are facing has probably to do with the version of CUDA® Toolkit. Tensorflow
is picky about the version of dependencies. Have a look inside C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2\bin
**. You should be able to find most^^ of the dlls needed by TensorFlow
there. You may notice that it contains cusolver64_11.dll
as opposed to the expected cusolver64_10.dll
as stated in the output.
Though the renaming hack mentioned in an answer above works, it's not guaranteed to work reliably all the time. The simple and correct solution is to install the correct dependencies, to begin with.
At the time of writing the compatible versions of CUDA Toolkit
and cuDNN
are
CUDA Toolkit 11.0 (May 2020)
cuDNN v8.0.4 (September 28th, 2020), for CUDA 11.0
from among the plethora of available versions of both, listed here & here.
More recent versions (I tested v11.0 onwards) aren't yet supported. I remember having the same problems with an earlier version of TensorFlow a few years back.
^ For ver >1.15, TensorFlow
has GPU support included by default hence the CUDA requirements. When unavailable, TensorFlow
works fine - it just reverts to CPU execution.
** Or wherever you installed the toolkit
^^ cudnn64_8.dll
comes with cuDNN SDK
Solution 3
I had the same problem. It turns out that CUDA 11.0 contains cusolver64_10.dll (that's probably why they indicate CUDA v11.0 in the tensorflow build guide here https://www.tensorflow.org/install/source_windows). Make sure to download cudnn as well!
Solution 4
For TensorFlow 2.4.1, the renaming hack will work if CUDA 11.2 needs to be installed. I suggest installing CUDA 11.0 + cuDNN 8.0.4 for TF 2.4.1, as @lineage wrote above, and then the renaming won't be necessary, and your GPU will be recognized.
For TensorFlow 2.5.0, I just got my GPU recognized using CUDA 11.2.2 + cuDNN 8.1.1. In that case, DO NOT rename the cusolver file. TF 2.5.0 expects the "cusolver64_11.dll" filename.
c> python
Python 3.9.4 | packaged by conda-forge | (default, May 10 2021, 22:10:34) [MSC v.1916 64 bit (AMD64)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import tensorflow as tf
2021-05-28 08:11:24.517894: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cudart64_110.dll
>>> print(tf.version.VERSION)
2.5.0
>>> print("Num GPUs Available: ", len(tf.config.list_physical_devices('GPU')),
... '\nDevice: ', tf.config.list_physical_devices('GPU'))
2021-05-28 08:12:19.501812: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library nvcuda.dll
2021-05-28 08:12:19.530869: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1733] Found device 0 with properties:
pciBusID: 0000:01:00.0 name: NVIDIA GeForce GTX 1080 with Max-Q Design computeCapability: 6.1
coreClock: 1.468GHz coreCount: 20 deviceMemorySize: 8.00GiB deviceMemoryBandwidth: 298.32GiB/s
2021-05-28 08:12:19.531377: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cudart64_110.dll
2021-05-28 08:12:19.597785: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cublas64_11.dll
2021-05-28 08:12:19.597992: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cublasLt64_11.dll
2021-05-28 08:12:19.618849: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cufft64_10.dll
2021-05-28 08:12:19.634321: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library curand64_10.dll
2021-05-28 08:12:19.677539: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library **cusolver64_11.dll**
2021-05-28 08:12:19.731541: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cusparse64_11.dll
2021-05-28 08:12:19.746271: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cudnn64_8.dll
2021-05-28 08:12:19.746674: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1871] Adding visible gpu devices: 0
Num GPUs Available: 1
Device: [PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]
>>>
Related videos on Youtube
Haseeb
Updated on June 01, 2021Comments
-
Haseeb almost 3 years
When i run
import tensorflow as tf tf.test.is_gpu_available( cuda_only=False, min_cuda_compute_capability=None )
I get the following error
-
Douglas Marttinen over 3 yearsCreating a symlink with powershell also works and is probably preferred:
New-Item -ItemType SymbolicLink -Path .\cusolver64_10.dll -Target .\cusolver64_11.dll
-
Haseeb over 3 years@DouglasMarttinen Thanks it will help others
-
Jerome Ariola about 3 yearsWhy doesn't it work with cusolver64_11? I had this same issue even after updating my tensorflow...
-
Daniel Argüelles about 3 yearsSeems not too profesional answer but it works. Thank you :)
-
Thulfiqar about 3 yearsthis will cause the problem in this question stackoverflow.com/questions/53698035/…
-
Nermin about 3 yearsWould it be more future proof to make a copy of the file, rename the copy and keep the original?
-
Hong Cheng almost 3 yearsI got this error"Could not load dynamic library 'cusolver64_11.dll'; dlerror: cusolver64_11.dll not found" and I copy the cusolver64_10.dll as cusolver64_11.dll to solve this problem. [My tensorflow-gpu version is 2.5, with cudnn64_8.dll]
-
Hong Cheng almost 3 yearsyou can look at my comments in the first answer.
-
Pythonic2020 almost 3 years@Hong Cheng, yes, I had the same experience as you with TF 2.5.0. I think that I had renamed the file to the "cusolver64_10.dll" name during previous attempts to get GPU support for TF 2.4.1. Using CUDA Toolkit 11+, use "cusolver64_10.dll" for TF 2.4.1, and keep the original file name, "cusolver64_11.dll," for TF 2.5.0.