Cuda 11.6 cudnn version?

what is the cuda 11.6 cudnn version?

25 Likes

Did you ever work this out?

2 Likes

Anyone? Which is the correct CUDNN version for CUDA 11.6? Does 8.3 work with 11.6?

1 Like

hey guys i have just tried it and it works
[cuDNN v8.3.1 (November 22nd, 2021), for CUDA 11.5] works fine with CUDA 11.6:

device_lib.list_local_devices() 2022-02-21 00:58:20.218829: I tensorflow/core/platform/cpu_feature_guard.cc:151] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations: AVX AVX2 To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags. 2022-02-21 00:58:33.987142: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1525] Created device /device:GPU:0 with 1373 MB memory: → device: 0, name: NVIDIA GeForce MX110, pci bus id: 0000:01:00.0, compute capability: 5.0 [name: “/device:CPU:0” device_type: “CPU” memory_limit: 268435456 locality { } incarnation: 1772131460318394456 xla_global_id: -1 , name: “/device:GPU:0” device_type: “GPU” memory_limit: 1440209307 locality { bus_id: 1 links { } } incarnation: 10646297292489860002 physical_device_desc: “device: 0, name: NVIDIA GeForce MX110, pci bus id: 0000:01:00.0, compute capability: 5.0” xla_global_id: 416903419 ] >>>

4 Likes

@mohamedmahdi.mechkene how did you get it to install. I get an error:
E: Version ‘8.3.2.44-1+cuda11.6’ for ‘libcudnn8’ was not found
should I just tell it to install for cuda 11.5?

hello,
did you manage to solve this error?
i installed CUDA 11.6 update 1
and iam installing cuDNN v8.3.1.22 [November 22nd, 2021] for CUDA 11.5

Yes, sorry should have put this up earlier. Download and install from the tar file.
That seemed to work ok, as you manually put the files from the tar where they belong and can ignore the version dependency.

I still don’t get why NVIDIA doesn’t make releases of CUDA and CUDNN synchronously in their repos. This shifts the complexity to the devs instead.

6 Likes

didn’t work for me … asserts in cudnnCreateConvolutionDescriptor. Had to abandon VS2022 and go back to VS2019 and Cuda 11.3 and v142 runtime

I can’t use 2022 and v142 as OneApi doesn’t show up unless you specify v143 runtime. What a mess. !

I was able to build with vs 2022 and use cudnn 8.3.3 works fine throw the cudart version warning but it still works

1 Like