I see that Cudnn for Cuda 11.x has some additional files versus cuda 10.x
cudnn_8.3.1.22.exe
cudnn_8.3.2.44_windows.exe
These files are twice smaller than the zipped SDKs, is it just a matter of using a better compression algorithm?
I see that Cudnn for Cuda 11.x has some additional files versus cuda 10.x
cudnn_8.3.1.22.exe
cudnn_8.3.2.44_windows.exe
These files are twice smaller than the zipped SDKs, is it just a matter of using a better compression algorithm?
cuDNN offers two methods of install for the Windows platform. The first is through a Zip file of the library, as detailed in the install steps here: Installation Guide :: NVIDIA Deep Learning cuDNN Documentation
The second is through an executable installer (The EXE files that you have mentioned). Both methods are equivalent.
As for the size of the executables, that is due to compression (quite good compression it seems).
Hey Mr @jhalabi !
you just stated that the ZIP and EXE installation methods are equivalent right?
If so, then could you please tell me - FOR GOD’S SAKE - why does the ZIP instructions say to unzip cuDNN files in the following directories…
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\vx.x\bin
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\vx.x\include
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\vx.x\lib\x64
while instead the EXE installer will install them in
C:\Program Files\NVIDIA\CUDNN\v8.3\bin
C:\Program Files\NVIDIA\CUDNN\v8.3\include
C:\Program Files\NVIDIA\CUDNN\v8.3\lib
…thus making cuDNN unfindable by other program requiring them at expecting them at their default location?
Also, would you please explain me WHY ON EARTH when the EXE installer gets to the “Select installation location” it will not allow me to pick a location, and at the same time it will not even care to tell me where it has decided to install the files? Why, after that, I can only click “Finished”? With not even the smallest piece of information regarding where the files have been placed?
Like, do you actually enjoy creating frustration to other people ???
I would bet $ 1000 dollars that almost anyone who came to Nvidia website to install CUDA & cuDNN get sucked up in this insanely stupid flow of events:
Another example: the step where it says to install the ZLIB package.
You don’t even tell me what ZLIB is ? Why do I need it ? Where I should extract it ?
But hey, you proceed to explain me that if the download doesn’t start, then I should right click on it and click “Save as”… Like, I’m installing some bloody ML libraries… I think I’m supposed to know how to download a file, what you think??
Why are you just adding noise to the confused and misleading information you already manage to pack up?
Anyway, sure, call this a rant … but … know what? I’m really so confident that it would only take one responsible employee at Nvidia to notice this, and it would only be a matter of time before some others there will find themselves without a good job anymore…
Hello @jhalabi could you please let me know where I can find the compressed exe for more recent builds of Cudnn?
Thanks in advance