How To: Install Deep Art Effects GPU

Support
#1

When you are running an NVIDIA Graphics Card, you can use our GPU optimized client!

Here you can check whether your Graphics Card is capable of CUDA

For the Deep Art Effects Desktop client to work with GPU you have to install CUDA 9.0 toolkit.

Windows:
On Windows, you’ll have to install the CUDA 9.0 Toolkit
Choose the setup for your system, download and install

Extract cudnn into cuda installation directory:
Win10: cudnn for win10
Win7 & Win8: cudnn for win7

Deep Art Effects for Desktop is ready to run on your GPU!
Download the Windows GPU version for Windows and install that.

Please note: The first time you use the Windows desktop version on GPU it may take some time to initialize the NVIDIA graphic card (up to 1 minute)

Linux:
Just install the CUDA 9.0 Toolkit. After that you have to install cuDNN. Please download it from here:

Afterward, you can install Deep Art Effects GPU for Linux.

Mac:
Unfortunately, it is currently not possible to run the GPU version on Mac.

AI initializing forever
Loading Styles... and nothing happens
Software Freezes when loading an image
Win 10 cuda9 installed Quadro M6000
RTX 3070 + GPU version
Software Freezes when loading an image
#2

Bonjour
Aucun changement.

Après installation de CUBA et de toutes les mises à jour, voici le message que j’obtiens lors de l’installation :
InstallDEEP

#3

Hello Sosten,

sorry for the inconvenience.
Is it possible that your harddrive is full?

#4

How to install cuDNN, it is only a .zip file without installer. Where do i need to extract the files to get it running ?

#5

Which operation system do you have?

#6

Owing to the lack of decent instructions for DAE I’ve been setting up some useful help pages. The web site is still a bit of a work in progress - easily fixed…

I’ve just finished installation instructions for CUDA + cuDNN for Win7+10

There’s an introductory page at https://peardox.com/gpu-acceleration/

Win7 installation - https://peardox.com/cuda-cudnn-windows-7/
Win10 installation - https://peardox.com/cuda-cudnn-windows-10/

Ubuntu 16.04 - I’ll do when I get a chance (I’ve got a load of stuff just about ready to publish and a LOAD of ideas for new ones)

#7

I installed the GPU version on my Win 10 machine, but don’t see how to start it. I only see the regular DAE CPU version. Did I have to uninstall the regular version first?
Thanks.

#8

Just install the GPU version - it (stupidly) overwrites the CPU version

Why do I consider this stupid?

Simple - unless you’ve got a GPU with a LOT of memory a large chunk size setting will crash DAE (and it won’t tell you)

There’s a messy work-around that lets you have both installed

#9

I have been following what you said to do on your website. After installing Cuda, I clicked to install the first patch. I adds files to a tmp folder and then I briefly see a flash of the NVIDIA Installer window and it goes away. I thought wow a quick install. But when I go to install patch 2 I get the image attached.
image
Do I need to worry about that? Just keep installing the patches?

#10

The ‘quick’ stuff was probably a requirement addition - like a Microsoft system DLL or whatever

#11

Thanks! It works.
The GPU version seems faster when rendering, but definitely hangs with other portion that were a lot faster previously. Like when “Art is created” message is up there.
This is using an RTX 2080.

I can’t private message on here yet. Do you have a Discord where you and other people chat about this stuff at all?

#12

You can PM here, just click on my Avatar. You can also leave comments on my website, if you want a Discord that’s https://discord.gg/DVjXwSf8h7 (there’s only me on it ATM as I have nothing big to discuss…)

RTX 2080 - plenty of cores so processing should be nippy.

I imagine you’ve got 8Gb in the thing (possibly more 6+ is usual with 8 the norm)

Here’s a gotcha. The GPU is very good at doing all the Math but you’ve gotta transfer the image to the GPU to start with (which takes time). Once it’s there the only memory the GPU can see is it’s own, essentially it’s a separate computer with 8G of memory.

AI takes a LOT of memory, if you look at server offerings (e.g. AWS etc) you can get stupid amounts of dedicated GPU RAM (128Gb, not a problem Sir - please empty your bank account into ours…)

When processing AI it creates a large number of variations of the images of various sizes then does some fancy tricks which results in your end image. The memory used here can get seriously demanding very quickly.

At some point the GPU will run out of memory (DAE responds to this by crashing and not telling you anyting’s wrong). Eventually you’ll work out its broken and kill it off / reboot.

Your PC has way more memory than your GPU - installing both the GPU + CPU versions is something worth doing. For big images that the GPU barfs at just run them thru the CPU version - it’ll take longer but it’ll get there in the end.

#13

For setting up for the GPU version of DAE can I use the latest version of the Cuda Toolkit? This is now version 11, which is supposed to be backwardly compatable. Thanks - Mike

#14

I was directed to this post by DAE support for my Nvidia 3060 issues with the GPU version of DAE but according to this post:

Cuda 9 only supports Nvidia cards that are in the 1xxx series.
Cuda 10 supports 2xxx cards and lower
Cuda 11 supports 3xxx cards and lower

I have a 3xxx series 3060 card.
Has anyone from the DAE team tested the GPU version on a 3xxx card?