- Jul 10, 2016
- 160
- 195
Last week I set up Deepstack for the first time, CPU version, and added the delivery models, and it worked really well. So I decided to take the next step and get the GPU version going. I am no expert in AI or 3D graphics cards but the CPU version "just worked" so I figured the GPU version would be as straightforward. It turns out it is a little more bleeding edge...
I got a new Quadro P620 card, physically installed it (no drivers) and went straight to perform these instructions for installing DS GPU. First sign of trouble was the CUDA 11.3.1. installer complaining that no compatible Visual Studio was installed. Uh-oh, nothing in the instructions about that. I went to install Visual Studio 2022 community (free version) and there were like 1000 options of what to install and none of jumped out at me as an appropriate choice. Desktop Development? Data science and analytical development? I'm not developing anything so rather than load up my perfectly working BI machine with a bunch of crap I skipped Visual Studio and continued installing CUDA without it.
Next I joined the nvdia developer program and downloaded cuDNN (v8.4.1 for CUDA 11.x). Uh-oh, there is no installer just a bunch of files. Best I could find were these instructions for installing cuDNN which I followed as best as I could except there was no C:\Program Files\NVIDIA\CUDNN\ folder on my machine so I ended up moving them to C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.3\ into the respective bin, include and lib folders based on this reddit post.
At this point, deepstack gpu does apparently start up automatically from BI. From the local machine does bring up the "DeepStack Activated" page and task manager shows it deepstack is running. However no alerts are getting confirmations. If I run an old clip and choose testing and tuning > analyze with AI, there is a long delay opening the clip with a circle of doom and when the clip finally plays nothing is recognized. Also the nvidia control panel takes a similarly long time to start up.
I'm pretty sure I screwed up something fundamental and since others seem to have this running straightforwardly, I am hoping someone can point me in the right direction.
I got a new Quadro P620 card, physically installed it (no drivers) and went straight to perform these instructions for installing DS GPU. First sign of trouble was the CUDA 11.3.1. installer complaining that no compatible Visual Studio was installed. Uh-oh, nothing in the instructions about that. I went to install Visual Studio 2022 community (free version) and there were like 1000 options of what to install and none of jumped out at me as an appropriate choice. Desktop Development? Data science and analytical development? I'm not developing anything so rather than load up my perfectly working BI machine with a bunch of crap I skipped Visual Studio and continued installing CUDA without it.
Next I joined the nvdia developer program and downloaded cuDNN (v8.4.1 for CUDA 11.x). Uh-oh, there is no installer just a bunch of files. Best I could find were these instructions for installing cuDNN which I followed as best as I could except there was no C:\Program Files\NVIDIA\CUDNN\ folder on my machine so I ended up moving them to C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.3\ into the respective bin, include and lib folders based on this reddit post.
At this point, deepstack gpu does apparently start up automatically from BI. From the local machine does bring up the "DeepStack Activated" page and task manager shows it deepstack is running. However no alerts are getting confirmations. If I run an old clip and choose testing and tuning > analyze with AI, there is a long delay opening the clip with a circle of doom and when the clip finally plays nothing is recognized. Also the nvidia control panel takes a similarly long time to start up.
I'm pretty sure I screwed up something fundamental and since others seem to have this running straightforwardly, I am hoping someone can point me in the right direction.
