vGPU for Caffe framework

Hi All,

My requirement is we need to train 50K images using caffe framework on GPU machine.

We have NVIDIA GRID K2 GPU machine.

As per the link(https://devtalk.nvidia.com/default/topic/921233/driver-for-linux-ubuntu-for-vgpu-nvidia-grid-k1-k2) we need to have the server loaded with VMware and then set up a VM with Ubuntu.

Please suggest if we can leverage on VM approach for image processing.

Thanks
vijay

Hi vijay,

I don’t see a reason why it shouldn’t work. Please have in mind that you need to run Passthrough to have CUDA on the K2.

Regards

Simon

Thanks simon,

What is the version of CUDA needs to be installed.

Can u share the relavant doc for the same.

Thanks
vijay

I need to run Caffe application on NVIDIA GRID K2 machine. Do i need to have CUDA installed if yes, what is the version need to be installed

Thanks
vijay

Hi vijay,

I’ve never installed DL stuff on Linux therefore I don’t know if there is a “special” requirement other than the NV drivers. In Passthrough CUDA is enabled and on Windows VMs there is nothing that needs to be installed in addition to run apps with CUDA.
I don’t think there is any difference compared to a local machine so if you ever installed Caffe on a local machine you should be able to replicate your knowledge for the VM…

Regards

Simon

Hi vijay,

Sean Messay wrote nice blog containing exactly what you’re looking for:
https://thevirtualhorizon.com/2017/10/18/configuring-a-headless-centos-virtual-machine-for-nvidia-grid-vgpu-blogtober/

Regards

Simon