Question How to build system to offload 'DRL GPU processing' from workstation to other box & update graphical sim on workstation based on DRL data results?

[s_quintanar]

Situation: I currently have a 13 year old workstation I built running Ubuntu 18.04 that runs well, so I don't want replace the motherboard or any other parts until they malfunction. The GPU on this computer is too old to run deep learning tasks with Webots simulations using CUDA.

Solutions 1: Build a new workstation with adequate GPU(s) where I can use the same monitors, keyboard & mouse for both computer via KVM switch. (This I know how to do)

Solution 2: Build a server with adequate GPU(s)

The problem is I don't know if I can use the server to graphically display a Webots simulation of a continuous deep learning training run on the workstation; this is the main thing I'd use the GPU server for. Is this doable? (This is the main important point)
Additionally, would I be able to access the server via ssh over wifi?

It seems needing, over time, to build workstation after workstation may not be as wise as continually modifying a single server?! Is this true?

 

[MrN1ce9uy]

What exactly are you referring to by "server?"

There's no real difference between a server and a workstation. If you are referring to Ubuntu Server and you are worried about a graphical user interface (GUI), then just use Ubuntu Desktop. It can do everything Ubuntu Server can do.

 

[s_quintanar]

What exactly are you referring to by "server?"

There's no real difference between a server and a workstation. If you are referring to Ubuntu Server and you are worried about a graphical user interface (GUI), then just use Ubuntu Desktop. It can do everything Ubuntu Server can do.

A server is a computer that can be called from a workstation using ssh (over wifi??) to provide services?!

Are you implying I can't use a server to run a graphical simulation display on a workstation? (this is what I'd like, and am trying to confirm the feasibility)

 

[MrN1ce9uy]

A server is a computer that can be called from a workstation using ssh (over wifi??) to provide services?!

You can call Ubuntu Desktop (with SSH server installed) using SSH from a workstation.

Are you implying I can't use a server to run a graphical simulation display on a workstation? (this is what I'd like, and am trying to confirm the feasibility)

No, I'm implying that you can use Ubuntu Desktop to do anything you can do on Ubuntu Server while not compromising a GUI.

 

[s_quintanar]

You can call Ubuntu Desktop (with SSH server installed) using SSH from a workstation.



No, I'm implying that you can use Ubuntu Desktop to do anything you can do on Ubuntu Server while not compromising a GUI.

Oh, I see what you're getting at; I'm not specifically referring to the Ubuntu Server software. In my case, I'm referring to a different computer(as the server) to access for displaying data and graphics on the workstation. The server can certainly run Ubuntu Desktop as you suggested.
Can I use the server(running Ubuntu Desktop and Webots and GPU-supported deep learning models) to graphically display Webots on the workstation?; ....or maybe run Webots on the workstation so it sends data to the server(so the server can compute deep learning models based on received data from workstation) and then the server sends data back to the the workstation so Webots can update the graphic simulation?

Normally, I would run a workstation with a good GPU and Webots so the simulation is updated as the deep learning model is processed.
I'm looking to offload the GPU processing on the server while maintaining a graphical display of the simulation on the workstation.

 

[USAFRet]

Oh, I see what you're getting at; I'm not specifically referring to the Ubuntu Server software. In my case, I'm referring to a different computer(as the server) to access for displaying data and graphics on the workstation. The server can certainly run Ubuntu Desktop as you suggested.
Can I use the server(running Ubuntu Desktop and Webots and GPU-supported deep learning models) to graphically display Webots on the workstation?

Change the word "server" to "system".

"Can I build a system to do X?"

 

[s_quintanar]

Change the word "server" to "system".

"Can I build a system to do X?"

Ok then, I concur. Please disregard 'server' references and replace with 'system'. I probably need to repost a whole new question with your better wording suggestion.

I modified my previous reply a few replies back to clarify my intent while you were replying...so take a quick gander.

Can I build a system to offload 'deep learning GPU processing' from a workstation to another computer while maintaining a graphical display of a simulation on the workstation based on the offloaded GPU processing data results?

Thank you for your insight! Can I end this thread and start a new one with the right question?