Hi guys!
I was given a Dell R730, 2x Xeons 2698 v4, 40 physical cores in total, 128gb RAM, and 2 Tesla P40s, which are natively vGPU capable, with the task to build a multi-purpose machine.
I am planning to (note, it is only a concept, and could change):
normal mode in PVE:
_______________________________________________________________
* use around the half of its power ( 1x CPU and 1x P40 as vGPU) to run 5 win10 VMs for day to day modelling (CAD and graphics modelling)
* use the other half of the resources (20 Physical Cores and the other P40 as vGPU) for midsized Rendering VM OR Simulation VM OR Lighter AI vm model to summarize PDF and translate libraries, on a switching basis given to the user.
_______________________________________________________________
Power mode in PVE:
_______________________________________________________________
* use all of the resources (40 cores, all RAM in, 2x P40s as vGPUs) for one VM (eveything else shutted off) for big simulations OR big rendering VM or bigger model / lighter learning
_______________________________________________________________
Extreme mode on bare metal:
_______________________________________________________________
* use the above but without PVE, or with direct pass-through for big simulations OR big rendering VM or bigger model / lighter learning
_______________________________________________________________
Is it possible to:
#1 Have 2x P40s in one VM Linux/windows for simulations or rendering as vGPUs? So far it works like a charm for Win10 in graphics mode with one Q6 Profile.
#2 is it a good idea to try AI models with vGPU, or they explcitly need pcie passthrough? if it is not, then:
#3 to have my P40s available as vGPU and direct pass-through through scripting in switching mode through API scripts?
In the end I am demanded to give the right to the user (they are small team of engineers) to actually choose which server mode he wants to use, so some other physical box will run a dashboard with a connection to the Proxmox's API, as an orchestrator, so which one of the machines palette is he needing - his normal mode in PVE, the power PVE mode, or the extreme bare metal mode.
#4 if you have understood the graduality, I would be happy to have any further tipps from you on how to set it up in a smart way, how I could save at least space (in the end it is only three machines, which are repeated but given different resources up to bare metal). Maybe, there is some resource profiling which I have missed to see so far in pve 8.3.3 and the task doesn't need that much VM implementations and the extreme bare metal mode is also useless, giving next to nothing?
Thank you once again and I love my the creation of my viennese fellows
I was given a Dell R730, 2x Xeons 2698 v4, 40 physical cores in total, 128gb RAM, and 2 Tesla P40s, which are natively vGPU capable, with the task to build a multi-purpose machine.
I am planning to (note, it is only a concept, and could change):
normal mode in PVE:
_______________________________________________________________
* use around the half of its power ( 1x CPU and 1x P40 as vGPU) to run 5 win10 VMs for day to day modelling (CAD and graphics modelling)
* use the other half of the resources (20 Physical Cores and the other P40 as vGPU) for midsized Rendering VM OR Simulation VM OR Lighter AI vm model to summarize PDF and translate libraries, on a switching basis given to the user.
_______________________________________________________________
Power mode in PVE:
_______________________________________________________________
* use all of the resources (40 cores, all RAM in, 2x P40s as vGPUs) for one VM (eveything else shutted off) for big simulations OR big rendering VM or bigger model / lighter learning
_______________________________________________________________
Extreme mode on bare metal:
_______________________________________________________________
* use the above but without PVE, or with direct pass-through for big simulations OR big rendering VM or bigger model / lighter learning
_______________________________________________________________
Is it possible to:
#1 Have 2x P40s in one VM Linux/windows for simulations or rendering as vGPUs? So far it works like a charm for Win10 in graphics mode with one Q6 Profile.
#2 is it a good idea to try AI models with vGPU, or they explcitly need pcie passthrough? if it is not, then:
#3 to have my P40s available as vGPU and direct pass-through through scripting in switching mode through API scripts?
In the end I am demanded to give the right to the user (they are small team of engineers) to actually choose which server mode he wants to use, so some other physical box will run a dashboard with a connection to the Proxmox's API, as an orchestrator, so which one of the machines palette is he needing - his normal mode in PVE, the power PVE mode, or the extreme bare metal mode.
#4 if you have understood the graduality, I would be happy to have any further tipps from you on how to set it up in a smart way, how I could save at least space (in the end it is only three machines, which are repeated but given different resources up to bare metal). Maybe, there is some resource profiling which I have missed to see so far in pve 8.3.3 and the task doesn't need that much VM implementations and the extreme bare metal mode is also useless, giving next to nothing?
Thank you once again and I love my the creation of my viennese fellows