Planning a VM-Setup

Hardware Hub
1

What I Need:
VDI for 3 to 5 medium-weight Catia users and 10 to 20 office seats.
Current machines a stack of aging towers with T400 and T1000 GPUs and 48GB RAM for the engineers to remote into when required.

Context:

  • This is for a business setup, so no DIY-bodge-jobs or card-board engineering.
  • While FOSS is preferable, this needs to be a maintained and maintainable setup to last a few years without someone (= me) babysitting it.
  • SAN is available.
  • VMware is still set up, we are strongly considering to make our exit towards Proxmox though.
  • Sourcing from Germany or EU is required.

My Rough Idea:
I was thinking to get one medium-beefy server, stick 3 GPUs in there and pass them through to “CAD worker”-VMs, the office VMs don’t really need gpu acceleration.
I really want to avoid paying for grid licenses, so preferably AMD or Intel for the cards.
I found UDS Enterprise to enable Proxmox to do VDI service, have not had any luck on user feedback for that though.

Suggestions please!

Edit: Changed the thread title as the requirements are now the same but different.

2

There is:

Apalrd did some write ups and videos about it, don’t know if this is collides with:

this needs to be a maintained and maintainable setup to last a few years without someone (= me) babysitting it.

But might at least be worth a look, this is around for 4+ years and has regular releases, latest is 3 month old.

Write ups:

Videos:

2 Likes
3

Will look into that.

I was also told to look into Windows Server for this, I have only basic Hyper-V knowledge, so I am out of my depths here.

4

I’ve been running Dassault’s other CAD package via VDI on a couple different systems.

I’ve got a “bare metal” VDI machine that use’s AMD’s “PRO” drivers to get GPU acceleration over RDP without any virtualization. This machine’s a i5-2600 and W7500.

I’ve got another machine that uses a R9 7900 and a RTX A4000 that does use virtualization via Hyper-V and uses DDA to pass the RTX A4000 to the VM without any extra Nvidia licensing.

​​​ ​ ​

The second machine is somewhat faster (as long as the guest’s request for turbo clocks actually filter up to the host, which isn’t a given if not correctly configured) but I think that comes down to more upscale hardware being used rather than the software stack.
If you want to consolidate multiple users onto a single system you’d have to use the second route over the first; install 1 GPU for each user’s VM to avoid Nvidia’s paravirtualization licensing scheme (and retain better performance).

I’d recommend leaving VMware, their future is too uncertain and they are becoming more hostile to their smaller customers.

2 Likes
5

New Requirements:

  1. Figure out how to replace VMware environment with Proxmox
  2. Make it 99.9% available
  3. Keep it “reasonable” budget wise.
  4. [Bonus] Do the Catia VDI-dance

My main concern is point 2 honestly.
Plugging three servers into our (aging) as I found out SAN would work, and then I get to rotate that SAN out of commission end of 2027.

State of things:
We have two VM-Hosts that will age out in Q2/27.
They each have:

  • 8c/16t Intel Xeon Silvers
  • 1024 GB RAM (45% load)
  • 240 GB Raid1 for boot

SAN:

  • Dell Powerstore
  • 10-ish TB storage (60% load)

I am currently working on redoing the entire network, but did not plan for randomly catching a “our SAN is old”-stray, so 10Gig is my limit without spending more (which is a pain, but here we are).

Very rough sketches:
1) In place replacement:

  • Buy 3 new same-spec-ish Servers
  • each with 768 GB RAM (casual 40% RAM load normal, spicy 90% in fail-over state)
  • some recent-ish CPUs (Epyc 9005-series?)
  • FC cards

2) Total rebuild, in place:

  • Buy 3 same-ish Servers
  • each with 768 GB (casual 40% RAM load normal, spicy 90% in fail-over state)
  • same-ish CPUs (Epyc 9005-series?)
  • “my first Proxmox VE cluster” but with real servers
  • Intel E830 100G-cards direct-connect

3) Total rebuild, but make use of 2 server rooms:

  • Since I have 2 fully featured server rooms to play with and am about to have 40 Gbit/s between them, why not?
  • Buy 2 Switches (thinking Mikrotik CRS520 or Extreme 8720)
  • Buy 5 Servers
  • smaller-ish CPUs (Epyc 9005-series?)
  • each with 384 GB (50% normal, 60% with 1 failed host, 80% with 2 failed hosts)
  • Intel E830 100G-cards, switched

Reasons for my insanity:

  • Option 1 is semi-searchably documented, since Broadcom screwed their customers, everyone was quick to welcome VMware-nomads

  • Option 2 is copy&paste the tutorial, since the likely servers have 2x 10G onboard, I can just wire that into the existing network and am ready to run

  • Option 3 is big scary. Not because my colleagues and I will potentially end up wrangling 7 pieces of expensive gear for a month, but because this is the only setup where “Building 1 died” would potentially pass by the IT with an alert mail. Option 1 and 2 are shutdown scenarios.

Yes, we have UPSs, with a projected runtime of 20 minutes

Anyone here running a “production grade” Proxmox setup with some stories?