ROS + Machine learning build help

Hello fellow enthusiasts,

I am hoping if you could help me plan to build a workstation for robotics+machine learning research(+personal) purposes along with some self hosted services. I have been reading forums, articles, benchmarks to get to know about the hardware/software possibilities, pros and cons, suitability and so on. So far, below is the specifications i could come up with:

Ryzen 7 3700x or Ryzen 3900x (if price difference is ~<100 EUR)
Be Quiet! Dark rock pro4
ASRock X570 Taichi (90-MXBA83-A0UAYZ)
2x Kingston Server Premier DIMM 32GB, DDR4-2666, CL19-19-19, ECC (KSM26ED8/32ME) or 2x Samsung DIMM 32GB, DDR4-2666, CL19-19-19, ECC (M391A4G43MB1-CTD)
RTX 3080 or 2080Ti (hoping present cost hype settles down by March’21)
2x Samsung SSD 970 EVO 500GB, M.2 (MZ-V7E500BW)
2x Seagate IronWolf Pro NAS HDD 2TB, SATA 6Gb/s (ST2000NE001)
2x Western Digital WD Red Plus 2TB, SATA 6Gb/s (WD20EFRX)
Be Quiet! Silent Base 801
Be Quiet! Dark power 11 750w

Note:
I am in a region with 28cent/kWh. So saving energy costs is a high priority, followed by overall stability.
To be placed in living room.
No overclocking any parts and No gaming is intended.

Main tasks: Using ROS (Robot Operating System) in Debian for robotics research, data analysis and visualization. Some ROS docker nodes is also possible. Machine learning frameworks for time series analysis.

Apart from this, i would like to have a self hosted gitea, Wireguard/Openvpn incase i have to work away from home (5 to 6 days a month assuming everything is back to normal), virtualize pfsense, some docker instances for nextcloud, etc.,

If headless operation is possible then i would prefer that as some of the calculations takes overnight or days to complete. Hence i am curious if Proxmox is a good option. This way i can run VMs, dockers and also keep a windows VM handy for ‘must’ needs like Solidworks, Fusion and some windows only applications.

  1. Does using Proxmox as host makes sense for this kind of setup ? or Is there any simpler alternatives i should look into?

  2. If proxmox is ok to use, then is using 2*M.2 as ZFS mirror for proxmox host os, VM ISOs and VM guests a good option ? I hear mixed information about placing all in this way is not good for M.2s and will require quicker replacement. So, i am pretty confused on this point.

  3. I am thinking of ZFS mirror for HDDs. Is there any better way to do ?

  4. Going Threadripper way seems costly and i might not need it as most of the works will be offloaded to GPUs. I can also use work place clusters for larger calculations. If I am wrong on this and I should look into Threadripper options, then let me know.

  5. Does anyone has experience with Samsung RAM listed above in combination with the Taichi board ? I see Kingston RAM is listed in QVL but not this Samsung RAM. Not sure whether it will work without any problems.

I planning to finish building this system by end of February or mid March. So, i guess i have time to explore more options and then decide. I can expand to second GPU in the future but not in next two years. Do let me know if I have missed any information and thanks for helping me out !

This topic was automatically closed 273 days after the last reply. New replies are no longer allowed.