Two years working with #proxmox now, and PCIe pass-through is still an adventure every time. I absolutely could not get my ASUS Z890 mobo working reliably and stably with my nVIDIA cards or even USB controller. No two machines have the same kernel parameters.
Especially for DIY AI explorations. NVIDIA is still really the only game in town. Any iGPU isn't going to cut it there, and even for gaming the full-fat card is going to blow away anything integrated. Even with oculink...
Depends on what you need and want. TrueNAS runs really well even on lower-end hardware, memory dependent. Beyond a few disks and DIY starts to look attractive, especially on price. And you can always DIY to insanity levels, which is good. 😜
Coffee-Lake can still be viable for a proxmox node, but this is going to be on the edge of useful - probably not a bad starter, especially if you can invest a bit more and stuff it with RAM. I believe you can put up to 128GB in these. Network expanding is an issue here too.
TLDR: It depends.
😀
How much lamdba-like? 'cause you can do serverless in Kubernetes. But it ain't gonna be cheaper...
Announcing a new open source utility - PCITUI. A text terminal TUI interface to pciutils, sprovide an interactive way to explore the PCI interface of your computer or VM. Also displays IOMMU groups for virtualization.
github.com/ccebelenski/...
#homelab #selfhosting #proxmox
Yeah me too. Twice. I like that it’s flexible but I don’t have time to build custom dashboards or even maintain it.
Training my first LORA for FLUX on one of my GPU equipped proxmox nodes. It’s pulling 800W so I guess I’m well out of the low-power server zone now! #proxmox #selfhosted #homelab
Town hall with US House whip Katherine Clark, overflow seating! Be involved in our government or lose it! #townhall #katherineclark #democrats
Tried to make this as clear as I could. We're so far from the correct path right now we can't even see the path anymore.
They seem alive and well, despite it all…
Yes, depending on lots of uncontrollable factors. Certainly great for homelab, but not sufficient for many commercial uses. But I’m glad to have that capability for-sure.
Correct, it’s not really GP. Although I have heard there may be some new host/server enhancement’s coming that move it more in that direction.
Took me just under a year, but yes. 12 servers and counting.
They would, since HA is not fault tolerant. It takes some time to move the traffic. True fault tolerance is the next level.
Congrats - that's next level stuff. Next in the list - HA and hyperconverged storage? 😀 (12 node Proxmox cluster here with Ceph and three NAS's, and two AI dedicated machines with multi-GPU's... So I know the struggle!)
Also if you run something like Watchtower, make sure you set it up to do automatic cleanup. That should be made to be the default, IMO. Otherwise one day you find you have a TB of old images. No, that's not a hypothetical..😱
I'm actually kind of impressed with Lenovo engineering on their workstation machines. Just got a P920 dual Xeon 2nd gen scalable machine, and it's a tank. If I had one complaint, their flex-bay arrangement I could do without.
Not in quite a while. You do it once, and you remember not to do that again. Accidents can happen.
Three is minimum, especially for ceph. Really, five is when ceph becomes something really useful. I also don't understand people trying to do ceph on 1 GbE networking - yes, it works, but it's really annoying and slow.
Yeah, I get the complaints when I take down the net. Probably time to fix that. Maybe next month - network reworking is too much like real work.
Of those two, postgresql. This for an existing app, or something new you're working on? If the latter, I'd probably look at noSQL type databases today, rather than SQL. But it depends on the application and what you're trying to do.
Yup. TrueNAS as a VM on one of my supermicro machines. Passthrough LSI HBA in IT mode, with a NetApp disk shelf of 24 disks. Works very nicely.
Love supermicro servers... I'm up to six now. What vintage are those drives? I see the Sun branding on them, so I assume used.
I'm busy that day and can't attend. Anyone who voted for Trump is going to find out how much that mistake costs. I just wish it didn't drag the rest of us with him.
That's been a background task for me too.. I decided to do it WIKI style, which means I also need to learn that. Good times, and glad I'm not in a rush to finish it.
I'd argue past time to do that. VMWare died when Broadcom acquired them. And ESX died before then when CPU architecture shifted to non-homogeneous cores for homelabs.
Nothing - these are 17 years old now, non-ECC memory 32-bit. Basically not worth the electricity to run them. If there's any interesting cards or storage then that might have value. Otherwise they're scrap. I doubt they sell any of those CPU's at that price either. Maybe $3-$5.
