I found I had over-provisioned some things, like using those red 4TB kingspec nvme drives in my Proxmox NUC cluster. When upgrading to Proxmox 9, I migrated everything off of one node and did a fresh install to a smaller 512GB drive so I could reclaim the 4TB.
My postgres servers have some over provisioned nvme drives, but I hesitate to touch those as they were for an Elastic Search index I swear I’m getting around to any day now!
I wanted the 4TB for a workstation that has a Radeon Pro W6800 (32GB), but only had a 512GB drive; not large enough to store most of the stable diffusion models I use on another box with an AMD R9700. I’ve had bad luck with those DDR SODIMM→DIMM adapters in the past, but I was able to get the ones photoed working. That let me push those 64GB SODIMMS in the same machine as the W6800. The SODIMM adapters are finicky, and they won’t work with my existing 32GB kit, so I guess I’ll add that to the spare pile.
nvme prices are coming down, but they’re still way higher than they were last year. I really hope I don’t need to replace anything major, because I really don’t want to buy any parts right now.
Is anyone else carefully tracking and organizing all their spares?
I do not have enough roles for all the parts that sit waiting to be used. But with the introduction of ML/LLMs, I do find it difficult to find parts that are sufficient to task. LLMs want very recent hardware features. My recent hardware is limited.
I am playing with using Vulkan to run inference, but some cards I try it on are either too old or too small.
Thankfully I consolidated just before the Ai boom from a 7 server stack of Proxmox and ceph to a single low-power idle 2950X threadripper behemoth, a partially powered mass backup of the same data, and misc SoC devices.
Sitting on all the previous setups when my health declined for a few years during this new Ai thing, I’m just now getting through the old stuff and finding gold mines like dozens of 32GB and 16GB and even decent SSDs. Even the P40s I was using for training before the Ai was released.
But I will say, the kids needed new gaming PCs and I’ve been having a rough time mixing, merging, and sourcing those parts - so I’ve had to re-inventory everything (the kids didn’t like me taking parts from their old server motherboard-now-gaming machines). But selling old gear high, and buying new/updated gear high is kinda washing things out.
It really sucks letting go of 256 GB 3200Mhz ECC, but we get a couple AM5 rigs of 32GB DDR5 and an 16GB VRAM GPU and a new WiFi 7 setup and everyone is happy. They have outgrown their childhood games and are moving into AAA titles, needing horsepower.
I have raided my stockpiles, either for my own projects or to help out friends.
I am down to 1 factory sealed Sata SSD and a few scavanged bits and pieces.
Honestly I’m a sucker, I’ve been giving away most of my extra Memory since it’s gotten so expensive. I’ve had a few friends and family in need of replacment/expansion and it’s chipped away at my pile of those black corsair vengance ddr4 dimms.
I am encouraged by everyone on here that are using their stashes altruistically help out friends and family. That does give me some warm fuzzies in this increasingly insane world.
I really don’t use local LLMs much. I get Copilot and Claude via work, but I still don’t understand how people are full on about them if they care about the code quality. Our manager left for another company and he loved the Claude and loved to merge things as I was leaving comments, so now I’m tech lead on a projects with tons of routes that don’t correctly do error checking, tests with queries copy-pasted into them and not using the actual DBO implementations (they literally test nothing) and all kids of other horrors I’m just making backlog tickets for.
I want to try some of the promising local coding models, but qwen3.6:35b-a3b will not load on either of my cards due to them only having 32GB of vram.
I mostly just use Comfy-UI to generate images. Random art, fantasy castles, anime characters, bizarre crossovers (Cause why not have Ruby Rose and Ann Takamaki fighting Gretta Thurnberg as she’s riding a dragon?)
My years of snagging budget leftovers off Ebay during good times has paid off overall. I maintain over a dozen machines for roughly half a dozen households in my circle of friends/family. Last month I dusted off components as old as a Radeon HD 7870 which is now in service on a nephew’s budget gaming rig he uses daily.
I also bought several good deals on AM4/AM5 Mobo+CPU+RAM deals between 2023 and 2025, which are almost all now running home office or gaming systems. I think I have one kit of DDR4 left, and I recently was testing a 4th gen Intel board on my bench to act as a backup in case something, somewhere dies in the meantime.
Resources are tight, and if I can avoid giving these insane, bloodsucking companies money at ridiculous prices it will have all been worth it. Hopefully I only have to buy a few things during this market.
I have spare parts for AM4 builds (CPU+RAM+SSD+PSU), but I may need a mITX board to “test” devices… as I am facing a broken 4070 with no means to test the GPU by itself at the moment.
I’m having mixed feelings in this regard. I have a small stockpile which is basically not doing anything. Stuff like dual 10GbE network cards, the motherboard, CPUs and RAM of an old server I decommissioned and a couple of SATA SSDs which are not large enough to be storage drives but not small enough to be used as (disposable) boot drives. On the other hand I’m decommissioning some of my AM4 servers in favour of a sWRX8 machine and transplanting some parts like memory and storage. The main benefit is consolidation. And then there is the new gaming server which will use some of the RAM currently in my NAS.
So there is definitely some rationing going on, but I’m renewing as well. I’m considering to sell the leftovers, instead of stockpiling the old stuff, because they won’t be very useful for much longer. Tech, however expensive it may be right now, is still moving forward. I guess all we can hope for is to have memory prices come down in time to not crash the whole retail market to failure.
Sadly I am down to basically 2x spare SSDs (lightly used) and 1X brand new spare 8TB HDD, purchased before the price explosion, in case of a disk failure in 1 of my ZFS arrays…
I will be running hardware until it LITERALLY dies - which happened recently to me, had a very old SSD die that was running my OPNsense firewall. But, because I keep backups, it was a matter of replace drive, install OPN, restore from backup and I was back online within a couple hours and taking my time to do some annual cleaning of the box.
But I think we’re all in a ‘hold and secure’ pattern for now…
I feel very ah, untrusting of Corporations & governments with how this all went. I feel they are pricing consumers out of the market so people will eventually be forced to ‘rent’ compute from evil Corps…. Which the Corps like for $ and the governments like for NO PRIVACY.
I am down to several sticks of 4gb ddr4 sodimm and a few sata ssds, everything I had went into my new NAS and new AI/gaming oculink setup. I plan on doing a review for both devices on here eventually. I am praying that sata hdds are somewhat reasonable on Prime day.
I’m in a different situation, as over the years I sold gear off to upgrade and put towards other goals. So I currently find myself with a growing interest in localizing my data and a growing disgust with how expensive storage and memory have gotten. I would like to move my data off my main rig and onto a dedicated server, but again, the cost of even doing that on a budget has increased. Back when storage was far cheaper, I just assumed that I’d have time to spec and build out - what an error on my behalf.
Other than a GPU, I don’t have parts laying around. I’ve moved most of everything to SBCs (however, even those aren’t all used or used efficiently for the ones that are).
Given all the flash shortages, I probably should write the guide for centralizing your storage and net-booting everything. That comes with its own challenges and the cost of an additional NIC (you can probably get away with a stable gigabit for the OS, but I’ll always recommend 2.5G or higher for the OS and storage access and use the built-in gigabit for the internet connection). If your mobos have 2 ports that’s perfect for netbooting.
Most computing tasks are bursty, so you can get away with a lot with a single, larger flash array. The largest problem I found in my own infrastructure is that I can’t really VLAN stuff off (netboot works with native VLAN, 'cuz it’s before the OS boots with your VLAN configs, duh!). You can still VLAN segregate your NAS, but ideally you should have a dedicated NIC for management and another 1 for storage anyway.
But that may come as an advantage because you can just buy a cheap unmanaged multi-gig switch for your storage net (and keep the internet NICs on managed gigabit switches, which should be pretty cheap).
