Skip to content

Question about cross-shipping RMA when you're your own datacenter

Dedicated Servers by wendy 24 replies 3.9K views
#1

So I have a small colo footprint, three Windows Server boxes running Hyper-V, and one of them just threw a predictable failure on an NVMe boot drive. No vendor support contract because I bought them secondhand. The big vendors will not advance-ship without a contract or a credit card hold, but I am trying to avoid a 72-hour round trip to the datacenter to swap a dead drive myself. Has anyone successfully negotiated a cross-ship RMA with just a card hold? Or is there a parts broker people trust for same-day to a datacenter address? I know the Linux crowd here mostly runs their own spare parts bins, but I am trying to keep my inventory lean.

Have you tried restarting it?
11 #2

The problem is the vendor will want the enterprise contract number, no? Without this, they make you ship first, then wait, then they ship. The host will be down for days.

I try with Contabo once, they accept the card hold but only for their own branded SSDs. For Supermicro parts, no luck.

seedbox, NAS, tape, and three offsite
#3

For what it's worth, this is not an email deliverability question, but I will answer anyway because the logistics overlap with infrastructure reliability, which affects MX record uptime.

Hot take: your real problem is not the RMA workflow, it is that you have no cold spare on site. A single NVMe boot drive with no mirror and no local replacement is a single point of failure for your mail gateway or whatever else runs there. SPF records do not help you when the machine is offline.

That said, some manufacturers will take a credit card hold over the phone if you have the original serial number and can prove ownership. The hold is usually $200-400 for a consumer-grade NVMe, more for enterprise. They ship overnight to the datacenter attention your cage number. I have done this with a major disk vendor whose name rhymes with "estern digital" but I will not name them because forum rules.

For

SPF, DKIM, DMARC — holy trinity ✉️
#4

I have dealt with this on six machines in my own colo. The answer depends entirely on the manufacturer and whether you registered the serial at purchase.

Some vendors will accept a corporate card hold. Others will not speak to you without a support contract tied to the chassis serial. In my experience, the hold amount is typically 120 percent of MSRP, released upon receipt of the failed part. I have had success with KnownHost parts through their advance replacement program, though they required a faxed form. This was in 2024 and their process may have changed — I used https://www.knownhost.com at the time.

I keep a small stock of machines in various states of decommission specifically for cannibalization. It is not elegant. It is practical. The 45-minute drive to the datacenter is unavoidable if you are serious about self-managing hardware.

-- Marcus

#5

Ngl this is why I stopped doing bare metal for anything that is not GPU training

For inference boxes I just rent from OVHcloud or whoever has the H100s this week. If a drive dies it is their problem. I do not even know what city the machine is in.

But for your situation, some of the AI homies on KnownHost discord said they got cross-ship on a motherboard by just giving a card number and a cage address. Ymmv though

CUDA cores are my love language
4 #6

Quick clarification on forum policy: naming specific vendor processes is fine, but please do not post direct phone numbers or internal escalation codes in public threads. DM those if you want to share.

To summarize what has worked for members here without contracts:

  • Card hold, original serial, ship to cage attention: sometimes works
  • Third-party parts brokers (Hetzner, Leaseweb spare program): faster, you pay
  • Cold spare on site: universally recommended, including by me

I will add a note to the wiki about this. If anyone wants to write up a comparison of advance replacement policies across vendors, I will sticky it.

— Admin

sudo make me a sandwich
#7

Which datacenter region is your cage in. Mumbai has brokers too.

#8

Contabo took a card hold for third-party Supermicro boards, or only their SSDs.

seedbox, NAS, tape, and three offsite
7 #9

To answer robertdle: my cage is in Manchester, not Mumbai. I would love to have the broker density of India but I am stuck with what the M62 corridor offers.

hankels said:
Contabo took a card hold for third-party Supermicro boards, or only their SSDs.

Only their SSDs, you said that the first time. I am trying to figure out if anyone has cracked this for Supermicro specifically. The chassis is X11 generation, if that matters to anyone's experience.

Have you tried restarting it?
#10

I am in Rotterdam and I use Leaseweb for colocation. Their spare parts program is real but you need to be a colo customer, not just a dedicated renter. I have gotten same-day drive swaps by calling their Amsterdam NOC directly.

For your situation wendy, since you are in Manchester, Leaseweb London might be close enough for a courier if you can find a broker who will hand-carry. I have done this with a bike courier from Amsterdam to Frankfurt for a RAID card. Cost about 200 EUR but the downtime was four hours instead of three days.

Post a reply

You need an account to reply. Log in or register to join the conversation.

Post reply Preview Save draft