Mining rig risers: why they are the usual suspects in "random" failures
Six cards in a rig mean six risers: small boards that connect the cards to the motherboard through a USB cable. A part that costs lunch — and one of the most common sources of "magic": the card or the whole rig drops at night, with no pattern, and a reboot fixes it. Here is how a riser is built, where it fails, and how to rule it out of the suspect list. Public community experience only.
The short version (for the no-time-to-read)
- A riser is not "just a cable". It converts PCIe x16 into x1 over a long run: link quality depends on the riser chip, the cable, and the power feed.
- Riser power is the key choice. A SATA connector is rated for 4.5 A total — marginal even for mid-range cards. A 6-pin (PCIe) feed is the sane default; "it survived on SATA" is a lottery.
- A cheap USB cable is the classic cause of overnight drop-offs. When the link goes, the card vanishes from the system entirely. Everyone knows the symptom: rig alive, half the cards showing zero.
- The fix costs pennies: swap cable, power connector, reseat everything — before you touch overclocking.
- Testing a riser = swapping parts. Exchange cable/riser/slot between cards and watch whether the problem moves with the part. Free and reliable.
Anatomy of a riser and what breaks
A standard mining riser is three parts: a board with a chip and a PCIe x16 socket, a short USB cable, and a dedicated power connector. The card's signal goes through the riser chip and the cable to an x1 slot on the motherboard, while the 12 V feed comes from the separate power connector (correctly so: an x1 slot cannot power a card under load by itself).
What fails:
- poor contact in sockets. Dust, oxidation, incomplete seating of the card into the riser's x16 socket or of the riser into the motherboard's x1 slot. Typical symptom: the card appears and disappears;
- the cable. Cheap cables do not hold the signal: the link gets noisy and collapses under sustained load. You see no errors — the card simply disappears;
- power. Undervoltage or a bad contact in the power connector: the card runs but behaves unpredictably — reboots, failures at load peaks;
- the riser chip. No-name chips are consistently worse than certified solutions; the gap shows on overnight runs, not five-minute tests.
Riser power: the choice that matters
Per the SATA spec, the connector is rated about 1.5 A per contact (three 12 V contacts, ~4.5 A total) — that is the ceiling, not the operating point. Cards take tens of watts on the slot feed; a power-hungry card lives on a SATA-powered riser only with a wide margin. The community rule of thumb:
- 6-pin (PCIe) riser power is the default. Higher rated current, better contact, straight from the PSU;
- SATA is acceptable for power-light cards — and even then with thick (non-split) runs and a hand check of connector heat;
- never take 12 V for a riser from a Molex branch already carrying other devices: Molex daisy-chains are a source of voltage drops.
Connector warmth is a free sensor: warm is normal, hot (uncomfortable for a finger) is a warning. A melted riser connector is a classic cause of "a rig that reboots out of nowhere" (more in our PSU article).
How to rule the riser out: test order
Swap parts by links
Card + cable + riser + slot — four links. Put another card in the same riser: if the trouble moved with the card, it is the card; if it stayed, it is the cable–riser–slot chain.
Change one thing at a time
Replace the cable, then the riser, then consider the motherboard slot. One change per step tells you exactly who is guilty.
Check seating and dust
Reseat the card in the riser socket and the riser in the motherboard; blow out the slots. Half of "dead risers" are cured by reseating.
Read the connector by hand
Warm is fine, hot means bad contact or an overloaded run. Move the power line to another PSU branch and check again in an hour.
Why "overnight drop-off" is the riser's signature
Link failures depend on temperature (contacts expand) and on voltage sag under load. Recognition pattern: the rig is alive, the miner runs, but some cards sit at zero — or the rig reboots, and a restart "fixes" it until next time. Automation does not replace hardware swaps — it is needed for another reason: OCBot reports dropped cards in Telegram immediately, not at noon, and avoids pointless reboots when the failure yields to a lighter intervention. The rest is your swap-by-order above.
This article is general information summarizing public practice (connector current ratings, community rules of thumb); riser brands are deliberately not named. OCBot is an independent project: not affiliated with NVIDIA, HiveOS or Hiveon. Mains-level work requires care; use at your own risk — see the disclaimer.