Third-Party Support for Mellanox and NVIDIA Networking Switches

A managed services provider in Bermuda contacted us about twelve Mellanox switches it needs to keep in service. Twelve is a typical number. Mellanox InfiniBand and Ethernet switches tend to arrive in fleets, behind an HPC cluster, a storage system or a virtualisation farm, and they tend to stay long after the OEM contract that came with them has expired.

Why Mellanox switches outlive their contracts

Mellanox became part of NVIDIA in 2020. The InfiniBand and Spectrum Ethernet lines continue under the NVIDIA networking brand, but older generations, the QDR and FDR InfiniBand switches and early SN-series Ethernet, have moved through end of sale and end of support. Many were also sold embedded in other vendors’ systems. Our catalog, for example, lists the Mellanox 36-port QDR InfiniBand switch that shipped inside Dell EMC Isilon clusters with an end of service life in August 2018.

The switches themselves are robust. A fixed-configuration InfiniBand switch has few moving parts: power supplies, fans, and the switch ASIC board. Failure rates are low and the failure modes are predictable. That combination, low failure rate and high replacement cost, is exactly where third-party maintenance earns its keep.

What coverage looks like

A third-party maintenance contract on a Mellanox or NVIDIA switch fleet covers:

• Hardware fault diagnosis and replacement of PSUs, fan modules and complete switch units where a board-level fault occurs.
• Spares held for the specific models in the fleet, including whole-unit spares for fixed-configuration switches, so a failure is a swap rather than a procurement exercise.
• Transceivers and cables where the contract includes them.
• Onsite response at next business day or 4 hours, with remote support 24x7.
• Coverage for 5+ years past the OEM end of service life.

Firmware is the usual caveat. After end of support, NVIDIA does not release new MLNX-OS or Onyx images for the generation in question. For an isolated fabric running a stable release that rarely matters; our engineers will tell you if the release you run has a known issue worth addressing.

Fleets, not units

Switch maintenance is priced per unit per year, and fleets attract a discount because the spares position is shared. Twelve identical switches need far fewer than twelve spare units on the shelf. The same logic applies when the switches sit alongside the servers and storage they connect: a single contract across the Mellanox fabric, the Dell or Supermicro nodes and the storage arrays costs less than three contracts and gives you one SLA and one phone number.

If you have a Mellanox or NVIDIA switch fleet coming off contract, send us the model numbers, the quantity and the site, and the response level you need. You will have an indicative price within 24 hours and a firm quote within 48.