QM9700 to QM8700

Infrastructure & Networking Adapters and Cables InfiniBand/VPI Adapter Cards
1

Hi!. We have an existing HDR fabric based on QM8700 switches. Soon there will be a QM9700 Quantum-2 switch with a number of servers attached which is supposed to become part of our fabric.
Obviously, we will have to connect this QM9700 to our QM8700-Spine-Switches.
I have been staring at Nvidias compatibility tables (not matrices!!) and searched the net but still am unsure about the possibilities:

  • The Quantum-2 switches have 32 “sockets” (Nvidia calls that “cages”, just as silly), 800Gb/s electrically but from the start meant to be two 400Gb/s ports. Formfactor is OSFP
  • The Quantum switches have 40 ports, formfactor is QSFP56.
    Obviously, I would need a cable that connects via OSFP-plug to the QM9700 and breaks out into four 200 Gb/s QSFP transceivers on the other side.
    Nvidia seems to produce a 2x400Gb/s-OSFP to 4x200 Gb/s-OSFP cable. I won’t be able to plug this one into our spine.
    OEM seem to produce 2x400Gb/s-OSFP to 4x200 Gb/s-QSFP112 cables. I suppose that as long as the form factor is QSFP, it won’t bother the spine switches that the cables inside could actually to double the speed (112 ↔ 56) ? So this type of cable would fit?
2

Good question — this is a pretty common point of confusion when mixing Quantum and Quantum-2.

Short answer: an OSFP → QSFP112 breakout won’t work with QM8700, even though the module fits mechanically.

A bit more detail:

  • QM9700 (Quantum-2) uses OSFP ports, electrically 2×400G, and supports breakout.
  • QM8700 uses QSFP56, 200G (4×50G PAM4).
  • QSFP112 is not backward-compatible with QSFP56. Even though the form factor looks the same, QSFP112 expects 112G-per-lane signaling, which the QM8700 ports don’t support.

So a 2×400G OSFP → 4×200G QSFP112 cable won’t come up on the QM8700 side.

What you actually need is a breakout that ends in QSFP56 (200G) on the QM8700 side — i.e. OSFP on the QM9700, 4×200G QSFP56 on the QM8700, using NVIDIA-qualified DAC/AOC or optics.

The NVIDIA OSFP → OSFP breakout doesn’t help here because of the form factor, and the QSFP112 option doesn’t work electrically.

Bottom line: stick to OSFP → QSFP56 (200G) breakouts that are explicitly validated for Quantum-2 ↔ Quantum interop.