Two Dell Pro Max GB10s, two HP ZGX Nano AI Stations, or one of each, connect with one QSFP cable between their ConnectX-7 ports. Both are built on NVIDIA's GB10 reference board with the same ConnectX-7 NIC and the same two QSFP ports as the DGX Spark. The cable, the network setup and the NCCL configuration are the same as for two Sparks. HP's product page says it plainly: pairing two ZGX Nanos requires a compatible QSFP cable, sold separately.
Below: which cable to buy, the setup, the firmware problem that caught Tom's Hardware out on a pair of Dell Pro Max units, and the few things Dell and HP do differently from NVIDIA.
Short answer: one QSFP cable, same as two DGX Sparks
Dell Pro Max GB10 and HP ZGX Nano pair with one cable — $174 or $84 at Resilient Tec
Plug one 0.5 m QSFP passive copper cable into a ConnectX-7 QSFP port on each unit and follow NVIDIA's two-node setup. Dell + Dell, HP + HP and Dell + HP all work the same way, at the full 200 Gb/s the ports support. Update firmware on both units from the command line before you test.
|
NJAAKK-N911 — QSFP112 400G stacking cable, 0.5 m Compatible with the NJAAKK-0006 NVIDIA names for GB10 systems. |
$174 · In stock at Resilient Tec Add to cart |
|
MCP1650-V00AE30 — 200G QSFP56 DAC, 0.5 m Full 200 Gb/s at the lowest cost. |
$84 · In stock at Resilient Tec Add to cart |
Resilient Tec sells the NJAAKK-N911 compatible QSFP112 stacking cable for $174 and the MCP1650-V00AE30 200G DAC for $84 directly at resilient-tec.com — new, lifetime warranty, shipping from Knoxville, Tennessee. Prices checked October 12, 2026. POs and net terms available.
Dell Pro Max GB10 and HP ZGX Nano at a Glance
| Dell Pro Max with GB10 | HP ZGX Nano G1n AI Station | |
|---|---|---|
| Board | NVIDIA GB10 reference | NVIDIA GB10 reference |
| Cluster NIC | ConnectX-7, 2 × 200G QSFP | ConnectX-7, 2 × 200G QSFP |
| OS | NVIDIA DGX OS | NVIDIA DGX OS |
| Clustering cable | Budget for one separately | Sold separately (per HP) |
| Vendor extras | 280 W USB-C adapter, front power LED, Dell support contracts | HP ZGX Toolkit (VS Code extension, runs from an x86 client) |
| What it pairs with | Any GB10 system | Any GB10 system |
None of the differences in that table touch the network. Each vendor's chassis, power adapter, SSD and software extras differ, but the ConnectX-7 ports the cable plugs into are identical.
Which Cable to Buy
NVIDIA's documentation for the GB10 platform names two approved cables: the Amphenol NJAAKK-N911 (400 mm, with NJAAKK0006 as its 0.5 m version) and the Luxshare LMTQF022-SD-R (400 mm). It also says each port is capped at 200 Gb/s and a faster cable brings no benefit. Your choice is between the cable type NVIDIA names and a 200G cable at half the price. On a Dell or HP GB10, both run at the same speed:
| Cable | Form factor | Link on Dell / HP GB10 | Price at Resilient Tec |
|---|---|---|---|
| NJAAKK-N911 compatible, 0.5 m | QSFP112, 400G-rated | 200 Gb/s | $174 |
| MCP1650-V00AE30, 0.5 m | QSFP56, 200G | 200 Gb/s | $84 |
If a Dell or HP support contract is part of why you bought these units, the QSFP112 cable is the easier one to explain in a ticket. Otherwise the 200G DAC does the same job. Full comparison: MCP1650-V00AE30 vs NJAAKK-N911.
How to Connect Them
-
Update both units first, from the command line. Run the DGX OS updates, then check
aptandfwupdfor anything held back, install it, and reboot. Don't rely on the dashboard alone (see the firmware trap below). - Plug in the cable between a QSFP port on each unit, pull-tab up. All ports on the Dell Pro Max are on the rear panel; check each vendor's manual for which QSFP port is which, since rear-panel layouts differ from the DGX Spark.
- Configure the link. Follow NVIDIA's "Connect Two Sparks" playbook: apply its netplan file on both units, or use NVIDIA Sync's Cluster Assistant to discover the units and apply the ConnectX-7 settings for you.
- Set up passwordless SSH between the units, with the same username on both.
- Run the NCCL bandwidth test from NVIDIA's playbook. Expect close to 25 GB/s (200 Gb/s).
The Firmware Trap
This is the problem most likely to cost you an afternoon. When Tom's Hardware clustered two Dell Pro Max GB10s, the first NCCL test returned only a small fraction of the expected bandwidth. The DGX Dashboard reported both systems fully up to date. On the command line, one unit had an fwupd update stuck behind a phased rollout. Once it was force-installed, a new round of firmware updates appeared. After applying those and rebooting, the link ran at close to the full 25 GB/s and stayed there across reboots.
HP owners see a version of the same problem. One ZGX Nano owner on NVIDIA's forums reported their unit's firmware running about two months behind the DGX Spark it was clustered with. Firmware covers the ConnectX-7 and the SoC, so a mismatch looks exactly like a bad cable. Check firmware on both units before you suspect the cable.
Two more causes of a half-speed link:
- Only one interface per port in use. Each QSFP port appears as two Linux interfaces, one per PCIe Gen5 x4 link. Both must carry traffic to reach 200 Gb/s. See GB10 stuck at 100 Gb/s.
- MTU mismatch. Set 9000 on the QSFP interfaces of both units.
One Dell and One HP
A mixed Dell + HP pair works the same way, because the cable, NIC and network layer are identical. Two things are different from a same-brand pair. First, software parity: Dell and HP ship their own builds of DGX OS on their own update schedules, so line up the OS, firmware, CUDA and NCCL versions by hand. Some owners reimage partner units with NVIDIA's DGX Spark recovery image so every node matches; check your warranty terms first. Second, support: each vendor supports its own box, so keep a single-unit reproduction ready for any ticket. More in Can You Cluster Different GB10 Systems?
Going to Three or Four
Three Dell or HP GB10s connect in a ring of three cables with no switch, the same layout NVIDIA documents for three Sparks. See How to Connect Three DGX Sparks for the port-by-port wiring. At four, the supported route is a 200G-capable switch. A four-cable ring with a custom NCCL build also works, outside NVIDIA support.
FAQ
What cable connects two Dell Pro Max GB10 systems?
One QSFP passive copper cable (DAC) between the ConnectX-7 QSFP ports on the back of each unit. The Dell Pro Max GB10 uses the same NVIDIA GB10 board and ConnectX-7 NIC as the DGX Spark, so the cables NVIDIA approves for the Spark (Amphenol NJAAKK-N911 and NJAAKK0006, Luxshare LMTQF022-SD-R) are correct. A 200G QSFP56 DAC such as the MCP1650-V00AE30 also links the pair at the full 200 Gb/s.
What cable connects two HP ZGX Nano AI Stations?
The same one. HP says connecting two ZGX Nano systems requires a compatible QSFP cable, sold separately. The ZGX Nano has the same ConnectX-7 NIC and two QSFP ports as every GB10 system, so the NVIDIA-approved QSFP112 stacking cables and 200G QSFP56 DACs both work, at 200 Gb/s.
Can I connect a Dell Pro Max GB10 to an HP ZGX Nano?
Yes. Both are built on NVIDIA's GB10 reference board with the same ConnectX-7 NIC and QSFP ports, so they connect with the same cable and setup as two units of one brand. Dell and HP each support only their own system, and we have not seen a public report of this exact pair yet, though Dell and HP units have each been clustered with other GB10 brands. Put the same DGX OS, firmware and NCCL versions on both before you benchmark.
How much does the cable to connect two GB10 systems cost?
At Resilient Tec, the NJAAKK-N911 compatible QSFP112 stacking cable, 0.5 m, is $174, and the MCP1650-V00AE30 200G QSFP56 DAC, 0.5 m, is $84. Both are new with a lifetime warranty and link two Dell Pro Max GB10 or HP ZGX Nano systems at 200 Gb/s. (Prices checked October 12, 2026.)
Where can I buy a QSFP cable for the Dell Pro Max GB10 or HP ZGX Nano?
Directly from Resilient Tec, shipping from Knoxville, Tennessee, with POs and net terms available. The NJAAKK-N911 compatible stacking cable is $174 at resilient-tec.com and the MCP1650-V00AE30 200G DAC is $84 at resilient-tec.com, both new with a lifetime warranty. (Prices checked October 12, 2026.)
Why is the link between my two GB10 systems slow?
Usually firmware. When Tom's Hardware clustered two Dell Pro Max GB10s, the first NCCL test showed a fraction of the expected bandwidth even though the dashboard said both units were up to date. One unit had a firmware update held back behind a phased package; after installing it from the command line, the link reached close to 25 GB/s (200 Gb/s). Also check that both interfaces of the port carry traffic and that MTU is 9000 on both units.
Can I connect three Dell Pro Max GB10 or HP ZGX Nano systems?
Yes, in a ring of three cables with no switch, using both QSFP ports on each unit. It is the same three-node layout NVIDIA documents for the DGX Spark. Four or more units go through a 200G-capable switch.
Ready to Buy?
The same parts link any Dell, HP or other GB10 system. New, lifetime warranty, shipping from Knoxville, Tennessee. Prices below checked October 12, 2026.
- NJAAKK-N911 compatible — QSFP112 400G stacking cable, 0.5 m — $174, in stock at Resilient Tec; one for two units, three for a ring
- LMTQF022-SD-R compatible — QSFP112 400G stacking cable, 0.5 m — $174, the same cable under Luxshare's part number
- MCP1650-V00AE30 — 200G QSFP56 DAC, 0.5 m — $84, in stock at Resilient Tec; full 200 Gb/s at the lowest cost
- MCP7H60-W002 — 400G QSFP-DD to 2×200G QSFP56 breakout, 2 m — $179.45, switch side only, one per pair of units at four or more
Buying for a team, or going past two units? Send us your models and node count. POs and net terms available.
You Might Also Need
- How to connect three DGX Sparks — the three-node ring, port by port
- How to connect two ASUS Ascent GX10s — the same setup on ASUS hardware
- MCP1650-V00AE30 vs NJAAKK-N911 — which cable to choose
- Can you cluster different GB10 systems? — mixing Dell, HP, ASUS and NVIDIA units
- mlxlink and mst field guide — diagnosing a ConnectX-7 link that won't come up