The MCP1650-V00AE30 is NVIDIA's part number for a 0.5 m, 200GbE QSFP56 passive copper cable. It is also the cheapest way to link two DGX Sparks at full speed. The Spark's ConnectX-7 ports top out at 200 Gb/s, which is exactly what this cable carries. The 400G-rated QSFP112 stacking cables NVIDIA names, the NJAAKK-N911 and LMTQF022-SD-R, run those same ports at the same 200 Gb/s. On a DGX Spark, the two cables deliver the same link. They differ in price, in form factor, and in whether the part number appears in NVIDIA's documentation.
Below: the full MCP1650-V00AE30 spec, a side-by-side comparison with the NJAAKK-N911, how many you need at two and three nodes, and when the more expensive cable is worth it.
Short answer: yes, it links two Sparks at full speed
MCP1650-V00AE30 is $84 at Resilient Tec — half the price of the stacking cable
The MCP1650-V00AE30 is a 0.5 m 200GbE QSFP56 passive DAC. It plugs into the DGX Spark's QSFP112 ports, and into the same ports on the ASUS Ascent GX10, Dell Pro Max GB10, HP ZGX Nano and other GB10 systems, and links at the full 200 Gb/s. The NJAAKK-N911 runs at the same 200 Gb/s on these ports; you pay more for the part number NVIDIA names.
|
MCP1650-V00AE30 — 200G QSFP56 DAC, 0.5 m Full 200 Gb/s per port. Lowest-cost Spark-to-Spark link. |
$84 · In stock at Resilient Tec Add to cart |
|
NJAAKK-N911 — QSFP112 400G stacking cable, 0.5 m Compatible with the NJAAKK-0006 NVIDIA names for DGX Spark. |
$174 · In stock at Resilient Tec Add to cart |
Resilient Tec sells the MCP1650-V00AE30 200G QSFP56 DAC for $84 and the NJAAKK-N911 compatible QSFP112 stacking cable for $174 directly at resilient-tec.com — new, lifetime warranty, shipping from Knoxville, Tennessee. Prices checked October 12, 2026. POs and net terms available.
MCP1650-V00AE30 Specifications
NVIDIA lists the MCP1650-V00AE30 as a passive copper cable for 200GbE, 200 Gb/s, QSFP56, LSZH jacket, 0.5 m, black pull-tab, 30 AWG. Here is what that means on a GB10 system:
| Specification | MCP1650-V00AE30 | What it means on a DGX Spark |
|---|---|---|
| Connector | QSFP56 both ends | Fits the Spark's QSFP112 cage, which is backward-compatible with QSFP56 plugs |
| Rate | 200GbE, 4 × 50G PAM4 | Matches the 200 Gb/s cap NVIDIA sets on each Spark port |
| Protocol coding | Ethernet (the "V" in the part number) | GB10 clustering runs over Ethernet with RoCE only |
| Type | Passive direct-attach copper | No optics, no firmware, effectively no power draw |
| Length | 0.5 m (1.6 ft) | Two units side by side, or stacked; a three-unit stack |
| Wire gauge | 30 AWG | Thin and flexible enough to dress around a desktop unit |
| Jacket | LSZH | Low smoke, zero halogen |
| Longer versions | 1 m, 1.5 m (30 AWG); 2 m, 2.5 m, 3 m (26 AWG) | Same family for switch runs and spread-out units |
MCP1650-V00AE30 vs NJAAKK-N911: Side by Side
NVIDIA's DGX Spark User Guide 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). The same page says to use cables that provide at least 200 Gb/s, and that a faster cable brings no benefit because the port is capped at 200 Gb/s. That is the whole case for the MCP1650-V00AE30:
| MCP1650-V00AE30 | NJAAKK-N911 (compatible) | |
|---|---|---|
| Form factor | QSFP56 | QSFP112 |
| Cable rating | 200G | 400G |
| Link speed on DGX Spark / GB10 | 200 Gb/s | 200 Gb/s |
| Named in NVIDIA's DGX Spark guide | No | Yes (as NJAAKK-N911 / NJAAKK0006) |
| Length at Resilient Tec | 0.5 m, with 1–3 m versions in the family | 0.5 m |
| Use beyond GB10 | Any 200GbE QSFP56 port: ConnectX-6/7 adapters, 200G switch ports | QSFP112 ports, where its 400G rating can be used |
| Price at Resilient Tec | $84 | $174 |
| Two-node cost | $84 | $174 |
| Three-node ring cost | $252 | $522 |
The two cables are the same where it counts: both are passive copper, both plug into the same port, and both run the link at 200 Gb/s. What you buy with the extra $90 is a part number that matches NVIDIA's documentation. If you ever open a support case about a cluster, "I'm using the cable in your guide" is a shorter conversation than "I'm using a 200G DAC."
Why the 400G Cable Doesn't Make the Spark Faster
The ConnectX-7 in a GB10 system connects to the SoC through two PCIe Gen5 x4 links, and NVIDIA caps each external QSFP port at 200 Gb/s. A cable cannot raise that ceiling. The NJAAKK-N911's 400G rating is headroom the port never uses. You can see the same thing in testing: Tom's Hardware reached close to the expected 25 GB/s between two Dell Pro Max GB10 systems, which is 200 Gb/s, and only after fixing a firmware mismatch on one unit. Firmware and NCCL configuration decide your throughput. The cable's rating does not.
One real-world note: each Spark port appears in Linux as two interfaces, one per PCIe x4 link. A port that is only half-loaded tops out near 100 Gb/s whatever cable you use. If you see 100G instead of 200G, see GB10 stuck at 100 Gb/s. It is a configuration problem, not a cable problem.
How Many Do You Need?
| Nodes | Topology | MCP1650-V00AE30 | NJAAKK-N911 |
|---|---|---|---|
| 2 | One cable, port to port | 1 × = $84 | 1 × = $174 |
| 3 | Ring, both ports on every node (NVIDIA-documented) | 3 × = $252 | 3 × = $522 |
| 4 | Ring with custom NCCL (works, not NVIDIA-supported) | 4 × = $336 | 4 × = $696 |
| 4+ | Through a switch (supported) | One 200G DAC per node, length to suit the switch | Rarely used — it is a short-reach part |
The MCP1650-V00AE30 handles every direct-cabled layout the NJAAKK-N911 does. For the four-node options, see Can You Build a 4-Node DGX Spark Cluster Without a Switch?
When to Pay More for the NJAAKK-N911
- Procurement or support requires the documented part. If a purchase order, an IT standard or a vendor support contract calls for the cable in NVIDIA's guide, buy that one. (If you need a genuine Amphenol or Luxshare part rather than a compatible build, ask us for a quote.)
- You want the cable to outlive the Spark. A QSFP112 cable can run at 400G in a QSFP112 port that supports it. The MCP1650-V00AE30 is a 200G cable wherever you plug it.
When the MCP1650-V00AE30 Is the Better Buy
- Price. Half the cost per link, which adds up across a three- or four-node ring.
- Length options. The MCP1650-V family runs from 0.5 m to 3 m, so a switch run later can use the same cable type. The QSFP112 stacking cable is a short-reach part.
- Reuse. It is a standard 200GbE QSFP56 cable, so it also works in ConnectX-6 and ConnectX-7 200G adapters and 200G switch ports elsewhere in a lab.
FAQ
Does the MCP1650-V00AE30 work with the DGX Spark?
Yes. The DGX Spark's two ConnectX-7 QSFP ports run Ethernet at up to 200 Gb/s each, and the MCP1650-V00AE30 is a 200GbE QSFP56 passive DAC that matches that rate. The Spark's QSFP112 cage accepts the QSFP56 plug. It is not one of the two part numbers NVIDIA names in the DGX Spark User Guide, but it links at the same 200 Gb/s, and Resilient Tec customers use it to connect Sparks and other GB10 systems.
Is the MCP1650-V00AE30 slower than the NJAAKK-N911?
No, not on a DGX Spark. The NJAAKK-N911 is a 400G-rated QSFP112 cable, but NVIDIA caps each Spark port at 200 Gb/s and states that a faster cable does not raise the link rate. Both cables run the link at 200 Gb/s. The extra headroom on the QSFP112 cable only matters in a port that runs faster than 200G.
What does the V in MCP1650-V00AE30 mean?
V marks NVIDIA's 200GbE Ethernet version of the MCP1650 QSFP56 DAC family. The H versions, such as MCP1650-H002E26, are HDR InfiniBand cables. GB10 systems cluster over Ethernet (RoCE) only, so the V version is the natural match. The rest of the part number encodes length and wire gauge: 00A is 0.5 m and E30 is 30 AWG.
How much does the MCP1650-V00AE30 cost?
The MCP1650-V00AE30 200G QSFP56 passive DAC, 0.5 m, is $84 at Resilient Tec, new, with a lifetime warranty. The NJAAKK-N911 compatible QSFP112 stacking cable is $174 at Resilient Tec. Two Sparks need one cable; a three-node ring needs three. (Prices checked October 12, 2026.)
Where can I buy the MCP1650-V00AE30?
Directly from Resilient Tec, shipping from Knoxville, Tennessee, with POs and net terms available: MCP1650-V00AE30, 0.5 m — $84 at resilient-tec.com. It is new, in stock at Resilient Tec and covered by a lifetime warranty. (Prices checked October 12, 2026.)
Should I buy the MCP1650-V00AE30 or the NJAAKK-N911?
Buy the MCP1650-V00AE30 if you want the lowest-cost 200 Gb/s link, or if you may need longer lengths later for a switch run. Buy the NJAAKK-N911 if you want the cable that matches the part numbers NVIDIA names for DGX Spark, which can make a support ticket simpler. On a GB10 system both run at 200 Gb/s.
Can I use the MCP1650-V00AE30 to connect a DGX Spark to a switch?
Yes, if the switch has a 200G-capable QSFP56 or QSFP-DD port set to 200G. At 0.5 m it only reaches a switch on the same shelf. For longer runs the MCP1650-V family continues at 1 m, 1.5 m, 2 m and 3 m, and a QSFP-DD switch with fewer ports than nodes uses an MCP7H60 breakout on the switch side instead.
Ready to Buy?
Both cables link any two GB10 systems at 200 Gb/s. New, lifetime warranty, shipping from Knoxville, Tennessee. Prices below checked October 12, 2026.
- MCP1650-V00AE30 — 200G QSFP56 DAC, 0.5 m — $84, in stock at Resilient Tec; one for two nodes, three for a ring
- NJAAKK-N911 compatible — QSFP112 400G stacking cable, 0.5 m — $174, in stock at Resilient Tec; the cable that matches NVIDIA's documented part
- LMTQF022-SD-R compatible — QSFP112 400G stacking cable, 0.5 m — $174, the same cable as the NJAAKK-N911 listing, under Luxshare's part number
- MCP1650-V001E30 — 200G QSFP56 DAC, 1 m — $97, not in stock yet; request a quote for lead time
- MCP1650-V002E26 — 200G QSFP56 DAC, 2 m — $115, not in stock yet; for switch runs
Linking three or more nodes, or a switch? Send us your node count and hardware and we will confirm exactly what to buy before you order.
You Might Also Need
- NJAAKK-N911 vs NJAAKK-0006 vs LMTQF022-SD-R — the part-number decoder for the NVIDIA-named stacking cables
- Can you cluster different GB10 systems? — the same cable links any GB10 brand to any other
- Can you build a 4-node DGX Spark cluster without a switch? — the switchless ring and when to buy a switch
- GB10 stuck at 100 Gb/s? — the configuration fixes that get you to 200G
- DGX Spark interconnect reference — part numbers and what to buy at two, three and four or more nodes