Sixteen ports carry the whole fabric on a Cisco Nexus 9316D-GX, each of them a QSFP-DD cage that will run at 400, 100 or 40 Gigabit Ethernet. Behind them Cisco publishes 12.8 Tbps of switching capacity and a forwarding rate of over 4.3 billion packets per second, all of it in one rack unit. Route and MAC scale is published for the wider N9300-GX series at 896,000 IPv4 routes and 512,000 MAC addresses.
Switching capacity across sixteen ports
Cisco rates the Nexus 9316D-GX at 12.8 Tbps of switching capacity and over 4.3 billion packets per second across sixteen QSFP-DD ports, each of which will run at 400, 100 or 40 Gigabit. Cisco's word is over, and it travels with the forwarding figure wherever that figure is quoted.
Sixteen is a small port count, and it is small deliberately. Every port on this switch is an attachment into a fabric rather than a socket for a server, so the platform is sized on what it can move rather than on how many things can plug into it. That is the difference between a switch bought for an access row and a switch bought for the layer above it: one is specified on density, the other on capacity, and reading the second as though it were the first makes the port count look like a limitation instead of a design choice.
Supporting the fabric is a four-core control processor, 80 MB of system buffer, 32 GB of memory and a 128 GB solid-state disk. Management sits on two ports, one RJ-45 and one SFP, with a single USB port and an RS-232 console alongside them.
Tri-rate QSFP-DD ports and what they accept
Cisco's specification table gives these ports as 16 x 400/100/40-Gbps QSFP-DD. They are tri-rate: the same cage runs at 400, 100 or 40 Gigabit depending on the transceiver fitted and the configuration applied. Cisco's shorter marketing string states them as 400 and 100 Gigabit, which is the same port described more briefly rather than a different specification. Our display name leads on 400 Gigabit because that is the headline rate; the table above is where the full tri-rate reality is recorded.
Breakout is published for all sixteen ports and it is generous: two 200 Gigabit, four 100 Gigabit, two 100 Gigabit, eight 50 Gigabit, four 50 Gigabit, two 50 Gigabit, four 25 Gigabit and four 10 Gigabit links, with a 10 Gigabit link reached through a QSA adapter. That map is what turns a sixteen-port chassis into a much larger attachment count when the far end is slower than the port.
One conflict inside Cisco's own material is worth stating rather than smoothing over. The specification sheet publishes a four-way 25 Gigabit breakout mode for these ports, while Cisco's transceiver compatibility matrix lists no 25 Gigabit optics and no SFP28 parts against this model. Two Cisco surfaces, two answers. We publish the breakout capability as the specification sheet states it, and we do not turn that into a claim about which 25 Gigabit optics will be supplied. If a design depends on 25 Gigabit breakout on this platform, confirm the optic against the specific configuration before it is quoted.
Route and MAC scale, and the scope Cisco publishes it at
Cisco publishes 896,000 IPv4 longest-prefix-match routes and 512,000 MAC addresses, and it publishes them for the N9300-GX series rather than for this model on its own. The route table holds the prefixes the switch can forward on; the MAC table holds the endpoint addresses it can learn on the segments it terminates. A switch in the layer above an access row is sized on those two numbers as much as on its bandwidth, because a fabric that exhausts either table starts punting traffic in ways that do not show up as a bandwidth problem.
The series scope is worth reading precisely: it covers a scalability table, not a parts list. The third member of the same series, the 9364C-GX, is a two-rack-unit chassis with a different fan arrangement and shares no fan, power supply or accessory part with this switch, including a high-voltage DC option that exists in one airflow direction rather than two. Series-wide scalability figures are legitimate for all three; series-wide hardware assumptions are not. Never treat a shared series name as evidence that a part fits.
| Specification | Cisco Nexus 9316D-GX |
|---|---|
| Ports | 16 x 400/100/40-Gbps QSFP-DD, tri-rate |
| Switching capacity | 12.8 Tbps |
| Forwarding rate | over 4.3 bpps |
| Form-Factor | 1RU |
| IPv4 LPM routes, N9300-GX series | 896,000 |
| MAC addresses, N9300-GX series | 512,000 |
| Memory | 32 GB |
| SSD | 128 GB |
| CPU | 4 cores |
| System buffer | 80 MB |
| Management | 2 ports, 1 RJ-45 and 1 SFP |
| USB | 1 |
| RS-232 | 1 |
| Power supplies | 2, in a 1+1 arrangement |
| Fan trays | 5+1 |
| PSU options | NXA-PAC-1100W-PI2, NXA-PAC-1100W-PE2, NXA-PDC-1100W-PI, NXA-PDC-1100W-PE, NXA-PHV-1100W-PI, NXA-PHV-1100W-PE |
| Fans | NXA-FAN-35CFM-PI, NXA-FAN-35CFM-PE |
| Airflow | Port-side intake and port-side exhaust both offered; PI is intake, PE is exhaust |
| Accessory kit | NXK-ACC-KIT-1RU, a single part with no airflow variant |
| Dimensions, H x W x D | 1.72 x 17.37 x 25.5 in, 4.37 x 44.13 x 64.8 cm |
| Acoustics | 73.2, 81.8 and 88.8 dBA at 50, 70 and 100% fan speed |
| Typical power | 420 W |
| Maximum power | 1010 W |
| MTBF | 323,140 hours |
| Input | 100-240 VAC, 100-277 HVAC, -40 to -72 VDC, -240 to -380 HVDC, 50-60 Hz |
| Minimum NX-OS | 9.3.3 |
| Minimum ACI | 14.2(2e) |
Power supplies and fans
Six power supplies are offered across three input families: NXA-PAC-1100W-PI2 and NXA-PAC-1100W-PE2 on AC, NXA-PDC-1100W-PI and NXA-PDC-1100W-PE on DC, and NXA-PHV-1100W-PI and NXA-PHV-1100W-PE on high-voltage input. The fan modules are NXA-FAN-35CFM-PI and NXA-FAN-35CFM-PE, and the accessory kit NXK-ACC-KIT-1RU has no directional variant to choose between.
The PI and PE suffixes carry the airflow direction. PI is port-side intake and PE is port-side exhaust. They have to agree: the fan modules and the power supplies in one chassis must pull air the same way, and a chassis assembled from a mixture of the two is an invalid build rather than a suboptimal one. Decide the direction from the aisle layout before choosing anything else, because it constrains every module that follows.
These parts belong to this platform and to the Cisco Nexus 93600CD-GX beside it. They do not carry across the whole series, and they do not carry back to the earlier Nexus 9300 platforms, whose supplies and fans are rated differently and use different part numbers entirely.
How the 9316D-GX compares with the Nexus 93600CD-GX
The two 400 Gigabit models in the range answer different questions. This one puts the larger fabric behind a single tier of sixteen ports; the other spreads a slightly smaller fabric across two tiers and two cage types.
| Specification | Nexus 9316D-GX | Nexus 93600CD-GX |
|---|---|---|
| Switching capacity | 12.8 Tbps | 12 Tbps |
| Forwarding rate | over 4.3 bpps | 4.0 bpps |
| Port face | 16 x 400/100/40G QSFP-DD, one tier on one cage | 28 x 100/40G QSFP28 plus 8 x 400/100G QSFP-DD, two tiers on two cages |
| Typical power | 420 W | 586 W |
| Maximum power | 1010 W | 1071 W |
| MTBF | 323,140 hours | 295,515 hours |
Read down the columns rather than across them. This switch draws less at both the typical and the maximum figure and Cisco rates it for a longer mean time between failures; the 93600CD-GX offers more ports and a broader set of optics classes. Where the rack needs 25 Gigabit server access instead of a fabric attachment, the Cisco Nexus 93180YC-EX is the platform for that position, and the rest of the range sits on the Cisco Nexus 9000 pages.
Condition, testing and warranty
The part number and the hardware are what Cisco builds; what differs is that the unit is refurbished rather than new. Each one is tested on receipt, disassembled, cleaned, inventoried, picked, reassembled, configured, and tested again on the way out. Whole systems carry a 2-year Enterasource warranty and ship free by FedEx Ground to the lower 48 states. A Cisco warranty does not transfer with second-hand hardware, and we do not suggest otherwise.
Optics, supplies, fan modules and rails are quoted as separate lines unless a listing includes them, and the airflow pairing is confirmed on the build sheet rather than left to the packer. Software licenses and subscriptions are bought separately from the chassis and stay with the account that bought them. The wider multi-vendor category is on our AI fabric and high-speed data center switches pages, and the full catalogue of refurbished Cisco switches and routers covers everything else in the range.
Frequently Asked Questions
What does 12.8 Tbps across sixteen ports mean for fabric sizing?
It means the switch is specified on throughput rather than on attachment count. Cisco publishes 12.8 Tbps of switching capacity and over 4.3 billion packets per second behind sixteen QSFP-DD ports, and those ports are intended as fabric attachments rather than server ports. Size the platform on the aggregate the layer above your access row has to carry, then check that sixteen attachments of the rate you plan to run will physically reach the devices you need to connect. Both questions have to be answered, and the port count answers the second one.
What does a tri-rate QSFP-DD port accept on a Nexus 9316D-GX?
It accepts 400, 100 or 40 Gigabit, and Cisco publishes a breakout map on top of that: two 200 Gigabit, four 100 Gigabit, two 100 Gigabit, eight 50 Gigabit, four 50 Gigabit, two 50 Gigabit, four 25 Gigabit and four 10 Gigabit links, with 10 Gigabit reached through a QSA adapter. One caution belongs with that list. The specification sheet publishes a 25 Gigabit breakout mode while Cisco's compatibility matrix lists no 25 Gigabit optics for this model, so the two Cisco surfaces disagree and we publish neither as settled.
What do 896,000 routes and 512,000 MAC addresses actually limit?
The first limits how many distinct prefixes the switch can hold in its longest-prefix-match forwarding table, and the second limits how many endpoint addresses it can learn on the segments it terminates. Exhausting either one produces symptoms that look like a performance fault rather than a capacity fault, which is why they are sized at design time. Both figures are published by Cisco for the N9300-GX series rather than per model, and we state them at that scope because that is the scope Cisco states them at.
Is the Nexus 9316D-GX a spine switch or a leaf switch?
Read from the hardware, it is built for the spine position: a single tier of high-rate ports, a very large fabric behind them, and no dense low-rate access face at all. That is our reading of the port layout and the published capacity rather than a role Cisco assigns to the model in the material we have verified. It can certainly be deployed as a leaf where the attached devices run at 400 or 100 Gigabit, but the shape of the port face is what a spine position asks for.
How is a Nexus 9316D-GX prepared before it ships?
It goes through the same sequence as everything else we ship: tested on arrival, taken apart, cleaned, inventoried, picked against the order, rebuilt, configured to the specification agreed, and tested once more before it is packed. Whole systems come with an Enterasource warranty, stated in full on the listing, and delivery is free FedEx Ground to the lower 48 states. We publish no burn-in duration, no pass rate and no certification name, because those are claims we cannot evidence.
What do you need in order to price a Nexus 9316D-GX?
The airflow direction the rack requires, the input family for the power supplies, and the rate you intend to run the ports at, including any breakout you plan to use. Those three answers determine the modules and the optics classes that go on the build sheet, and the build sheet is what gets priced. We source this platform rather than stocking every variant, so availability is confirmed together with the price rather than assumed in advance of it.
