CN6000 800G Omni‑Path SuperNIC and Switch
Get more from every XPU. 800G on the fabric HPC already trusts.
What is CN6000?
CN6000 Omni-Path is an end-to-end 800G fabric for HPC and performance-intensive AI: the Omni-Path SuperNIC, the CN6000 Omni-Path Switch and OPX software. The SuperNIC brings 1.6 billion messages per second, sub-microsecond MPI latency and 800 Gb/s to the lossless Omni-Path® fabric, so accelerators spend more time computing and less time waiting.1 The switch adds 48 ports at 800G and is natively designed for multi-plane fabrics.
Key takeaways:
Highest message rate: 1.6 billion messages per second and sub-1 µs MPI latency keep accelerators computing.
Lossless at gigascale: Credit-based flow control and adaptive routing hold up in fabrics of more than 500,000 endpoints.
Multi-plane by design: The CN6000 Omni-Path Switch holds two independent 400G planes, so one cable per node reaches both.
Built for disaggregated inference: Sub-microsecond latency and 1.6 billion messages per second speed KV cache handoffs and attention-FFN exchanges.
Upgrade at your own pace: Backward compatible with CN5000, with the same OPX provider and host driver.
Omni-Path performance: four headline figures for the SuperNIC and switch.
- 1.6BMessages per second, bidirectional1
- <1 µsMPI latency1
- 76.8 Tb/sFull-duplex switching bandwidth, CN6000 Switch
- 500K+Endpoints in one Omni-Path fabric
¹ Design targets. Results vary with configuration, workload and topology.
New to CN6000? The family page covers multi-plane fabric design, partner perspectives and how to buy.
The network AI and HPC has been waiting for
Omni-Path is the fabric HPC already trusts. Credit-based flow control keeps it lossless, and adaptive routing steers traffic around busy links. CN6000 brings it to 800G and can share one fabric with your CN5000 SuperNICs, so you can upgrade at your own pace.
Performance at gigawatt-AI scale
In gigawatt-class AI and large HPC systems, collective communication can stall traditional Ethernet. This SuperNIC delivers 1.6 billion messages per second, sub-1 µs MPI latency and 800 Gb/s per port. GPUs stay busy and CPU clusters keep moving, across fabrics of more than 500,000 endpoints.
Lossless and self-correcting
Credit-based flow control prevents packet loss, and link-level retry fixes errors on the link. Fine-grained adaptive routing, static dispersive routing and hardware congestion management spread traffic and ease hot spots, so long jobs keep running.
Ecosystem-ready, end to end
Pair it with the CN6000 Omni-Path Switch for a complete 800G fabric. It is backward compatible with CN5000 infrastructure and uses the same OPX provider and host driver, so you can upgrade at your own pace.
Multi-plane by design
The CN6000 Omni-Path Switch holds two independent 400G planes in one 1U switch. Each SuperNIC reaches both planes at up to 800 Gb/s, and a fault in one plane leaves the other running, so the cluster keeps working while you service it.
Secured by post-quantum cryptography
A Root of Trust with post-quantum cryptography uses SPDM 1.4 with ML-DSA-87 and ECDSA P-384. It helps protect firmware and device identity against current and future cryptographic threats.
Open and built for every environment
OPX is built on libfabric, the open interface Cornelis originally developed. PyTorch, NCCL, RCCL, UCX and MPI run on it, with GPUDirect for NVIDIA and AMD accelerators.
Which CN6000 SuperNIC fits your cluster
CN6000 offers two SuperNICs, each with two 400 Gb/s data paths on a low-profile PCIe 6.0 x16 card. The Omni-Path SuperNIC pairs with the CN6000 Omni-Path Switch. The Ethernet SuperNIC connects to standards-based Ethernet switches.
CN6000 Omni-Path SuperNICThis Page
- Best Fit
- HPC and performance-intensive AI, including disaggregated inference
CN6000 Ethernet SuperNIC
- Best Fit
- Fine-tuning and inference on Ethernet networks
Description
CN6000 Omni-Path SuperNICThis Page
- Protocol
- Omni-Path
CN6000 Ethernet SuperNIC
- Protocol
- RoCEv2; Ultra Ethernet compliant (Preview)
Description
CN6000 Omni-Path SuperNICThis Page
- Fabric
- The Omni-Path SuperNIC and the CN6000 Omni-Path Switch form an end-to-end 800G fabric, run by OPX software
CN6000 Ethernet SuperNIC
- Fabric
- Connects to the standards-based Ethernet switches you choose
Description
CN6000 Omni-Path SuperNICThis Page
- Switches
- CN6000 Omni-Path Switch
CN6000 Ethernet SuperNIC
- Switches
- Standards-based Ethernet switches
Description
CN6000 Omni-Path SuperNICThis Page
- Lossless delivery
- Credit-based flow control
CN6000 Ethernet SuperNIC
- Lossless delivery
- Priority Flow Control, 8 priorities
Description
CN6000 Omni-Path SuperNICThis Page
- Software
- OPX, which runs on libfabric (OFI)
CN6000 Ethernet SuperNIC
- Software
- Compatible with PyTorch, NCCL, RCCL, UCX, MPI and OpenFabrics
Description
CN6000 Omni-Path SuperNICThis Page
- Existing deployments
- CN5000 and CN6000 share one driver, and applications carry forward
CN6000 Ethernet SuperNIC
- Existing deployments
- Interoperates with standards-based Ethernet switches and endpoints
Description
Omni-Path SuperNIC: built for MPI, SHMEM and collectives at scale
The SuperNIC sends data only when the receiving end has room for it, fixes errors on the link itself and steers traffic around busy paths. That keeps MPI, SHMEM and collective traffic moving without packet loss.
| Omni-Path protocolLossless, credit-based transport proven at national labs. | Optimized buffer managementSend and receive paths tuned for high message rates. | SR-IOVShare one SuperNIC across virtual functions. |
| MSI-X interruptsSpread interrupt load across CPU cores. | Configurable virtual lanes1 to 4 data VLs plus 1 management VL. | Configurable MTU2, 4, 8 or 10 KB per VL. |
| PCIe 6.0 x16Host bandwidth to match 800 Gb/s. | Air or conduction coolingFit the server design you run. |
| Advanced congestion managementHardware keeps hot spots from slowing the fabric. |
| Fine-grained adaptive routing (FGAR)Steers traffic around busy links as load shifts. |
| Static dispersive routing (SDR)Spreads traffic across multiple paths to balance load. |
| Post-quantum Root of TrustSPDM 1.4 with ML-DSA-87 and ECDSA P-384. |
| Scalable MPI and SHMEMTuned for message rate and latency in HPC codes. |
| Standard telemetryPLDM, MCTP and Redfish for system integration and firmware updates. |
| Integrated management processorMCTP over PCIe and SMBus; PLDM monitoring, firmware update and Redfish Device Enablement; SPDM 1.4. |
A 1U switch, natively designed for multi-plane
The CN6000 Omni-Path Switch pairs with the Omni-Path SuperNIC for an end-to-end 800G fabric. It holds two independent 48-port 400G planes in one 1U switch. With one cable per node, each SuperNIC reaches both planes at up to 800 Gb/s, while each plane keeps the radix and low hop count of a 400G fabric.
| 48 x 800GOr 96 ports at 400G, over QSFP-DD. | Two independent planesTwo 48-port 400G planes in one switch. | Low switch latencyAs low as 105 ns. |
| Configurable virtual lanes1 to 8 data VLs plus 1 management VL. | Liquid coolingBuilt for high-density racks. | Redundant powerHot-swappable supplies with N+N redundancy. |
| Credit-based flow controlIncast-aware, so the fabric stays lossless under load. |
| Adaptive routingFine-grained adaptive and static dispersive routing balance traffic. |
| Link-level retransmissionLow-latency error recovery keeps long jobs running. |
| OpenBMC-based managementRedfish and PLDM, in-band or out-of-band. |
| CLIOver 10/100/1000BASE-T Ethernet, with NTP and LDAP support. |
| Secure bootSecure key EEPROM and protected key storage. |
| Real-time telemetryLink, congestion and buffer metrics. |
Open software that runs the fabric
Cornelis OPX is an open-source software stack for the Omni-Path SuperNIC and Switch. It spans host drivers, fabric services and switch management, so applications carry forward from CN5000 to CN6000 and on to CN7000 with a fully compatible software stack.
OPX highlights
Multi-rail: The OPX libfabric provider drives both data paths for up to 800 Gb/s per node.
Same driver as CN5000: CN5000 and CN6000 SuperNICs share one Omni-Path fabric.
Frameworks you already run: MPI, OpenSHMEM, NCCL, RCCL, UCX, PyTorch and OpenFabrics.
Built on libfabric: The open interface Cornelis originally developed.
GPUDirect: Direct RDMA placement for NVIDIA and AMD accelerators.
Specifications
The Omni-Path SuperNIC is a low-profile PCIe 6.0 x16 card with dual QSFP-DD ports and two independent 400 Gb/s data paths. The CN6000 Omni-Path Switch is a 1U, liquid-cooled switch with 48 ports at 800G. See the SuperNIC product brief and the switch product brief for full details.
| Omni-Path Switch | Omni-Path SuperNIC | |
|---|---|---|
| Performance and fabric | ||
| Protocol | Omni-Path | Omni-Path |
| Bandwidth | 800 Gb/s; 1.6 Tb/s bidirectional | 76.8 Tb/s full duplex |
| Ports | Dual x4-lane QSFP-DD, 112 Gb/s per lane | 48 x 800G QSFP-DD, or 96 x 400G |
| Architecture | Two independent 400 Gb/s data paths | Two independent 48-port 400G planes |
| Host interface | PCIe 6.0 x16 | — |
| Message rate | 1.6B messages/s, bidirectional (design target)1 | — |
| Latency | MPI latency under 1 µs (design target)1 | Switch latency as low as 105 ns |
| Virtual lanes | 1 to 4 data VLs plus 1 management VL | 1 to 8 data VLs plus 1 management VL |
| MTU | 2, 4, 8 or 10 KB | — |
| Routing | Fine-grained adaptive; static dispersive | Fine-grained adaptive; static dispersive |
| Software and management | ||
| Software | OPX (libfabric); MPI, OpenSHMEM, NCCL, RCCL, UCX, PyTorch, OpenFabrics | — |
| Accelerators | GPUDirect for NVIDIA and AMD | — |
| Management | SR-IOV; MCTP, PLDM, Redfish (RDE) | OpenBMC; Redfish, PLDM; in-band and out-of-band; CLI |
| Security | PQC Root of Trust; SPDM 1.4 | Secure boot; protected key storage |
| Physical | ||
| Form factor | PCIe low profile, 3.12 x 6.6 in (7.92 x 16.76 cm) | 1U; 1.7 x 17.2 x 26.1 in (4.4 x 43.8 x 66.3 cm) |
| Brackets | Low profile installed; standard in package | — |
| Cooling | Air-cooled or conduction-cooled | Liquid-cooled |
| Power input | — | 240 to 277 VAC or 48 VDC; N+N hot-swap supplies |
| Power (typ/max) | AOC 57.5/62.2 W; DAC 42.4/46.0 W | DAC 856/929 W; AOC 1,663/1,767 W |
| Weight | Air-cooled 0.62 lb (0.280 kg); conduction-cooled 0.52 lb (0.235 kg) | AC 38.5 lb (17.5 kg); DC 36.5 lb (16.6 kg) |
- ¹ Design targets. Results vary with configuration, workload and topology.
Ordering information
CN6000 is sold through server OEMs and channel partners in configured systems. Ask Cornelis or a partner for current availability.
| Item Name | Omni-Path SKU, air |
| Item number | Item no. |
| Description | CN6000 Omni-Path SuperNIC, QSFP-DD, x16 PCIe 6.0, air-cooled, low profile, single |
| Weight | 0.62 lb (0.280 kg) |
| Item Name | Omni-Path SKU, conduction |
| Item number | Item no. |
| Description | CN6000 Omni-Path SuperNIC, QSFP-DD, x16 PCIe 6.0, conduction-cooled, low profile, single |
| Weight | 0.52 lb (0.235 kg) |
| Item Name | CN6SWE48G2WF |
| Item number | 980245 |
| Description | CN6000 Omni-Path Switch, 48- port, 240/277 VAC, dual power supply, liquid-cooled, fan-to- port airflow |
| Weight | 38.5 lb (17.5 kg) |
| Item Name | CN6SWE48G2WP |
| Item number | Item no. |
| Description | CN6000 Omni-Path Switch, 48- port, 240/277 VAC, dual power supply, liquid-cooled, port-to- fan airflow |
| Weight | 38.5 lb (17.5 kg) |
| Item Name | CN6SWE48D2WF |
| Item number | Item no. |
| Description | CN6000 Omni-Path Switch, 48-port, 48 VDC, dual power supply, liquid-cooled, fan-to-port airflow |
| Weight | 36.5 lb (16.6 kg) |
Safety, environment and operating conditions
Use these listings to confirm the SuperNIC and switch fit your site conditions and your region's regulatory requirements before you plan a deployment.
| Temperature0 to 55 °CSuperNIC, operating10 to 35 °CSwitch, operatingBoth derate 1 °C per 175 m above 900 m; storage -40 to 70 °C | Humidity10% to 85%SuperNIC, operating5% to 85%Switch, operatingStorage 5% to 95%; non-condensing | Altitude0 to 3,200 mSuperNIC, operating0 to 3,000 mSwitch, operatingNon-operating 0 to 12,000 m |
| US and Canada | NRTL 62368-1; CSA 22.2 No. 62368-1 |
| Europe | EN62368-1 |
| International | CB Scheme: IEC 62368-1 |
| RoHS | RoHS II Directive 2011/65/EU; Directive 2015/863 |
| REACH | Regulation (EC) No 1907/2006 |
| US and Canada | NRTL 62368-1; CSA 22.2 No. 62368-1 |
| Europe | EN62368-1 |
| International | CB Scheme: IEC 62368-1 |
Frequently asked questions
-
What is the CN6000 Omni-Path SuperNIC?
It is an 800G network adapter for AI and HPC that runs the Omni-Path fabric. It is designed to deliver 1.6 billion messages per second and sub-microsecond MPI latency on a lossless fabric. -
What is the CN6000 Omni-Path Switch?
It is a 1U, liquid-cooled switch with 48 ports at 800G, or 96 at 400G, and 76.8 Tb/s of full-duplex bandwidth. It is natively designed for multi-plane fabrics: two independent 400G planes let one cable per node reach both. -
Is there a CN6000 SuperNIC for Ethernet networks?
Yes. Cornelis also offers the CN6000 Ethernet SuperNIC, which runs RoCEv2 on standards-based Ethernet switches and is Ultra Ethernet compliant (Preview). The CN6000 Omni-Path Switch is Omni-Path only. -
How does CN6000 Omni-Path help disaggregated inference?
Disaggregated inference splits prefill from decode, and attention from FFN or expert layers, onto separate nodes. KV cache then moves from prefill to decode for each request, and small activations move between attention and FFN layers for each token. CN6000 Omni-Path is designed to bring sub-microsecond latency, 1.6 billion messages per second and 800 Gb/s per node to that traffic. Credit-based flow control keeps the fabric lossless, so no transfer waits on a retransmit. -
Does it work with my existing CN5000 fabric?
Yes. It is backward compatible with CN5000 infrastructure and uses the same OPX provider and host driver, so CN5000 and CN6000 SuperNICs can share one Omni-Path fabric. Applications carry forward to CN7000 with a fully compatible software stack. -
Which switches does it connect to?
Connect it to the CN6000 Omni-Path Switch for a full 800G dual-plane fabric. It also works with existing CN5000 edge and director-class switches. -
Which software runs on CN6000 Omni-Path?
Cornelis OPX, an open-source stack built on libfabric, runs the SuperNIC and the switch. MPI, OpenSHMEM, NCCL, RCCL, UCX and PyTorch run on it, and multi-rail drives both data paths for up to 800 Gb/s per node.