CN6000 800G Omin-Path® SuperNIC and Ethernet SuperNIC
The first multi-protocol SuperNIC unifying Omni-Path, RoCEv2, and Ultra Ethernet at 800Gbps on PCIe 6.0 — delivering the industry’s highest performance for AI and HPC at any scale.
The first multi-protocol SuperNIC unifying Omni-Path, RoCEv2, and Ultra Ethernet at 800Gbps on PCIe 6.0 — delivering the industry’s highest performance for AI and HPC at any scale.
The CN6000 SuperNIC is the first adapter to simultaneously support Omni-Path, Ethernet RoCEv2, and Ultra Ethernet Compliance on a single silicon platform. Each port is independently configurable, allowing operators to run Omni-Path for maximum-performance HPC fabrics on one port and RoCEv2 or Ultra Ethernet on the other — connecting seamlessly to any 400G/800G Ethernet switch ecosystem without a forklift upgrade.
Powered by Cornelis’ production-proven pipeline architecture and a new Ethernet Adaptation Layer, CN6000 brings the performance DNA of Omni-Path — credit-based flow control, fine-grained adaptive routing, and sub-microsecond latency — to the world’s most widely deployed networking standard. The result: deterministic, congestion-free behavior in heterogeneous environments that previously required separate fabrics.
As AI factories scale toward gigawatt-class deployments, traditional Ethernet architectures buckle under collective communication pressure. The CN6000 800G SuperNIC eliminates that bottleneck. Delivering 1.6 billion messages per second — the highest in the industry — with sub- 1 μs MPI latency and 800 Gbps of bidirectional bandwidth, the SuperNIC keeps GPUs saturated and CPU clusters moving.
The CN6000 SuperNIC pairs with CN6000-series Omni-Path edge switches for a fully optimized end-to-end fabric, and interoperates with Arista, Cisco, Juniper, NVIDIA Spectrum, and other standards-based Ethernet end points. CN6000 SuperNIC is backward compatible with CN5000 infrastructure allowing customers to upgrade at their own pace.
Hardware-accelerated RDMA, lightweight queue pairs, and a RoCEv2 In-Flight tracking table enable the SuperNIC to maintain millions of concurrent operations at line rate, providing deterministic latency and throughput as cluster sizes grow from thousands to hundreds of thousands of nodes.
The result: Lower AllReduce completion time versus alternate solutions, higher GPU utilization — directly translating into faster training runs, lower power and cooling costs, and higher token yield per dollar.
Security is first-class citizen with CN6000 with a PQC-enabled Root-of-Trust with SPDM 1.4 (ML-DSA-87 + ECDSA P-384) that protects every deployment against both current and future cryptographic threats.
The SuperNIC optimizes real-time inferencing, shortens feedback loops, and reduces time-to-results. Its robust and efficient design makes it the ideal choice for highly parallelized applications at any scale.
CN6000 is designed around openness at every layer. As a founding contributor to libfabric — now the transport foundation of the Ultra Ethernet Consortium — Cornelis brings deep protocol expertise to Ultra Ethernet 1.0’s full feature set: Reliable Unordered Delivery (RUD), Reliable Ordered Delivery (ROD), and unreliable delivery modes.
The CN6000 PCIe 6.0 x16 SuperNIC also supports industry standard virtualization with SR-IOV, GPU-Direct for NVIDIA and AMD accelerators, and compatibility with PyTorch, NCCL, RCCL, UCX, MPI, and OpenFabrics ensure it integrates into any AI or HPC software stack. Air- and conduction-cooling options support the most demanding rack densities.
The Cornelis CN6000 Omni-Path product family is a flexible, high-performance end-to-end networking solution that addresses diverse infrastructure needs and is comprised of the Switch, SuperNIC, cables, and open-source OPX Host and Management Software. CN6000 is also interoperable with the CN5000 product family.
Industry’s highest message rate: 1.6Billion msg/s
Concurrent Omni-Path, RoCEv2 and Ultra Ethernet
1.6Tbps bidirectional (800Gbps Tx and 800Gbps Rx)
< 1 μs MPI latency
Multi-protocol – Omni-Path and RoCEv2, concurrently
Enhanced resource scaling
Optimized send and receive buffer management for high message rate applications
Supports Single Root I/O Virtualization (SR-IOV)
Supports industry standard management interfaces
Supports MSI-X interrupt handling
Configurable from one to four VLs plus one management
Configurable MTU size of 2 KB, 4 KB, 8 KB, or 10 KB VL
PCI Express Base Specification 6.0 compliant
Air and conduction cooling options
Advanced Congestion Management
Fine-Grained Adaptive Routing (FGAR)
Static Dispersive Routing (SDR)
Root of Trust secured with Post Quantum Cryptography (PQC)Management Features
Optimized for scalable MPI and SHMEM message rate and latency
Includes support for standard telemetry protocols (PLDM, MCTP, Redfish) to streamline system integration and firmware updates
Integrated management processor supports host monitoring, query, and updates. Supported interfaces include:
MCTP over PCIe
MCTP over SMBus
PLDM for Platform Monitoring and Control
PLDM for Firmware Update
PLDM for Redfish Device Enablement (RDE) support
Security Protocol and Data Model (SPDM) 1.4
| Bandwidth | 800G |
| Port Options | Dual x4 lane QSFP-DD (112 Gbps per lane) |
| PCI Low-Profile Card | 3.12 x 6.6 in. |
| Profile Card (HxL) | 7.92 x 16.76 cm |
| Tall Bracket included | Low-Profile (installed) Standard (in package) |
| Cooling Options | Air- and Conduction-Cooled |
| Power Consumption (Typ/Max) | AOC: |
| Weight | Air-Cooled: Conduction-Cooled: |
| Item Name | Item Number | Item Description |
|---|---|---|
| CN6HFA1D6ALS | 980178 | CN6000 SuperNIC, QSFP-DD, x16 PCIe 6.0, Air-Cooled, Low Profile, Single |
| CN6HFA1D6NLS | 980182 | CN6000 SuperNIC, QSFP-DD, x16 PCIe 6.0, Conduction-Cooled, Low Profile, Single |
US/Canada | NRTL 62368-1, CSA 22.2. No. 62368-1 |
Europe | EN62368-1 |
International | CB Scheme: IEC 62368-1 |
Temperature | Operating: 0 to 55 °C (derated 1 °C/175 m above 900 m) Storage: -40 to 70°C |
Humidity | Operating: 10% to 85% non-condensing Storage: 5% to 95% non-condensing |
Altitude | Operating: 0-3,200m Non-operating: 0-12,000m |
US | FCC Part 15, Subpart B, Class A |
Canada | CAN ICES-3(A)/NMB-3(A) Issue 7 |
Europe | EN55032 (CISPR32), |
Japan | VCCI, Class A |
New Zealand/ Australia | As/NZS CISPR 32, Class A |
Korea | Emissions: KS C 9832 Class A |
Taiwan | BSMI (CNS 15936), Class A |
RoHS | RoHS II Directive 2011/65/EU2, RoHS Directive 2015/863 |
REACH | REACH Regulation (EC) No 1907/2006 |
From training large language models and powering agentic AI frameworks, to real-time fraud detection in finance and computer vision in autonomous systems - CN5000 Omni-Path AI solutions deliver the high-performance, lossless networking foundation that next-generation AI demands.
Push the boundaries of scientific discovery and engineering innovation with CN5000 Omni-Path-based HPC solutions. From advancing climate modeling and seismic analysis, to accelerating simulation and design in aerospace and automotive, and enabling complex genomic research in life sciences, CN5000 Omni-Path solutions power breakthrough results—faster, smarter, and at scale.