ConnectX-8 SuperNIC, 800 Gb/s
NVIDIA ConnectX-8 SuperNIC with 800 Gb/s InfiniBand XDR or Ethernet, PCIe Gen6 x16 and an OSFP port, for servers on Quantum-X800 and Spectrum-X fabrics.
- Speed
- 800 Gb/s, InfiniBand XDR or 800 GbE, one OSFP port; 2 x 400 Gb/s configurations available
- Host interface
- PCIe Gen6 x16, with multi host and socket direct options
- Offloads
- RDMA, GPUDirect RDMA, Spectrum-X adaptive routing and congestion control, in network computing
- Form factor
- PCIe card with short or tall bracket, passive heatsink
- Use
- One per GPU pair in PCIe servers, one per GPU in HGX class systems
Overview
ConnectX-8 is the SuperNIC generation that matches Blackwell Ultra. It provides 800 Gb/s of InfiniBand XDR or Ethernet bandwidth over a PCIe Gen6 x16 host interface, and it is the endpoint for both Quantum-X800 InfiniBand and Spectrum-X Ethernet fabrics, with the adaptive routing and congestion control features those platforms depend on.
In HGX B300 servers the ConnectX-8 devices are integrated on the baseboard. The add in card version is for PCIe GPU servers, CPU and storage nodes that need an 800 Gb/s port, and for upgrades of existing systems to the new fabric speed. Port splitting lets one adapter present two 400 Gb/s ports for sites that are still on NDR or 400 GbE.
We supply ConnectX-8 as an add in card with the correct bracket, as part of server configurations, and with matching OSFP optics and cables. Tell us the server model and PCIe slot layout and we confirm fit, airflow and firmware before shipment.
Use cases
- 800G GPU fabric endpoints
- Storage servers at 800 Gb/s
- Upgrading PCIe GPU servers
- CPU nodes on XDR InfiniBand
Specifications
| Speed | Up to 800 Gb/s per adapter |
|---|---|
| Protocols | InfiniBand XDR, NDR and HDR; Ethernet 400 GbE and below per port, 800 Gb/s end to end with compatible platforms |
| Ports | OSFP or QSFP112 module, port splitting supported |
| PCIe | Gen6 x16 edge connector; Gen5 auxiliary card for socket direct and multi host |
|---|---|
| Offloads | RDMA and RoCE, GPUDirect RDMA, GPUDirect Storage, NVIDIA SHARP in network computing |
| Security | Hardware root of trust, secure firmware update |
| Form factor | PCIe add in card, short or tall bracket |
|---|---|
| Cooling | Passive heatsink, requires server airflow validated for the adapter |
| Cables | 800G OSFP transceivers and cables, see the transceiver product |
Figures describe the system class. Exact configuration, weights and power draw are confirmed on the quotation.
What's included
- Original manufacturer part in retail or bulk packaging
- Compatibility check against your server or workstation model
- Manufacturer warranty passed through, advance replacement options
- Short and tall brackets
Compliance
This product is not a data center GPU system and does not require NVIDIA end user registration. Standard export documentation applies. Read about export compliance.
Related
Related systems
ConnectX-7 400 Gb/s adapter
- Speed
- 400 Gb/s single port, or 2 x 200 Gb/s
- Host interface
- PCIe Gen5 x16
- Offloads
- RDMA and RoCE, GPUDirect RDMA
- Form factor
- PCIe HHHL card with OSFP or QSFP112 port
800G Quantum-X800 InfiniBand switch
- Ports
- 144 x 800 Gb/s XDR InfiniBand through 72 OSFP cages
- Throughput
- 115.2 Tb/s aggregate switching
- In network computing
- SHARP fourth generation, adaptive routing
- Form factor
- 4U, air cooled
800G OSFP transceivers and cables
- Transceivers
- 800G OSFP, twin port 2 x 400G
- Copper
- 800G OSFP active and passive copper cables for in rack and adjacent rack links
- Optical
- 800G active optical cables and breakout cables to 2 x 400G
- Fiber
- MPO and LC patch cords
Compare it with the next class up or down
The nearest classes we quote are the ConnectX-7 400 Gb/s adapter and the 800G Quantum-X800 InfiniBand switch. If neither is quite right, tell us the workload and we specify the system between them.
- (858) 613-1818
- sales@computermodules.com
- Monday to Friday, 8am to 5pm Pacific