ConnectX-7 400 Gb/s adapter
NVIDIA ConnectX-7 adapter with 400 Gb/s InfiniBand NDR or 400 GbE, PCIe Gen5 x16, single or dual port, the standard fabric and storage adapter we fit.
- Speed
- 400 Gb/s single port, or 2 x 200 Gb/s, InfiniBand NDR or Ethernet
- Host interface
- PCIe Gen5 x16, socket direct and OCP 3.0 variants
- Offloads
- RDMA and RoCE, GPUDirect RDMA, GPUDirect Storage, hardware timestamping
- Form factor
- PCIe HHHL card with OSFP or QSFP112 port, or OCP 3.0 mezzanine
- Use
- GPU pair adapter in PCIe servers, fabric port for CPU and storage nodes, frontend on HGX systems
Overview
ConnectX-7 is the 400 Gb/s adapter that appears throughout this catalog: one per GPU pair in the PCIe GPU servers, the fabric port on CPU and storage nodes, and the frontend option on HGX systems. It supports InfiniBand NDR and 400 GbE on the same silicon, so one part covers both fabric choices.
GPUDirect RDMA and GPUDirect Storage let data move between the network and GPU memory without passing through the host, which is what makes a PCIe GPU server a credible cluster node. RoCE v2 with hardware congestion control, NVMe over Fabrics offload and inline crypto make it equally useful on storage servers and in multi tenant environments.
We supply it as a PCIe card in single and dual port versions and as an OCP 3.0 mezzanine for servers with that slot, with optics and cables matched to your switches. For new 800 Gb/s fabrics choose ConnectX-8; for existing NDR and 400 GbE networks ConnectX-7 remains the standard.
Use cases
- PCIe GPU server fabric adapter
- Storage server network ports
- CPU node fabric ports
- 400 GbE frontend on HGX servers
Specifications
| Speed | 400 Gb/s per adapter, 1 x 400 Gb/s or 2 x 200 Gb/s |
|---|---|
| Protocols | InfiniBand NDR, HDR and EDR; Ethernet 400, 200, 100, 50 and 25 GbE |
| Ports | OSFP or QSFP112, single or dual port |
| PCIe | Gen5 x16, socket direct option across two x8 links |
|---|---|
| Offloads | RDMA, RoCE v2, GPUDirect RDMA and Storage, NVMe over Fabrics, TLS and IPsec crypto |
| Security | Hardware root of trust |
| Form factor | HHHL PCIe card or OCP 3.0 SFF mezzanine |
|---|---|
| Cooling | Passive heatsink |
| Cables | 400G OSFP or QSFP112 transceivers, direct attach and active optical cables |
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
- Full height and low profile 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-8 SuperNIC, 800 Gb/s
- Speed
- 800 Gb/s, InfiniBand XDR or 800 GbE
- Host interface
- PCIe Gen6 x16
- Offloads
- RDMA, GPUDirect RDMA
- Form factor
- PCIe card with short or tall bracket
400G Ethernet leaf switch
- Ports
- 32 x 400 GbE QSFP112
- Throughput
- 25.6 Tb/s aggregate switching
- Features
- RoCE v2, VXLAN and EVPN
- Form factor
- 1U, air cooled
BlueField-3 DPU, 400 Gb/s
- Speed
- 400 Gb/s, 1 x 400 Gb/s or 2 x 200 Gb/s
- Compute
- Arm cores with onboard DDR memory running the DOCA software stack
- Offloads
- Software defined networking
- Host interface
- PCIe Gen5 x16
Compare it with the next class up or down
The nearest classes we quote are the ConnectX-8 SuperNIC, 800 Gb/s and the 400G Ethernet leaf 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