{"id":"network-adapter","level":"server","name":"Network interface adapter","shortName":"NIC","role":"Moves data between the server and the cluster network, often with direct-memory and protocol offloads.","description":"A NIC converts server memory transfers into network packets and receives packets from other machines. High-performance adapters can perform remote direct memory access, allowing registered buffers to be transferred with less involvement from the remote CPU. GPU-aware paths can also reduce unnecessary staging through host memory when the platform and software support them. These capabilities do not eliminate synchronization, congestion, or network topology. On a supported path, a tensor fragment leaves a GPU buffer, passes through the adapter and reaches a remote buffer; control and synchronization govern when the transfer is usable. The NIC belongs to the scale-out path between systems; it should not be confused with NVLink’s local GPU fabric or with the ordinary management network.","specs":[{"label":"ConnectX-7 example","value":"Up to 400 Gb/s throughput","context":"NVIDIA adapter family rating; selected card and port modes vary."},{"label":"Byte-rate equivalent","value":"50 GB/s at 400 Gb/s","context":"Arithmetic conversion before protocol overhead; not 400 GB/s."},{"label":"Acceleration","value":"RDMA / RoCE and offloads","context":"Availability depends on adapter, firmware, topology, and software."}],"interfaces":[{"name":"Local memory transfer","direction":"in","signal":"Registered host or supported GPU buffers"},{"name":"Network packets","direction":"out","signal":"Tensor, storage, and application traffic"},{"name":"Received packets","direction":"in","signal":"Remote data and protocol responses"}],"tradeoff":"Faster ports help only if the host path and network can sustain them. Queueing, congestion control, and placement can dominate tail latency even below the nominal line rate.","whyAI":"Distributed training exchanges gradients and parameters across servers. Distributed inference and storage access also depend on efficient movement beyond one rack’s local GPU domain.","related":["pcie-interface","nvlink-port","host-cpu","compute-baseboard","leaf-switch"],"sources":[{"title":"NVIDIA ConnectX Ethernet adapters","url":"https://www.nvidia.com/en-us/networking/ethernet-adapters/"},{"title":"NVIDIA ConnectX-7 adapter hardware manual","url":"https://networking-docs.nvidia.com/connectx7hw/introduction"}],"scale":"Centimeter-scale adapter or module with external copper/optical links","kind":"rdma-buffer-to-packets","mechanism":["Software or a supported accelerator path identifies buffers and initiates a network operation.","The NIC reads data, applies the appropriate protocol processing, and transmits packets through the network.","The remote adapter places received data into supported buffers and reports completion so dependent work can continue."],"misconception":"A 400 Gb/s port is not a 400 GB/s port. Dividing by eight gives 50 GB/s before network and protocol overhead.","check":{"question":"What does RDMA principally reduce in a supported transfer?","options":["The remote CPU’s work copying and processing the payload.","The number of bits in the model vocabulary.","The need for any network fabric."],"answer":0,"explanation":"RDMA places data directly into registered memory with less remote CPU intervention. It still requires a functioning network, protocol, and synchronization."},"branch":"Connectivity","subbranch":"Server scale-out communication","science":[{"title":"Small messages and large transfers have different bottlenecks","equation":"Tmessage ≈ α + M/Bpayload","explanation":"α represents fixed software, adapter, and path latency; M is payload bytes; Bpayload is sustained payload bandwidth in bytes per second. For small M, startup dominates. For large M, bandwidth dominates. This explains why buying a faster line-rate adapter may help bulk gradient transfers more than tiny synchronization messages, and why scheduling and aggregation matter.","assumptions":"A single uncontended path and approximately constant startup cost. Congestion, multiple hops, retransmission, protocol offloads, and overlap can change both terms. Convert bits per second to bytes per second before comparing them."}],"history":[],"prerequisites":["pcie-interface"],"learningObjective":"Separate fixed message startup time from payload transfer time on a network path.","engineeringExample":"Synthetic uncontended transfer: α = 5 μs and payload bandwidth = 25 GB/s. A 1 MB message needs 40 μs of serialization, giving about 45 μs total. A 1 kB message needs only 0.04 μs of serialization but still about 5.04 μs total, so startup dominates.","evidenceNotes":"The lesson’s latency-plus-payload model uses decimal bytes and one direction. These are synthetic rates; the linked network references do not make this a promise about RDMA, GPU access, congestion or any deployed adapter.","labId":"engineering","scope":"Vary ranks, payload, effective bandwidth and startup in an ideal ring all-reduce; this is not a topology-specific or every-collective traffic model.","deepLabId":"network","investigationId":"execution","labScope":"Vary ranks, payload, effective bandwidth and startup in an ideal ring all-reduce; this is not a topology-specific or every-collective traffic model."}