Which link should carry traffic between two GPUs in the same HGX system?
Why A is the answer
NVLink
NVLink joins GPUs directly at high bandwidth.
In detail
- NVLink joins GPUs directly.
- It joins GPUs directly.
Use case
An illustrative scenario, not a real incident.
Two GPUs in one HGX system exchange data over NVLink, not the network.
See it in different ways
flowchart TD A["GPUs in one server"] -->|"NVLink and NVSwitch"| B["Scale up"] B --> C["Server"] C -->|"InfiniBand or Ethernet"| D["Scale out to a cluster"]
mindmap
root((AI cluster))
Compute
GPU nodes
NVLink
Fabric
InfiniBand
Ethernet
Storage
High throughput
Management
Out-of-band networksequenceDiagram participant P0 as GPU 1 participant P1 as GPU 2 P0->>P1: NVLink transfer P1-->>P0: NVLink transfer
Key point: NVLink carries local GPU traffic.
Why the other options are wrong
- B. A public internet link
- That is far too slow.
- C. A management VLAN
- That is for management.
- D. A serial console cable
- That is for administration.