DPU Offload vs Host Cores: When the 100G NIC Stops Being Free
A serving node that pushes 100 Gbps of model traffic does not do it for free. Depending on kernel, MTU and how many small requests you handle, the host burns somewhere between two and six cores on network stack work: interrupts, softirq, socket copies, TLS record processing. On a 64-core machine th…
Read the full story at DEV Community — AI ↗
Timeline · 1 report
- 2026-11-03 19:12 · DEV Community — AI
DPU Offload vs Host Cores: When the 100G NIC Stops Being Free